paddlex 3.0.0rc0__py3-none-any.whl → 3.0.1__py3-none-any.whl

paddlex/inference/pipelines/layout_parsing/utils.py

@@ -1,4 +1,4 @@
-# copyright (c) 2024 PaddlePaddle Authors. All Rights Reserve.
+# Copyright (c) 2024 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,22 +14,20 @@
 
 __all__ = [
     "get_sub_regions_ocr_res",
-    "get_layout_ordering",
-    "get_single_block_parsing_res",
     "get_show_color",
     "sorted_layout_boxes",
 ]
 
-import numpy as np
-from PIL import Image
-import uuid
 import re
-from pathlib import Path
 from copy import deepcopy
-from typing import Optional, Union, List, Tuple, Dict, Any
-from ..ocr.result import OCRResult
-from ...models.object_detection.result import DetResult
+from typing import Dict, List, Optional, Tuple, Union
+
+import numpy as np
+from PIL import Image
+
 from ..components import convert_points_to_boxes
+from ..ocr.result import OCRResult
+from .setting import BLOCK_LABEL_MAP, REGION_SETTINGS
 
 
 def get_overlap_boxes_idx(src_boxes: np.ndarray, ref_boxes: np.ndarray) -> List:
@@ -173,808 +171,453 @@ def sorted_layout_boxes(res, w):
     return new_res
 
 
-def _calculate_overlap_area_div_minbox_area_ratio(
-    bbox1: Union[list, tuple],
-    bbox2: Union[list, tuple],
+def calculate_projection_overlap_ratio(
+    bbox1: List[float],
+    bbox2: List[float],
+    direction: str = "horizontal",
+    mode="union",
 ) -> float:
     """
-    Calculate the ratio of the overlap area between bbox1 and bbox2
-    to the area of the smaller bounding box.
+    Calculate the IoU of lines between two bounding boxes.
 
     Args:
-        bbox1 (list or tuple): Coordinates of the first bounding box [x_min, y_min, x_max, y_max].
-        bbox2 (list or tuple): Coordinates of the second bounding box [x_min, y_min, x_max, y_max].
+        bbox1 (List[float]): First bounding box [x_min, y_min, x_max, y_max].
+        bbox2 (List[float]): Second bounding box [x_min, y_min, x_max, y_max].
+        direction (str): direction of the projection, "horizontal" or "vertical".
 
     Returns:
-        float: The ratio of the overlap area to the area of the smaller bounding box.
-    """
-    bbox1 = list(map(int, bbox1))
-    bbox2 = list(map(int, bbox2))
-
-    x_left = max(bbox1[0], bbox2[0])
-    y_top = max(bbox1[1], bbox2[1])
-    x_right = min(bbox1[2], bbox2[2])
-    y_bottom = min(bbox1[3], bbox2[3])
-
-    if x_right <= x_left or y_bottom <= y_top:
-        return 0.0
-
-    intersection_area = (x_right - x_left) * (y_bottom - y_top)
-    area_bbox1 = (bbox1[2] - bbox1[0]) * (bbox1[3] - bbox1[1])
-    area_bbox2 = (bbox2[2] - bbox2[0]) * (bbox2[3] - bbox2[1])
-    min_box_area = min(area_bbox1, area_bbox2)
-
-    if min_box_area <= 0:
-        return 0.0
-
-    return intersection_area / min_box_area
-
-
-def _whether_y_overlap_exceeds_threshold(
-    bbox1: Union[list, tuple],
-    bbox2: Union[list, tuple],
-    overlap_ratio_threshold: float = 0.6,
-) -> bool:
+        float: Line overlap ratio. Returns 0 if there is no overlap.
     """
-    Determines whether the vertical overlap between two bounding boxes exceeds a given threshold.
+    start_index, end_index = 1, 3
+    if direction == "horizontal":
+        start_index, end_index = 0, 2
 
-    Args:
-        bbox1 (list or tuple): The first bounding box defined as (left, top, right, bottom).
-        bbox2 (list or tuple): The second bounding box defined as (left, top, right, bottom).
-        overlap_ratio_threshold (float): The threshold ratio to determine if the overlap is significant.
-                                         Defaults to 0.6.
-
-    Returns:
-        bool: True if the vertical overlap divided by the minimum height of the two bounding boxes
-              exceeds the overlap_ratio_threshold, otherwise False.
-    """
-    _, y1_0, _, y1_1 = bbox1
-    _, y2_0, _, y2_1 = bbox2
+    intersection_start = max(bbox1[start_index], bbox2[start_index])
+    intersection_end = min(bbox1[end_index], bbox2[end_index])
+    overlap = intersection_end - intersection_start
+    if overlap <= 0:
+        return 0
 
-    overlap = max(0, min(y1_1, y2_1) - max(y1_0, y2_0))
-    min_height = min(y1_1 - y1_0, y2_1 - y2_0)
+    if mode == "union":
+        ref_width = max(bbox1[end_index], bbox2[end_index]) - min(
+            bbox1[start_index], bbox2[start_index]
+        )
+    elif mode == "small":
+        ref_width = min(
+            bbox1[end_index] - bbox1[start_index], bbox2[end_index] - bbox2[start_index]
+        )
+    elif mode == "large":
+        ref_width = max(
+            bbox1[end_index] - bbox1[start_index], bbox2[end_index] - bbox2[start_index]
+        )
+    else:
+        raise ValueError(
+            f"Invalid mode {mode}, must be one of ['union', 'small', 'large']."
+        )
 
-    return (overlap / min_height) > overlap_ratio_threshold
+    return overlap / ref_width if ref_width > 0 else 0.0
 
 
-def _adjust_span_text(span: List[str], prepend: bool = False, append: bool = False):
+def calculate_overlap_ratio(
+    bbox1: Union[list, tuple], bbox2: Union[list, tuple], mode="union"
+) -> float:
     """
-    Adjust the text of a span by prepending or appending a newline.
+    Calculate the overlap ratio between two bounding boxes.
 
     Args:
-        span (list): A list where the second element is the text of the span.
-        prepend (bool): If True, prepend a newline to the text.
-        append (bool): If True, append a newline to the text.
+        bbox1 (list or tuple): The first bounding box, format [x_min, y_min, x_max, y_max]
+        bbox2 (list or tuple): The second bounding box, format [x_min, y_min, x_max, y_max]
+        mode (str): The mode of calculation, either 'union', 'small', or 'large'.
 
     Returns:
-        None: The function modifies the span in place.
+        float: The overlap ratio value between the two bounding boxes
     """
-    if prepend:
-        span[1] = "\n" + span[1]
-    if append:
-        span[1] = span[1] + "\n"
-    return span
+    x_min_inter = max(bbox1[0], bbox2[0])
+    y_min_inter = max(bbox1[1], bbox2[1])
+    x_max_inter = min(bbox1[2], bbox2[2])
+    y_max_inter = min(bbox1[3], bbox2[3])
 
+    inter_width = max(0, x_max_inter - x_min_inter)
+    inter_height = max(0, y_max_inter - y_min_inter)
 
-def _format_line(
-    line: List[List[Union[List[int], str]]],
-    layout_min: int,
-    layout_max: int,
-    is_reference: bool = False,
-) -> None:
-    """
-    Format a line of text spans based on layout constraints.
+    inter_area = inter_width * inter_height
 
-    Args:
-        line (list): A list of spans, where each span is a list containing a bounding box and text.
-        layout_min (int): The minimum x-coordinate of the layout bounding box.
-        layout_max (int): The maximum x-coordinate of the layout bounding box.
-        is_reference (bool): A flag indicating whether the line is a reference line, which affects formatting rules.
+    bbox1_area = caculate_bbox_area(bbox1)
+    bbox2_area = caculate_bbox_area(bbox2)
 
-    Returns:
-        None: The function modifies the line in place.
-    """
-    first_span = line[0]
-    end_span = line[-1]
-
-    if not is_reference:
-        if first_span[0][0] - layout_min > 10:
-            first_span = _adjust_span_text(first_span, prepend=True)
-        if layout_max - end_span[0][2] > 10:
-            end_span = _adjust_span_text(end_span, append=True)
+    if mode == "union":
+        ref_area = bbox1_area + bbox2_area - inter_area
+    elif mode == "small":
+        ref_area = min(bbox1_area, bbox2_area)
+    elif mode == "large":
+        ref_area = max(bbox1_area, bbox2_area)
     else:
-        if first_span[0][0] - layout_min < 5:
-            first_span = _adjust_span_text(first_span, prepend=True)
-        if layout_max - end_span[0][2] > 20:
-            end_span = _adjust_span_text(end_span, append=True)
+        raise ValueError(
+            f"Invalid mode {mode}, must be one of ['union', 'small', 'large']."
+        )
 
-    line[0] = first_span
-    line[-1] = end_span
+    if ref_area == 0:
+        return 0.0
 
-    return line
+    return inter_area / ref_area
 
 
-def split_boxes_if_x_contained(boxes, offset=1e-5):
-    """
-    Check if there is any complete containment in the x-direction
-    between the bounding boxes and split the containing box accordingly.
+def group_boxes_into_lines(ocr_rec_res, line_height_iou_threshold):
+    rec_boxes = ocr_rec_res["boxes"]
+    rec_texts = ocr_rec_res["rec_texts"]
+    rec_labels = ocr_rec_res["rec_labels"]
 
-    Args:
-        boxes (list of lists): Each element is a list containing an ndarray of length 4, a description, and a label.
-        offset (float): A small offset value to ensure that the split boxes are not too close to the original boxes.
-    Returns:
-        A new list of boxes, including split boxes, with the same `rec_text` and `label` attributes.
-    """
+    text_boxes = [
+        rec_boxes[i] for i in range(len(rec_boxes)) if rec_labels[i] == "text"
+    ]
+    text_orientation = calculate_text_orientation(text_boxes)
 
-    def is_x_contained(box_a, box_b):
-        """Check if box_a completely contains box_b in the x-direction."""
-        return box_a[0][0] <= box_b[0][0] and box_a[0][2] >= box_b[0][2]
+    match_direction = "vertical" if text_orientation == "horizontal" else "horizontal"
 
-    new_boxes = []
+    line_start_index = 1 if text_orientation == "horizontal" else 0
+    line_end_index = 3 if text_orientation == "horizontal" else 2
 
-    for i in range(len(boxes)):
-        box_a = boxes[i]
-        is_split = False
-        for j in range(len(boxes)):
-            if i == j:
-                continue
-            box_b = boxes[j]
-            if is_x_contained(box_a, box_b):
-                is_split = True
-                # Split box_a based on the x-coordinates of box_b
-                if box_a[0][0] < box_b[0][0]:
-                    w = box_b[0][0] - offset - box_a[0][0]
-                    if w > 1:
-                        new_boxes.append(
-                            [
-                                np.array(
-                                    [
-                                        box_a[0][0],
-                                        box_a[0][1],
-                                        box_b[0][0] - offset,
-                                        box_a[0][3],
-                                    ]
-                                ),
-                                box_a[1],
-                                box_a[2],
-                            ]
-                        )
-                if box_a[0][2] > box_b[0][2]:
-                    w = box_a[0][2] - box_b[0][2] + offset
-                    if w > 1:
-                        box_a = [
-                            np.array(
-                                [
-                                    box_b[0][2] + offset,
-                                    box_a[0][1],
-                                    box_a[0][2],
-                                    box_a[0][3],
-                                ]
-                            ),
-                            box_a[1],
-                            box_a[2],
-                        ]
-            if j == len(boxes) - 1 and is_split:
-                new_boxes.append(box_a)
-        if not is_split:
-            new_boxes.append(box_a)
-
-    return new_boxes
+    spans = list(zip(rec_boxes, rec_texts, rec_labels))
+    sort_index = 1
+    reverse = False
+    if text_orientation == "vertical":
+        sort_index = 0
+        reverse = True
+    spans.sort(key=lambda span: span[0][sort_index], reverse=reverse)
+    spans = [list(span) for span in spans]
 
+    lines = []
+    line = [spans[0]]
+    line_region_box = spans[0][0].copy()
+    line_heights = []
+    # merge line
+    for span in spans[1:]:
+        rec_bbox = span[0]
+        if (
+            calculate_projection_overlap_ratio(
+                line_region_box, rec_bbox, match_direction, mode="small"
+            )
+            >= line_height_iou_threshold
+        ):
+            line.append(span)
+            line_region_box[line_start_index] = min(
+                line_region_box[line_start_index], rec_bbox[line_start_index]
+            )
+            line_region_box[line_end_index] = max(
+                line_region_box[line_end_index], rec_bbox[line_end_index]
+            )
+        else:
+            line_heights.append(
+                line_region_box[line_end_index] - line_region_box[line_start_index]
+            )
+            lines.append(line)
+            line = [span]
+            line_region_box = rec_bbox.copy()
 
-def _sort_line_by_x_projection(
-    input_img: np.ndarray,
-    general_ocr_pipeline: Any,
-    line: List[List[Union[List[int], str]]],
-) -> None:
-    """
-    Sort a line of text spans based on their vertical position within the layout bounding box.
+    lines.append(line)
+    line_heights.append(
+        line_region_box[line_end_index] - line_region_box[line_start_index]
+    )
 
-    Args:
-        input_img (ndarray): The input image used for OCR.
-        general_ocr_pipeline (Any): The general OCR pipeline used for text recognition.
-        line (list): A list of spans, where each span is a list containing a bounding box and text.
+    min_height = min(line_heights) if line_heights else 0
+    max_height = max(line_heights) if line_heights else 0
 
-    Returns:
-        list: The sorted line of text spans.
-    """
-    splited_boxes = split_boxes_if_x_contained(line)
-    splited_lines = []
-    if len(line) != len(splited_boxes):
-        splited_boxes.sort(key=lambda span: span[0][0])
-        text_rec_model = general_ocr_pipeline.text_rec_model
-        for span in splited_boxes:
-            if span[2] == "text":
-                crop_img = input_img[
-                    int(span[0][1]) : int(span[0][3]),
-                    int(span[0][0]) : int(span[0][2]),
-                ]
-                span[1] = next(text_rec_model([crop_img]))["rec_text"]
-            splited_lines.append(span)
-    else:
-        splited_lines = line
+    if max_height > min_height * 2 and text_orientation == "vertical":
+        line_heights = np.array(line_heights)
+        min_height_num = np.sum(line_heights < min_height * 1.1)
+        if min_height_num < len(lines) * 0.4:
+            condition = line_heights > min_height * 1.1
+            lines = [value for value, keep in zip(lines, condition) if keep]
 
-    return splited_lines
+    return lines, text_orientation, np.mean(line_heights)
 
 
-def _sort_ocr_res_by_y_projection(
-    input_img: np.ndarray,
-    general_ocr_pipeline: Any,
-    label: Any,
-    block_bbox: Tuple[int, int, int, int],
-    ocr_res: Dict[str, List[Any]],
-    line_height_iou_threshold: float = 0.7,
-) -> Dict[str, List[Any]]:
+def calculate_minimum_enclosing_bbox(bboxes):
     """
-    Sorts OCR results based on their spatial arrangement, grouping them into lines and blocks.
+    Calculate the minimum enclosing bounding box for a list of bounding boxes.
 
     Args:
-        input_img (ndarray): The input image used for OCR.
-        general_ocr_pipeline (Any): The general OCR pipeline used for text recognition.
-        label (Any): The label associated with the OCR results. It's not used in the function but might be
-                     relevant for other parts of the calling context.
-        block_bbox (Tuple[int, int, int, int]): A tuple representing the layout bounding box, defined as
-                                                 (left, top, right, bottom).
-        ocr_res (Dict[str, List[Any]]): A dictionary containing OCR results with the following keys:
-            - "boxes": A list of bounding boxes, each defined as [left, top, right, bottom].
-            - "rec_texts": A corresponding list of recognized text strings for each box.
-        line_height_iou_threshold (float): The threshold for determining whether two boxes belong to
-                                           the same line based on their vertical overlap. Defaults to 0.7.
+        bboxes (list): A list of bounding boxes represented as lists of four integers [x1, y1, x2, y2].
 
     Returns:
-        Dict[str, List[Any]]: A dictionary with the same structure as `ocr_res`, but with boxes and texts sorted
-                              and grouped into lines and blocks.
+        list: The minimum enclosing bounding box represented as a list of four integers [x1, y1, x2, y2].
     """
-    assert (
-        ocr_res["boxes"] and ocr_res["rec_texts"]
-    ), "OCR results must contain 'boxes' and 'rec_texts'"
+    if not bboxes:
+        raise ValueError("The list of bounding boxes is empty.")
 
-    boxes = ocr_res["boxes"]
-    rec_texts = ocr_res["rec_texts"]
-    rec_labels = ocr_res["rec_labels"]
+    # Convert the list of bounding boxes to a NumPy array
+    bboxes_array = np.array(bboxes)
 
-    x_min, _, x_max, _ = block_bbox
-    inline_x_min = min([box[0] for box in boxes])
-    inline_x_max = max([box[2] for box in boxes])
+    # Compute the minimum and maximum values along the respective axes
+    min_x = np.min(bboxes_array[:, 0])
+    min_y = np.min(bboxes_array[:, 1])
+    max_x = np.max(bboxes_array[:, 2])
+    max_y = np.max(bboxes_array[:, 3])
 
-    spans = list(zip(boxes, rec_texts, rec_labels))
+    # Return the minimum enclosing bounding box
+    return [min_x, min_y, max_x, max_y]
 
-    spans.sort(key=lambda span: span[0][1])
-    spans = [list(span) for span in spans]
-
-    lines = []
-    current_line = [spans[0]]
-    current_y0, current_y1 = spans[0][0][1], spans[0][0][3]
 
-    for span in spans[1:]:
-        y0, y1 = span[0][1], span[0][3]
-        if _whether_y_overlap_exceeds_threshold(
-            (0, current_y0, 0, current_y1),
-            (0, y0, 0, y1),
-            line_height_iou_threshold,
-        ):
-            current_line.append(span)
-            current_y0 = min(current_y0, y0)
-            current_y1 = max(current_y1, y1)
-        else:
-            lines.append(current_line)
-            current_line = [span]
-            current_y0, current_y1 = y0, y1
-
-    if current_line:
-        lines.append(current_line)
-
-    new_lines = []
-    for line in lines:
-        line.sort(key=lambda span: span[0][0])
-
-        ocr_labels = [span[2] for span in line]
-        if "formula" in ocr_labels:
-            line = _sort_line_by_x_projection(input_img, general_ocr_pipeline, line)
-        if label == "reference":
-            line = _format_line(line, inline_x_min, inline_x_max, is_reference=True)
-        elif label != "content":
-            line = _format_line(line, x_min, x_max)
-        new_lines.append(line)
-
-    ocr_res["boxes"] = [span[0] for line in new_lines for span in line]
-    if label == "content":
-        ocr_res["rec_texts"] = [
-            "".join(f"{span[1]} " for span in line).rstrip() for line in new_lines
-        ]
-    else:
-        ocr_res["rec_texts"] = [span[1] + " " for line in new_lines for span in line]
-    return ocr_res, len(new_lines)
-
-
-def _process_text(input_text: str) -> str:
+def calculate_text_orientation(
+    bboxes: List[List[int]], orientation_ratio: float = 1.5
+) -> bool:
     """
-    Process the input text to handle spaces.
-
-    The function removes multiple consecutive spaces between Chinese characters and ensures that
-    only a single space is retained between Chinese and non-Chinese characters.
+    Calculate the orientation of the text based on the bounding boxes.
 
     Args:
-        input_text (str): The text to be processed.
+        bboxes (list): A list of bounding boxes.
+        orientation_ratio (float): Ratio for determining orientation. Default is 1.5.
 
     Returns:
-        str: The processed text with properly formatted spaces.
+        str: "horizontal" or "vertical".
     """
 
-    def handle_spaces_(text: str) -> str:
-        """
-        Handle spaces in the text by removing multiple consecutive spaces and inserting a single space
-        between Chinese and non-Chinese characters.
-
-        Args:
-            text (str): The text to handle spaces for.
-
-        Returns:
-            str: The text with properly formatted spaces.
-        """
-        spaces = re.finditer(r"\s+", text)
-        processed_text = list(text)
-
-        for space in reversed(list(spaces)):
-            start, end = space.span()
-            prev_char = processed_text[start - 1] if start > 0 else ""
-            next_char = processed_text[end] if end < len(processed_text) else ""
-
-            is_prev_chinese = (
-                re.match(r"[\u4e00-\u9fff]", prev_char) if prev_char else False
-            )
-            is_next_chinese = (
-                re.match(r"[\u4e00-\u9fff]", next_char) if next_char else False
+    horizontal_box_num = 0
+    for bbox in bboxes:
+        if len(bbox) != 4:
+            raise ValueError(
+                "Invalid bounding box format. Expected a list of length 4."
             )
+        x1, y1, x2, y2 = bbox
+        width = x2 - x1
+        height = y2 - y1
+        horizontal_box_num += 1 if width * orientation_ratio >= height else 0
 
-            if is_prev_chinese and is_next_chinese:
-                processed_text[start:end] = []
-            else:
-                processed_text[start:end] = [" "]
+    return "horizontal" if horizontal_box_num >= len(bboxes) * 0.5 else "vertical"
 
-        return "".join(processed_text)
 
-    text_without_spaces = handle_spaces_(input_text)
+def is_english_letter(char):
+    return bool(re.match(r"^[A-Za-z]$", char))
 
-    final_text = re.sub(r"\s+", " ", text_without_spaces).strip()
-    return final_text
 
+def is_numeric(char):
+    return bool(re.match(r"^[\d.]+$", char))
 
-def get_single_block_parsing_res(
-    general_ocr_pipeline: Any,
-    overall_ocr_res: OCRResult,
-    layout_det_res: DetResult,
-    table_res_list: list,
-    seal_res_list: list,
-) -> OCRResult:
+
+def is_non_breaking_punctuation(char):
     """
-    Extract structured information from OCR and layout detection results.
+    判断一个字符是否是不需要换行的标点符号，包括全角和半角的符号。
 
-    Args:
-        overall_ocr_res (OCRResult): An object containing the overall OCR results, including detected text boxes and recognized text. The structure is expected to have:
-            - "input_img": The image on which OCR was performed.
-            - "dt_boxes": A list of detected text box coordinates.
-            - "rec_texts": A list of recognized text corresponding to the detected boxes.
+    :param char: str, 单个字符
+    :return: bool, 如果字符是不需要换行的标点符号，返回True，否则返回False
+    """
+    non_breaking_punctuations = {
+        ",",  # 半角逗号
+        "，",  # 全角逗号
+        "、",  # 顿号
+        ";",  # 半角分号
+        "；",  # 全角分号
+        ":",  # 半角冒号
+        "：",  # 全角冒号
+        "-",  # 连字符
+    }
 
-        layout_det_res (DetResult): An object containing the layout detection results, including detected layout boxes and their labels. The structure is expected to have:
-            - "boxes": A list of dictionaries with keys "coordinate" for box coordinates and "block_label" for the type of content.
+    return char in non_breaking_punctuations
 
-        table_res_list (list): A list of table detection results, where each item is a dictionary containing:
-            - "block_bbox": The bounding box of the table layout.
-            - "pred_html": The predicted HTML representation of the table.
 
-        seal_res_list (List): A list of seal detection results. The details of each item depend on the specific application context.
+def format_line(
+    line: List[List[Union[List[int], str]]],
+    text_direction: int,
+    block_width: int,
+    block_start_coordinate: int,
+    block_stop_coordinate: int,
+    line_gap_limit: int = 10,
+    block_label: str = "text",
+) -> None:
+    """
+    Format a line of text spans based on layout constraints.
 
+    Args:
+        line (list): A list of spans, where each span is a list containing a bounding box and text.
+        block_left_coordinate (int): The text line directional minimum coordinate of the layout bounding box.
+        block_stop_coordinate (int): The text line directional maximum x-coordinate of the layout bounding box.
+        first_line_span_limit (int): The limit for the number of pixels before the first span that should be considered part of the first line. Default is 10.
+        line_gap_limit (int): The limit for the number of pixels after the last span that should be considered part of the last line. Default is 10.
+        block_label (str): The label associated with the entire block. Default is 'text'.
     Returns:
-        list: A list of structured boxes where each item is a dictionary containing:
-            - "block_label": The label of the content (e.g., 'table', 'chart', 'image').
-            - The label as a key with either table HTML or image data and text.
-            - "block_bbox": The coordinates of the layout box.
+        None: The function modifies the line in place.
     """
+    first_span_box = line[0][0]
+    last_span_box = line[-1][0]
 
-    single_block_layout_parsing_res = []
-    input_img = overall_ocr_res["doc_preprocessor_res"]["output_img"]
-    seal_index = 0
-    with_doc_title = False
-    max_block_area = 0.0
-    paragraph_title_indexs = []
-
-    layout_det_res_list, _ = _remove_overlap_blocks(
-        deepcopy(layout_det_res["boxes"]),
-        threshold=0.5,
-        smaller=True,
-    )
-
-    for box_idx, box_info in enumerate(layout_det_res_list):
-        block_bbox = box_info["coordinate"]
-        label = box_info["label"]
-        rec_res = {"boxes": [], "rec_texts": [], "rec_labels": [], "flag": False}
-        seg_start_coordinate = float("inf")
-        seg_end_coordinate = float("-inf")
-        num_of_lines = 1
-
-        if label == "doc_title":
-            with_doc_title = True
-        elif label == "paragraph_title":
-            paragraph_title_indexs.append(box_idx)
-
-        block_area = (block_bbox[2] - block_bbox[0]) * (block_bbox[3] - block_bbox[1])
-        max_block_area = max(max_block_area, block_area)
-
-        if label == "table":
-            for table_res in table_res_list:
-                if len(table_res["cell_box_list"]) == 0:
-                    continue
-                if (
-                    _calculate_overlap_area_div_minbox_area_ratio(
-                        block_bbox, table_res["cell_box_list"][0]
-                    )
-                    > 0.5
-                ):
-                    single_block_layout_parsing_res.append(
-                        {
-                            "block_label": label,
-                            "block_content": table_res["pred_html"],
-                            "block_bbox": block_bbox,
-                        },
-                    )
-                    break
-        elif label == "seal":
-            if len(seal_res_list) > 0:
-                single_block_layout_parsing_res.append(
-                    {
-                        "block_label": label,
-                        "block_content": _process_text(
-                            ", ".join(seal_res_list[seal_index]["rec_texts"])
-                        ),
-                        "block_bbox": block_bbox,
-                    },
-                )
-                seal_index += 1
-        else:
-            overall_text_boxes = overall_ocr_res["rec_boxes"]
-            for box_no in range(len(overall_text_boxes)):
-                if (
-                    _calculate_overlap_area_div_minbox_area_ratio(
-                        block_bbox, overall_text_boxes[box_no]
-                    )
-                    > 0.5
-                ):
-                    rec_res["boxes"].append(overall_text_boxes[box_no])
-                    rec_res["rec_texts"].append(
-                        overall_ocr_res["rec_texts"][box_no],
-                    )
-                    rec_res["rec_labels"].append(
-                        overall_ocr_res["rec_labels"][box_no],
-                    )
-                    rec_res["flag"] = True
-
-            if rec_res["flag"]:
-                rec_res, num_of_lines = _sort_ocr_res_by_y_projection(
-                    input_img, general_ocr_pipeline, label, block_bbox, rec_res, 0.7
-                )
-                seg_start_coordinate = rec_res["boxes"][0][0]
-                seg_end_coordinate = rec_res["boxes"][-1][2]
-                if label == "formula":
-                    rec_res["rec_texts"] = [
-                        rec_res_text.replace("$", "")
-                        for rec_res_text in rec_res["rec_texts"]
-                    ]
-
-            if label in ["chart", "image"]:
-                x_min, y_min, x_max, y_max = list(map(int, block_bbox))
-                img_path = f"imgs/img_in_table_box_{x_min}_{y_min}_{x_max}_{y_max}.jpg"
-                img = Image.fromarray(input_img[y_min:y_max, x_min:x_max, ::-1])
-                single_block_layout_parsing_res.append(
-                    {
-                        "block_label": label,
-                        "block_content": _process_text("".join(rec_res["rec_texts"])),
-                        "block_image": {img_path: img},
-                        "block_bbox": block_bbox,
-                    },
-                )
-            else:
-                if label in ["doc_title"]:
-                    content = " ".join(rec_res["rec_texts"])
-                elif label in ["content"]:
-                    content = "\n".join(rec_res["rec_texts"])
+    for span in line:
+        if span[2] == "formula" and block_label != "formula":
+            formula_rec = span[1]
+            if not formula_rec.startswith("$") and not formula_rec.endswith("$"):
+                if len(line) > 1:
+                    span[1] = f"${span[1]}$"
                 else:
-                    content = "".join(rec_res["rec_texts"])
-                    if label != "reference":
-                        content = _process_text(content)
-                single_block_layout_parsing_res.append(
-                    {
-                        "block_label": label,
-                        "block_content": content,
-                        "block_bbox": block_bbox,
-                        "seg_start_coordinate": seg_start_coordinate,
-                        "seg_end_coordinate": seg_end_coordinate,
-                        "num_of_lines": num_of_lines,
-                        "block_area": block_area,
-                    },
-                )
+                    span[1] = f"\n${span[1]}$"
+
+    line_text = ""
+    for span in line:
+        _, text, label = span
+        line_text += text
+        if len(text) > 0 and is_english_letter(line_text[-1]) or label == "formula":
+            line_text += " "
+
+    if text_direction == "horizontal":
+        text_start_index = 0
+        text_stop_index = 2
+    else:
+        text_start_index = 1
+        text_stop_index = 3
 
+    need_new_line = False
     if (
-        not with_doc_title
-        and len(paragraph_title_indexs) == 1
-        and single_block_layout_parsing_res[paragraph_title_indexs[0]].get(
-            "block_area", 0
-        )
-        > max_block_area * 0.3
+        len(line_text) > 0
+        and not is_english_letter(line_text[-1])
+        and not is_non_breaking_punctuation(line_text[-1])
     ):
-        single_block_layout_parsing_res[paragraph_title_indexs[0]][
-            "block_label"
-        ] = "doc_title"
-
-    if len(layout_det_res_list) == 0:
-        for ocr_rec_box, ocr_rec_text in zip(
-            overall_ocr_res["rec_boxes"], overall_ocr_res["rec_texts"]
-        ):
-            single_block_layout_parsing_res.append(
-                {
-                    "block_label": "text",
-                    "block_content": ocr_rec_text,
-                    "block_bbox": ocr_rec_box,
-                    "seg_start_coordinate": ocr_rec_box[0],
-                    "seg_end_coordinate": ocr_rec_box[2],
-                },
+        if (
+            text_direction == "horizontal"
+            and block_stop_coordinate - last_span_box[text_stop_index] > line_gap_limit
+        ) or (
+            text_direction == "vertical"
+            and (
+                block_stop_coordinate - last_span_box[text_stop_index] > line_gap_limit
+                or first_span_box[1] - block_start_coordinate > line_gap_limit
             )
+        ):
+            need_new_line = True
+
+    if line_text.endswith("-"):
+        line_text = line_text[:-1]
+    elif (
+        len(line_text) > 0 and is_english_letter(line_text[-1])
+    ) or line_text.endswith("$"):
+        line_text += " "
+    elif (
+        len(line_text) > 0
+        and not is_english_letter(line_text[-1])
+        and not is_non_breaking_punctuation(line_text[-1])
+        and not is_numeric(line_text[-1])
+    ) or text_direction == "vertical":
+        if block_stop_coordinate - last_span_box[text_stop_index] > block_width * 0.4:
+            line_text += "\n"
+        if (
+            first_span_box[text_start_index] - block_start_coordinate
+            > block_width * 0.4
+        ):
+            line_text = "\n" + line_text
 
-    single_block_layout_parsing_res = get_layout_ordering(
-        single_block_layout_parsing_res,
-        no_mask_labels=[
-            "text",
-            "formula",
-            "algorithm",
-            "reference",
-            "content",
-            "abstract",
-        ],
-    )
-
-    return single_block_layout_parsing_res
+    return line_text, need_new_line
 
 
-def _projection_by_bboxes(boxes: np.ndarray, axis: int) -> np.ndarray:
+def split_boxes_by_projection(spans: List[List[int]], direction, offset=1e-5):
     """
-    Generate a 1D projection histogram from bounding boxes along a specified axis.
+    Check if there is any complete containment in the x-direction
+    between the bounding boxes and split the containing box accordingly.
 
     Args:
-        boxes: A (N, 4) array of bounding boxes defined by [x_min, y_min, x_max, y_max].
-        axis: Axis for projection; 0 for horizontal (x-axis), 1 for vertical (y-axis).
-
+        spans (list of lists): Each element is a list containing an ndarray of length 4, a text string, and a label.
+        direction: 'horizontal' or 'vertical', indicating whether the spans are arranged horizontally or vertically.
+        offset (float): A small offset value to ensure that the split boxes are not too close to the original boxes.
     Returns:
-        A 1D numpy array representing the projection histogram based on bounding box intervals.
-    """
-    assert axis in [0, 1]
-    max_length = np.max(boxes[:, axis::2])
-    projection = np.zeros(max_length, dtype=int)
-
-    # Increment projection histogram over the interval defined by each bounding box
-    for start, end in boxes[:, axis::2]:
-        projection[start:end] += 1
-
-    return projection
-
-
-def _split_projection_profile(arr_values: np.ndarray, min_value: float, min_gap: float):
+        A new list of boxes, including split boxes, with the same `rec_text` and `label` attributes.
     """
-    Split the projection profile into segments based on specified thresholds.
-
-    Args:
-        arr_values: 1D array representing the projection profile.
-        min_value: Minimum value threshold to consider a profile segment significant.
-        min_gap: Minimum gap width to consider a separation between segments.
 
-    Returns:
-        A tuple of start and end indices for each segment that meets the criteria.
-    """
-    # Identify indices where the projection exceeds the minimum value
-    significant_indices = np.where(arr_values > min_value)[0]
-    if not len(significant_indices):
-        return
-
-    # Calculate gaps between significant indices
-    index_diffs = significant_indices[1:] - significant_indices[:-1]
-    gap_indices = np.where(index_diffs > min_gap)[0]
-
-    # Determine start and end indices of segments
-    segment_starts = np.insert(
-        significant_indices[gap_indices + 1],
-        0,
-        significant_indices[0],
-    )
-    segment_ends = np.append(
-        significant_indices[gap_indices],
-        significant_indices[-1] + 1,
-    )
+    def is_projection_contained(box_a, box_b, start_idx, end_idx):
+        """Check if box_a completely contains box_b in the x-direction."""
+        return box_a[start_idx] <= box_b[start_idx] and box_a[end_idx] >= box_b[end_idx]
 
-    return segment_starts, segment_ends
+    new_boxes = []
+    if direction == "horizontal":
+        projection_start_index, projection_end_index = 0, 2
+    else:
+        projection_start_index, projection_end_index = 1, 3
 
+    for i in range(len(spans)):
+        span = spans[i]
+        is_split = False
+        for j in range(i, len(spans)):
+            box_b = spans[j][0]
+            box_a, text, label = span
+            if is_projection_contained(
+                box_a, box_b, projection_start_index, projection_end_index
+            ):
+                is_split = True
+                # Split box_a based on the x-coordinates of box_b
+                if box_a[projection_start_index] < box_b[projection_start_index]:
+                    w = (
+                        box_b[projection_start_index]
+                        - offset
+                        - box_a[projection_start_index]
+                    )
+                    if w > 1:
+                        new_bbox = box_a.copy()
+                        new_bbox[projection_end_index] = (
+                            box_b[projection_start_index] - offset
+                        )
+                        new_boxes.append(
+                            [
+                                np.array(new_bbox),
+                                text,
+                                label,
+                            ]
+                        )
+                if box_a[projection_end_index] > box_b[projection_end_index]:
+                    w = (
+                        box_a[projection_end_index]
+                        - box_b[projection_end_index]
+                        + offset
+                    )
+                    if w > 1:
+                        box_a[projection_start_index] = (
+                            box_b[projection_end_index] + offset
+                        )
+                        span = [
+                            np.array(box_a),
+                            text,
+                            label,
+                        ]
+            if j == len(spans) - 1 and is_split:
+                new_boxes.append(span)
+        if not is_split:
+            new_boxes.append(span)
 
-def _recursive_yx_cut(
-    boxes: np.ndarray, indices: List[int], res: List[int], min_gap: int = 1
-):
-    """
-    Recursively project and segment bounding boxes, starting with Y-axis and followed by X-axis.
+    return new_boxes
 
-    Args:
-        boxes: A (N, 4) array representing bounding boxes.
-        indices: List of indices indicating the original position of boxes.
-        res: List to store indices of the final segmented bounding boxes.
-        min_gap (int): Minimum gap width to consider a separation between segments on the X-axis. Defaults to 1.
 
-    Returns:
-        None: This function modifies the `res` list in place.
+def remove_extra_space(input_text: str) -> str:
     """
-    assert len(boxes) == len(
-        indices
-    ), "The length of boxes and indices must be the same."
-
-    # Sort by y_min for Y-axis projection
-    y_sorted_indices = boxes[:, 1].argsort()
-    y_sorted_boxes = boxes[y_sorted_indices]
-    y_sorted_indices = np.array(indices)[y_sorted_indices]
-
-    # Perform Y-axis projection
-    y_projection = _projection_by_bboxes(boxes=y_sorted_boxes, axis=1)
-    y_intervals = _split_projection_profile(y_projection, 0, 1)
-
-    if not y_intervals:
-        return
-
-    # Process each segment defined by Y-axis projection
-    for y_start, y_end in zip(*y_intervals):
-        # Select boxes within the current y interval
-        y_interval_indices = (y_start <= y_sorted_boxes[:, 1]) & (
-            y_sorted_boxes[:, 1] < y_end
-        )
-        y_boxes_chunk = y_sorted_boxes[y_interval_indices]
-        y_indices_chunk = y_sorted_indices[y_interval_indices]
-
-        # Sort by x_min for X-axis projection
-        x_sorted_indices = y_boxes_chunk[:, 0].argsort()
-        x_sorted_boxes_chunk = y_boxes_chunk[x_sorted_indices]
-        x_sorted_indices_chunk = y_indices_chunk[x_sorted_indices]
-
-        # Perform X-axis projection
-        x_projection = _projection_by_bboxes(boxes=x_sorted_boxes_chunk, axis=0)
-        x_intervals = _split_projection_profile(x_projection, 0, min_gap)
-
-        if not x_intervals:
-            continue
-
-        # If X-axis cannot be further segmented, add current indices to results
-        if len(x_intervals[0]) == 1:
-            res.extend(x_sorted_indices_chunk)
-            continue
-
-        # Recursively process each segment defined by X-axis projection
-        for x_start, x_end in zip(*x_intervals):
-            x_interval_indices = (x_start <= x_sorted_boxes_chunk[:, 0]) & (
-                x_sorted_boxes_chunk[:, 0] < x_end
-            )
-            _recursive_yx_cut(
-                x_sorted_boxes_chunk[x_interval_indices],
-                x_sorted_indices_chunk[x_interval_indices],
-                res,
-            )
-
+    Process the input text to handle spaces.
 
-def _recursive_xy_cut(
-    boxes: np.ndarray, indices: List[int], res: List[int], min_gap: int = 1
-):
-    """
-    Recursively performs X-axis projection followed by Y-axis projection to segment bounding boxes.
+    The function removes multiple consecutive spaces between Chinese characters and ensures that
+    only a single space is retained between Chinese and non-Chinese characters.
 
     Args:
-        boxes: A (N, 4) array representing bounding boxes with [x_min, y_min, x_max, y_max].
-        indices: A list of indices representing the position of boxes in the original data.
-        res: A list to store indices of bounding boxes that meet the criteria.
-        min_gap (int): Minimum gap width to consider a separation between segments on the X-axis. Defaults to 1.
+        input_text (str): The text to be processed.
 
     Returns:
-        None: This function modifies the `res` list in place.
+        str: The processed text with properly formatted spaces.
     """
-    # Ensure boxes and indices have the same length
-    assert len(boxes) == len(
-        indices
-    ), "The length of boxes and indices must be the same."
-
-    # Sort by x_min to prepare for X-axis projection
-    x_sorted_indices = boxes[:, 0].argsort()
-    x_sorted_boxes = boxes[x_sorted_indices]
-    x_sorted_indices = np.array(indices)[x_sorted_indices]
-
-    # Perform X-axis projection
-    x_projection = _projection_by_bboxes(boxes=x_sorted_boxes, axis=0)
-    x_intervals = _split_projection_profile(x_projection, 0, 1)
-
-    if not x_intervals:
-        return
-
-    # Process each segment defined by X-axis projection
-    for x_start, x_end in zip(*x_intervals):
-        # Select boxes within the current x interval
-        x_interval_indices = (x_start <= x_sorted_boxes[:, 0]) & (
-            x_sorted_boxes[:, 0] < x_end
-        )
-        x_boxes_chunk = x_sorted_boxes[x_interval_indices]
-        x_indices_chunk = x_sorted_indices[x_interval_indices]
-
-        # Sort selected boxes by y_min to prepare for Y-axis projection
-        y_sorted_indices = x_boxes_chunk[:, 1].argsort()
-        y_sorted_boxes_chunk = x_boxes_chunk[y_sorted_indices]
-        y_sorted_indices_chunk = x_indices_chunk[y_sorted_indices]
-
-        # Perform Y-axis projection
-        y_projection = _projection_by_bboxes(boxes=y_sorted_boxes_chunk, axis=1)
-        y_intervals = _split_projection_profile(y_projection, 0, min_gap)
-
-        if not y_intervals:
-            continue
-
-        # If Y-axis cannot be further segmented, add current indices to results
-        if len(y_intervals[0]) == 1:
-            res.extend(y_sorted_indices_chunk)
-            continue
-
-        # Recursively process each segment defined by Y-axis projection
-        for y_start, y_end in zip(*y_intervals):
-            y_interval_indices = (y_start <= y_sorted_boxes_chunk[:, 1]) & (
-                y_sorted_boxes_chunk[:, 1] < y_end
-            )
-            _recursive_xy_cut(
-                y_sorted_boxes_chunk[y_interval_indices],
-                y_sorted_indices_chunk[y_interval_indices],
-                res,
-            )
 
+    # Remove spaces between Chinese characters
+    text_without_spaces = re.sub(
+        r"(?<=[\u4e00-\u9fff])\s+(?=[\u4e00-\u9fff])", "", input_text
+    )
 
-def sort_by_xycut(
-    block_bboxes: Union[np.ndarray, List[List[int]]],
-    direction: int = 0,
-    min_gap: int = 1,
-) -> List[int]:
-    """
-    Sort bounding boxes using recursive XY cut method based on the specified direction.
+    # Ensure single space between Chinese and non-Chinese characters
+    text_with_single_spaces = re.sub(
+        r"(?<=[\u4e00-\u9fff])\s+(?=[^\u4e00-\u9fff])|(?<=[^\u4e00-\u9fff])\s+(?=[\u4e00-\u9fff])",
+        " ",
+        text_without_spaces,
+    )
 
-    Args:
-        block_bboxes (Union[np.ndarray, List[List[int]]]): An array or list of bounding boxes,
-                                                           where each box is represented as
-                                                           [x_min, y_min, x_max, y_max].
-        direction (int): Direction for the initial cut. Use 1 for Y-axis first and 0 for X-axis first.
-                         Defaults to 0.
-        min_gap (int): Minimum gap width to consider a separation between segments. Defaults to 1.
+    # Reduce any remaining consecutive spaces to a single space
+    final_text = re.sub(r"\s+", " ", text_with_single_spaces).strip()
 
-    Returns:
-        List[int]: A list of indices representing the order of sorted bounding boxes.
-    """
-    block_bboxes = np.asarray(block_bboxes).astype(int)
-    res = []
-    if direction == 1:
-        _recursive_yx_cut(
-            block_bboxes,
-            np.arange(len(block_bboxes)).tolist(),
-            res,
-            min_gap,
-        )
-    else:
-        _recursive_xy_cut(
-            block_bboxes,
-            np.arange(len(block_bboxes)).tolist(),
-            res,
-            min_gap,
-        )
-    return res
+    return final_text
 
 
 def gather_imgs(original_img, layout_det_objs):
     imgs_in_doc = []
     for det_obj in layout_det_objs:
-        if det_obj["label"] in ("image", "chart"):
+        if det_obj["label"] in BLOCK_LABEL_MAP["image_labels"]:
+            label = det_obj["label"]
             x_min, y_min, x_max, y_max = list(map(int, det_obj["coordinate"]))
-            img_path = f"imgs/img_in_table_box_{x_min}_{y_min}_{x_max}_{y_max}.jpg"
+            img_path = f"imgs/img_in_{label}_box_{x_min}_{y_min}_{x_max}_{y_max}.jpg"
             img = Image.fromarray(original_img[y_min:y_max, x_min:x_max, ::-1])
             imgs_in_doc.append(
                 {
@@ -1008,10 +651,10 @@ def _get_minbox_if_overlap_by_ratio(
             The selected bounding box or None if the overlap ratio is not exceeded.
     """
     # Calculate the areas of both bounding boxes
-    area1 = (bbox1[2] - bbox1[0]) * (bbox1[3] - bbox1[1])
-    area2 = (bbox2[2] - bbox2[0]) * (bbox2[3] - bbox2[1])
+    area1 = caculate_bbox_area(bbox1)
+    area2 = caculate_bbox_area(bbox2)
     # Calculate the overlap ratio using a helper function
-    overlap_ratio = _calculate_overlap_area_div_minbox_area_ratio(bbox1, bbox2)
+    overlap_ratio = calculate_overlap_ratio(bbox1, bbox2, mode="small")
     # Check if the overlap ratio exceeds the threshold
     if overlap_ratio > ratio:
         if (area1 <= area2 and smaller) or (area1 >= area2 and not smaller):
@@ -1021,7 +664,7 @@ def _get_minbox_if_overlap_by_ratio(
     return None
 
 
-def _remove_overlap_blocks(
+def remove_overlap_blocks(
     blocks: List[Dict[str, List[int]]], threshold: float = 0.65, smaller: bool = True
 ) -> Tuple[List[Dict[str, List[int]]], List[Dict[str, List[int]]]]:
     """
@@ -1036,13 +679,13 @@ def _remove_overlap_blocks(
         Tuple[List[Dict[str, List[int]]], List[Dict[str, List[int]]]]:
             A tuple containing the updated list of blocks and a list of dropped blocks.
     """
-    dropped_blocks = []
     dropped_indexes = set()
-
+    blocks = deepcopy(blocks)
+    overlap_image_blocks = []
     # Iterate over each pair of blocks to find overlaps
-    for i, block1 in enumerate(blocks):
-        for j in range(i + 1, len(blocks)):
-            block2 = blocks[j]
+    for i, block1 in enumerate(blocks["boxes"]):
+        for j in range(i + 1, len(blocks["boxes"])):
+            block2 = blocks["boxes"][j]
             # Skip blocks that are already marked for removal
             if i in dropped_indexes or j in dropped_indexes:
                 continue
@@ -1054,1332 +697,255 @@ def _remove_overlap_blocks(
                 smaller=smaller,
             )
             if overlap_box_index is not None:
-                # Determine which block to remove based on overlap_box_index
-                if overlap_box_index == 1:
-                    drop_index = i
+                is_block1_image = block1["label"] == "image"
+                is_block2_image = block2["label"] == "image"
+
+                if is_block1_image != is_block2_image:
+                    # 如果只有一个块在视觉标签中，删除在视觉标签中的那个块
+                    drop_index = i if is_block1_image else j
+                    overlap_image_blocks.append(blocks["boxes"][drop_index])
                 else:
-                    drop_index = j
+                    # 如果两个块都在或都不在视觉标签中，根据 overlap_box_index 决定删除哪个块
+                    drop_index = i if overlap_box_index == 1 else j
+
                 dropped_indexes.add(drop_index)
 
     # Remove marked blocks from the original list
     for index in sorted(dropped_indexes, reverse=True):
-        dropped_blocks.append(blocks[index])
-        del blocks[index]
-
-    return blocks, dropped_blocks
-
+        del blocks["boxes"][index]
 
-def _get_text_median_width(blocks: List[Dict[str, any]]) -> float:
-    """
-    Calculate the median width of blocks labeled as "text".
-
-    Args:
-        blocks (List[Dict[str, any]]): List of block dictionaries, each containing a 'block_bbox' and 'label'.
-
-    Returns:
-        float: The median width of text blocks, or infinity if no text blocks are found.
-    """
-    widths = [
-        block["block_bbox"][2] - block["block_bbox"][0]
-        for block in blocks
-        if block.get("block_label") == "text"
-    ]
-    return np.median(widths) if widths else float("inf")
+    return blocks
 
 
-def _get_layout_property(
-    blocks: List[Dict[str, any]],
-    median_width: float,
-    no_mask_labels: List[str],
-    threshold: float = 0.8,
-) -> Tuple[List[Dict[str, any]], bool]:
+def get_bbox_intersection(bbox1, bbox2, return_format="bbox"):
     """
-    Determine the layout (single or double column) of text blocks.
+    Compute the intersection of two bounding boxes, supporting both 4-coordinate and 8-coordinate formats.
 
     Args:
-        blocks (List[Dict[str, any]]): List of block dictionaries containing 'label' and 'block_bbox'.
-        median_width (float): Median width of text blocks.
-        no_mask_labels (List[str]): Labels of blocks to be considered for layout analysis.
-        threshold (float): Threshold for determining layout overlap.
+        bbox1 (tuple): The first bounding box, either in 4-coordinate format (x_min, y_min, x_max, y_max)
+                       or 8-coordinate format (x1, y1, x2, y2, x3, y3, x4, y4).
+        bbox2 (tuple): The second bounding box in the same format as bbox1.
+        return_format (str): The format of the output intersection, either 'bbox' or 'poly'.
 
     Returns:
-        Tuple[List[Dict[str, any]], bool]: Updated list of blocks with layout information and a boolean
-        indicating if the double layout area is greater than the single layout area.
-    """
-    blocks.sort(
-        key=lambda x: (
-            x["block_bbox"][0],
-            (x["block_bbox"][2] - x["block_bbox"][0]),
-        ),
-    )
-    check_single_layout = {}
-    page_min_x, page_max_x = float("inf"), 0
-    double_label_area = 0
-    single_label_area = 0
-
-    for i, block in enumerate(blocks):
-        page_min_x = min(page_min_x, block["block_bbox"][0])
-        page_max_x = max(page_max_x, block["block_bbox"][2])
-    page_width = page_max_x - page_min_x
-
-    for i, block in enumerate(blocks):
-        if block["block_label"] not in no_mask_labels:
-            continue
-
-        x_min_i, _, x_max_i, _ = block["block_bbox"]
-        layout_length = x_max_i - x_min_i
-        cover_count, cover_with_threshold_count = 0, 0
-        match_block_with_threshold_indexes = []
-
-        for j, other_block in enumerate(blocks):
-            if i == j or other_block["block_label"] not in no_mask_labels:
-                continue
-
-            x_min_j, _, x_max_j, _ = other_block["block_bbox"]
-            x_match_min, x_match_max = max(
-                x_min_i,
-                x_min_j,
-            ), min(x_max_i, x_max_j)
-            match_block_iou = (x_match_max - x_match_min) / (x_max_j - x_min_j)
-
-            if match_block_iou > 0:
-                cover_count += 1
-                if match_block_iou > threshold:
-                    cover_with_threshold_count += 1
-                    match_block_with_threshold_indexes.append(
-                        (j, match_block_iou),
-                    )
-                x_min_i = x_match_max
-                if x_min_i >= x_max_i:
-                    break
-
-        if (
-            layout_length > median_width * 1.3
-            and (cover_with_threshold_count >= 2 or cover_count >= 2)
-        ) or layout_length > 0.6 * page_width:
-            # if layout_length > median_width * 1.3 and (cover_with_threshold_count >= 2):
-            block["layout"] = "double"
-            double_label_area += (block["block_bbox"][2] - block["block_bbox"][0]) * (
-                block["block_bbox"][3] - block["block_bbox"][1]
-            )
-        else:
-            block["layout"] = "single"
-            check_single_layout[i] = match_block_with_threshold_indexes
-
-    # Check single-layout block
-    for i, single_layout in check_single_layout.items():
-        if single_layout:
-            index, match_iou = single_layout[-1]
-            if match_iou > 0.9 and blocks[index]["layout"] == "double":
-                blocks[i]["layout"] = "double"
-                double_label_area += (
-                    blocks[i]["block_bbox"][2] - blocks[i]["block_bbox"][0]
-                ) * (blocks[i]["block_bbox"][3] - blocks[i]["block_bbox"][1])
-            else:
-                single_label_area += (
-                    blocks[i]["block_bbox"][2] - blocks[i]["block_bbox"][0]
-                ) * (blocks[i]["block_bbox"][3] - blocks[i]["block_bbox"][1])
-
-    return blocks, (double_label_area > single_label_area)
-
-
-def _get_bbox_direction(input_bbox: List[float], ratio: float = 1.0) -> bool:
-    """
-    Determine if a bounding box is horizontal or vertical.
-
-    Args:
-        input_bbox (List[float]): Bounding box [x_min, y_min, x_max, y_max].
-        ratio (float): Ratio for determining orientation. Default is 1.0.
-
-    Returns:
-        bool: True if the bounding box is considered horizontal, False if vertical.
-    """
-    width = input_bbox[2] - input_bbox[0]
-    height = input_bbox[3] - input_bbox[1]
-    return width * ratio >= height
-
-
-def _get_projection_iou(
-    input_bbox: List[float], match_bbox: List[float], is_horizontal: bool = True
-) -> float:
-    """
-    Calculate the IoU of lines between two bounding boxes.
-
-    Args:
-        input_bbox (List[float]): First bounding box [x_min, y_min, x_max, y_max].
-        match_bbox (List[float]): Second bounding box [x_min, y_min, x_max, y_max].
-        is_horizontal (bool): Whether to compare horizontally or vertically.
-
-    Returns:
-        float: Line IoU. Returns 0 if there is no overlap.
-    """
-    if is_horizontal:
-        x_match_min = max(input_bbox[0], match_bbox[0])
-        x_match_max = min(input_bbox[2], match_bbox[2])
-        overlap = max(0, x_match_max - x_match_min)
-        input_width = min(input_bbox[2] - input_bbox[0], match_bbox[2] - match_bbox[0])
+        tuple or None: The intersection bounding box in the specified format, or None if there is no intersection.
+    """
+    bbox1 = np.array(bbox1)
+    bbox2 = np.array(bbox2)
+    # Convert both bounding boxes to rectangles
+    rect1 = bbox1 if len(bbox1.shape) == 1 else convert_points_to_boxes([bbox1])[0]
+    rect2 = bbox2 if len(bbox2.shape) == 1 else convert_points_to_boxes([bbox2])[0]
+
+    # Calculate the intersection rectangle
+
+    x_min_inter = max(rect1[0], rect2[0])
+    y_min_inter = max(rect1[1], rect2[1])
+    x_max_inter = min(rect1[2], rect2[2])
+    y_max_inter = min(rect1[3], rect2[3])
+
+    # Check if there is an intersection
+    if x_min_inter >= x_max_inter or y_min_inter >= y_max_inter:
+        return None
+
+    if return_format == "bbox":
+        return np.array([x_min_inter, y_min_inter, x_max_inter, y_max_inter])
+    elif return_format == "poly":
+        return np.array(
+            [
+                [x_min_inter, y_min_inter],
+                [x_max_inter, y_min_inter],
+                [x_max_inter, y_max_inter],
+                [x_min_inter, y_max_inter],
+            ],
+            dtype=np.int16,
+        )
     else:
-        y_match_min = max(input_bbox[1], match_bbox[1])
-        y_match_max = min(input_bbox[3], match_bbox[3])
-        overlap = max(0, y_match_max - y_match_min)
-        input_width = min(input_bbox[3] - input_bbox[1], match_bbox[3] - match_bbox[1])
-
-    return overlap / input_width if input_width > 0 else 0.0
+        raise ValueError("return_format must be either 'bbox' or 'poly'.")
 
 
-def _get_sub_category(
-    blocks: List[Dict[str, Any]], title_labels: List[str]
-) -> Tuple[List[Dict[str, Any]], List[float]]:
+def shrink_supplement_region_bbox(
+    supplement_region_bbox,
+    ref_region_bbox,
+    image_width,
+    image_height,
+    block_idxes_set,
+    block_bboxes,
+) -> List:
     """
-    Determine the layout of title and text blocks and collect pre_cuts.
+    Shrink the supplement region bbox according to the reference region bbox and match the block bboxes.
 
     Args:
-        blocks (List[Dict[str, Any]]): List of block dictionaries.
-        title_labels (List[str]): List of labels considered as titles.
+        supplement_region_bbox (list): The supplement region bbox.
+        ref_region_bbox (list): The reference region bbox.
+        image_width (int): The width of the image.
+        image_height (int): The height of the image.
+        block_idxes_set (set): The indexes of the blocks that intersect with the region bbox.
+        block_bboxes (dict): The dictionary of block bboxes.
 
     Returns:
-        List[Dict[str, Any]]: Updated list of blocks with title-text layout information.
-        Dict[float]: Dict of pre_cuts coordinates.
-    """
-
-    sub_title_labels = ["paragraph_title"]
-    vision_labels = ["image", "table", "chart", "figure"]
-    vision_title_labels = ["figure_title", "chart_title", "table_title"]
-    all_labels = title_labels + sub_title_labels + vision_labels + vision_title_labels
-    special_pre_cut_labels = sub_title_labels
-
-    # single doc title is irregular,pre cut not applicable
-    num_doc_title = 0
-    for block in blocks:
-        if block["block_label"] == "doc_title":
-            num_doc_title += 1
-            if num_doc_title == 2:
-                special_pre_cut_labels = title_labels + sub_title_labels
-                break
-    if len(blocks) == 0:
-        return blocks, {}
-
-    min_x = min(block["block_bbox"][0] for block in blocks)
-    min_y = min(block["block_bbox"][1] for block in blocks)
-    max_x = max(block["block_bbox"][2] for block in blocks)
-    max_y = max(block["block_bbox"][3] for block in blocks)
-    region_bbox = (min_x, min_y, max_x, max_y)
-    region_x_center = (region_bbox[0] + region_bbox[2]) / 2
-    region_y_center = (region_bbox[1] + region_bbox[3]) / 2
-    region_width = region_bbox[2] - region_bbox[0]
-    region_height = region_bbox[3] - region_bbox[1]
-
-    pre_cuts = {}
-
-    for i, block1 in enumerate(blocks):
-        block1.setdefault("title_text", [])
-        block1.setdefault("sub_title", [])
-        block1.setdefault("vision_footnote", [])
-        block1.setdefault("sub_label", block1["block_label"])
-
-        if block1["block_label"] not in all_labels:
-            continue
-
-        bbox1 = block1["block_bbox"]
-        x1, y1, x2, y2 = bbox1
-        is_horizontal_1 = _get_bbox_direction(block1["block_bbox"])
-        left_up_title_text_distance = float("inf")
-        left_up_title_text_index = -1
-        left_up_title_text_direction = None
-        right_down_title_text_distance = float("inf")
-        right_down_title_text_index = -1
-        right_down_title_text_direction = None
-
-        # pre-cuts
-        # Condition 1: Length is greater than half of the layout region
-        if is_horizontal_1:
-            block_length = x2 - x1
-            required_length = region_width / 2
-        else:
-            block_length = y2 - y1
-            required_length = region_height / 2
-        if block1["block_label"] in special_pre_cut_labels:
-            length_condition = True
-        else:
-            length_condition = block_length > required_length
-
-        # Condition 2: Centered check (must be within ±20 in both horizontal and vertical directions)
-        block_x_center = (x1 + x2) / 2
-        block_y_center = (y1 + y2) / 2
-        tolerance_len = block_length // 5
-        if block1["block_label"] in special_pre_cut_labels:
-            tolerance_len = block_length // 10
-        if is_horizontal_1:
-            is_centered = abs(block_x_center - region_x_center) <= tolerance_len
-        else:
-            is_centered = abs(block_y_center - region_y_center) <= tolerance_len
-
-        # Condition 3: Check for surrounding text
-        has_left_text = False
-        has_right_text = False
-        has_above_text = False
-        has_below_text = False
-        for block2 in blocks:
-            if block2["block_label"] != "text":
-                continue
-            bbox2 = block2["block_bbox"]
-            x1_2, y1_2, x2_2, y2_2 = bbox2
-            if is_horizontal_1:
-                if x2_2 <= x1 and not (y2_2 <= y1 or y1_2 >= y2):
-                    has_left_text = True
-                if x1_2 >= x2 and not (y2_2 <= y1 or y1_2 >= y2):
-                    has_right_text = True
-            else:
-                if y2_2 <= y1 and not (x2_2 <= x1 or x1_2 >= x2):
-                    has_above_text = True
-                if y1_2 >= y2 and not (x2_2 <= x1 or x1_2 >= x2):
-                    has_below_text = True
-
-            if (is_horizontal_1 and has_left_text and has_right_text) or (
-                not is_horizontal_1 and has_above_text and has_below_text
-            ):
-                break
-
-        no_text_on_sides = (
-            not (has_left_text or has_right_text)
-            if is_horizontal_1
-            else not (has_above_text or has_below_text)
-        )
-
-        # Add coordinates if all conditions are met
-        if is_centered and length_condition and no_text_on_sides:
-            if is_horizontal_1:
-                pre_cuts.setdefault("y", []).append(y1)
-            else:
-                pre_cuts.setdefault("x", []).append(x1)
-
-        for j, block2 in enumerate(blocks):
-            if i == j:
-                continue
-
-            bbox2 = block2["block_bbox"]
-            x1_prime, y1_prime, x2_prime, y2_prime = bbox2
-            is_horizontal_2 = _get_bbox_direction(bbox2)
-            match_block_iou = _get_projection_iou(
-                bbox2,
-                bbox1,
-                is_horizontal_1,
+        list: The new region bbox and the matched block idxes.
+    """
+    x1, y1, x2, y2 = supplement_region_bbox
+    x1_prime, y1_prime, x2_prime, y2_prime = ref_region_bbox
+    index_conversion_map = {0: 2, 1: 3, 2: 0, 3: 1}
+    edge_distance_list = [
+        (x1_prime - x1) / image_width,
+        (y1_prime - y1) / image_height,
+        (x2 - x2_prime) / image_width,
+        (y2 - y2_prime) / image_height,
+    ]
+    edge_distance_list_tmp = edge_distance_list[:]
+    min_distance = min(edge_distance_list)
+    src_index = index_conversion_map[edge_distance_list.index(min_distance)]
+    if len(block_idxes_set) == 0:
+        return supplement_region_bbox, []
+    for _ in range(3):
+        dst_index = index_conversion_map[src_index]
+        tmp_region_bbox = supplement_region_bbox[:]
+        tmp_region_bbox[dst_index] = ref_region_bbox[src_index]
+        iner_block_idxes, split_block_idxes = [], []
+        for block_idx in block_idxes_set:
+            overlap_ratio = calculate_overlap_ratio(
+                tmp_region_bbox, block_bboxes[block_idx], mode="small"
             )
-
-            def distance_(is_horizontal, is_left_up):
-                if is_horizontal:
-                    if is_left_up:
-                        return (y1 - y2_prime + 2) // 5 + x1_prime / 5000
-                    else:
-                        return (y1_prime - y2 + 2) // 5 + x1_prime / 5000
-
-                else:
-                    if is_left_up:
-                        return (x1 - x2_prime + 2) // 5 + y1_prime / 5000
-                    else:
-                        return (x1_prime - x2 + 2) // 5 + y1_prime / 5000
-
-            block_iou_threshold = 0.1
-            if block1["block_label"] in sub_title_labels:
-                block_iou_threshold = 0.5
-
-            if is_horizontal_1:
-                if match_block_iou >= block_iou_threshold:
-                    left_up_distance = distance_(True, True)
-                    right_down_distance = distance_(True, False)
-                    if (
-                        y2_prime <= y1
-                        and left_up_distance <= left_up_title_text_distance
-                    ):
-                        left_up_title_text_distance = left_up_distance
-                        left_up_title_text_index = j
-                        left_up_title_text_direction = is_horizontal_2
-                    elif (
-                        y1_prime > y2
-                        and right_down_distance < right_down_title_text_distance
-                    ):
-                        right_down_title_text_distance = right_down_distance
-                        right_down_title_text_index = j
-                        right_down_title_text_direction = is_horizontal_2
-            else:
-                if match_block_iou >= block_iou_threshold:
-                    left_up_distance = distance_(False, True)
-                    right_down_distance = distance_(False, False)
-                    if (
-                        x2_prime <= x1
-                        and left_up_distance <= left_up_title_text_distance
-                    ):
-                        left_up_title_text_distance = left_up_distance
-                        left_up_title_text_index = j
-                        left_up_title_text_direction = is_horizontal_2
-                    elif (
-                        x1_prime > x2
-                        and right_down_distance < right_down_title_text_distance
-                    ):
-                        right_down_title_text_distance = right_down_distance
-                        right_down_title_text_index = j
-                        right_down_title_text_direction = is_horizontal_2
-
-        height = bbox1[3] - bbox1[1]
-        width = bbox1[2] - bbox1[0]
-        title_text_weight = [0.8, 0.8]
-
-        title_text, sub_title, vision_footnote = [], [], []
-
-        def get_sub_category_(
-            title_text_direction,
-            title_text_index,
-            label,
-            is_left_up=True,
-        ):
-            direction_ = [1, 3] if is_left_up else [2, 4]
-            if (
-                title_text_direction == is_horizontal_1
-                and title_text_index != -1
-                and (label == "text" or label == "paragraph_title")
+            if overlap_ratio > REGION_SETTINGS.get(
+                "match_block_overlap_ratio_threshold", 0.8
             ):
-                bbox2 = blocks[title_text_index]["block_bbox"]
-                if is_horizontal_1:
-                    height1 = bbox2[3] - bbox2[1]
-                    width1 = bbox2[2] - bbox2[0]
-                    if label == "text":
-                        if (
-                            _nearest_edge_distance(bbox1, bbox2)[0] <= 15
-                            and block1["block_label"] in vision_labels
-                            and width1 < width
-                            and height1 < 0.5 * height
-                        ):
-                            blocks[title_text_index]["sub_label"] = "vision_footnote"
-                            vision_footnote.append(bbox2)
-                        elif (
-                            height1 < height * title_text_weight[0]
-                            and (width1 < width or width1 > 1.5 * width)
-                            and block1["block_label"] in title_labels
-                        ):
-                            blocks[title_text_index]["sub_label"] = "title_text"
-                            title_text.append((direction_[0], bbox2))
-                    elif (
-                        label == "paragraph_title"
-                        and block1["block_label"] in sub_title_labels
-                    ):
-                        sub_title.append(bbox2)
-                else:
-                    height1 = bbox2[3] - bbox2[1]
-                    width1 = bbox2[2] - bbox2[0]
-                    if label == "text":
-                        if (
-                            _nearest_edge_distance(bbox1, bbox2)[0] <= 15
-                            and block1["block_label"] in vision_labels
-                            and height1 < height
-                            and width1 < 0.5 * width
-                        ):
-                            blocks[title_text_index]["sub_label"] = "vision_footnote"
-                            vision_footnote.append(bbox2)
-                        elif (
-                            width1 < width * title_text_weight[1]
-                            and block1["block_label"] in title_labels
-                        ):
-                            blocks[title_text_index]["sub_label"] = "title_text"
-                            title_text.append((direction_[1], bbox2))
-                    elif (
-                        label == "paragraph_title"
-                        and block1["block_label"] in sub_title_labels
-                    ):
-                        sub_title.append(bbox2)
-
-        if (
-            is_horizontal_1
-            and abs(left_up_title_text_distance - right_down_title_text_distance) * 5
-            > height
-        ) or (
-            not is_horizontal_1
-            and abs(left_up_title_text_distance - right_down_title_text_distance) * 5
-            > width
-        ):
-            if left_up_title_text_distance < right_down_title_text_distance:
-                get_sub_category_(
-                    left_up_title_text_direction,
-                    left_up_title_text_index,
-                    blocks[left_up_title_text_index]["block_label"],
-                    True,
-                )
-            else:
-                get_sub_category_(
-                    right_down_title_text_direction,
-                    right_down_title_text_index,
-                    blocks[right_down_title_text_index]["block_label"],
-                    False,
-                )
-        else:
-            get_sub_category_(
-                left_up_title_text_direction,
-                left_up_title_text_index,
-                blocks[left_up_title_text_index]["block_label"],
-                True,
-            )
-            get_sub_category_(
-                right_down_title_text_direction,
-                right_down_title_text_index,
-                blocks[right_down_title_text_index]["block_label"],
-                False,
-            )
-
-        if block1["block_label"] in title_labels:
-            if blocks[i].get("title_text") == []:
-                blocks[i]["title_text"] = title_text
-
-        if block1["block_label"] in sub_title_labels:
-            if blocks[i].get("sub_title") == []:
-                blocks[i]["sub_title"] = sub_title
-
-        if block1["block_label"] in vision_labels:
-            if blocks[i].get("vision_footnote") == []:
-                blocks[i]["vision_footnote"] = vision_footnote
-
-    return blocks, pre_cuts
-
-
-def get_layout_ordering(
-    parsing_res_list: List[Dict[str, Any]],
-    no_mask_labels: List[str] = [],
-) -> None:
-    """
-    Process layout parsing results to remove overlapping bounding boxes
-    and assign an ordering index based on their positions.
-
-    Modifies:
-        The 'parsing_res_list' list by adding an 'index' to each block.
-
-    Args:
-        parsing_res_list (List[Dict[str, Any]]): List of block dictionaries with 'block_bbox' and 'block_label'.
-        no_mask_labels (List[str]): Labels for which overlapping removal is not performed.
-    """
-    title_text_labels = ["doc_title"]
-    title_labels = ["doc_title", "paragraph_title"]
-    vision_labels = ["image", "table", "seal", "chart", "figure"]
-    vision_title_labels = ["table_title", "chart_title", "figure_title"]
-
-    parsing_res_list, pre_cuts = _get_sub_category(parsing_res_list, title_text_labels)
-
-    parsing_res_by_pre_cuts_list = []
-    if len(pre_cuts) > 0:
-        block_bboxes = [block["block_bbox"] for block in parsing_res_list]
-        for axis, cuts in pre_cuts.items():
-            axis_index = 1 if axis == "y" else 0
-
-            max_val = max(bbox[axis_index + 2] for bbox in block_bboxes)
-
-            intervals = []
-            prev = 0
-            for cut in sorted(cuts):
-                intervals.append((prev, cut))
-                prev = cut
-            intervals.append((prev, max_val))
-
-            for start, end in intervals:
-                mask = [
-                    (bbox[axis_index] >= start) and (bbox[axis_index] < end)
-                    for bbox in block_bboxes
-                ]
-                parsing_res_by_pre_cuts_list.append(
-                    [parsing_res_list[i] for i, m in enumerate(mask) if m]
-                )
-    else:
-        parsing_res_by_pre_cuts_list = [parsing_res_list]
-
-    final_parsing_res_list = []
-    num_index = 0
-    num_sub_index = 0
-    for parsing_res_by_pre_cuts in parsing_res_by_pre_cuts_list:
-
-        doc_flag = False
-        median_width = _get_text_median_width(parsing_res_by_pre_cuts)
-        parsing_res_by_pre_cuts, projection_direction = _get_layout_property(
-            parsing_res_by_pre_cuts,
-            median_width,
-            no_mask_labels=no_mask_labels,
-            threshold=0.3,
-        )
-        # Convert bounding boxes to float and remove overlaps
-        (
-            double_text_blocks,
-            title_text_blocks,
-            title_blocks,
-            vision_blocks,
-            vision_title_blocks,
-            vision_footnote_blocks,
-            other_blocks,
-        ) = ([], [], [], [], [], [], [])
-
-        drop_indexes = []
-
-        for index, block in enumerate(parsing_res_by_pre_cuts):
-            label = block["sub_label"]
-            block["block_bbox"] = list(map(int, block["block_bbox"]))
-
-            if label == "doc_title":
-                doc_flag = True
-
-            if label in no_mask_labels:
-                if block["layout"] == "double":
-                    double_text_blocks.append(block)
-                    drop_indexes.append(index)
-            elif label == "title_text":
-                title_text_blocks.append(block)
-                drop_indexes.append(index)
-            elif label == "vision_footnote":
-                vision_footnote_blocks.append(block)
-                drop_indexes.append(index)
-            elif label in vision_title_labels:
-                vision_title_blocks.append(block)
-                drop_indexes.append(index)
-            elif label in title_labels:
-                title_blocks.append(block)
-                drop_indexes.append(index)
-            elif label in vision_labels:
-                vision_blocks.append(block)
-                drop_indexes.append(index)
-            else:
-                other_blocks.append(block)
-                drop_indexes.append(index)
-
-        for index in sorted(drop_indexes, reverse=True):
-            del parsing_res_by_pre_cuts[index]
-
-        if len(parsing_res_by_pre_cuts) > 0:
-            # single text label
-            if (
-                len(double_text_blocks) > len(parsing_res_by_pre_cuts)
-                or projection_direction
+                iner_block_idxes.append(block_idx)
+            elif overlap_ratio > REGION_SETTINGS.get(
+                "split_block_overlap_ratio_threshold", 0.4
             ):
-                parsing_res_by_pre_cuts.extend(title_blocks + double_text_blocks)
-                title_blocks = []
-                double_text_blocks = []
-                block_bboxes = [
-                    block["block_bbox"] for block in parsing_res_by_pre_cuts
-                ]
-                block_bboxes.sort(
-                    key=lambda x: (
-                        x[0] // max(20, median_width),
-                        x[1],
-                    ),
-                )
-                block_bboxes = np.array(block_bboxes)
-                sorted_indices = sort_by_xycut(block_bboxes, direction=1, min_gap=1)
-            else:
-                block_bboxes = [
-                    block["block_bbox"] for block in parsing_res_by_pre_cuts
-                ]
-                block_bboxes.sort(key=lambda x: (x[0] // 20, x[1]))
-                block_bboxes = np.array(block_bboxes)
-                sorted_indices = sort_by_xycut(block_bboxes, direction=0, min_gap=20)
-
-            sorted_boxes = block_bboxes[sorted_indices].tolist()
-
-            for block in parsing_res_by_pre_cuts:
-                block["index"] = num_index + sorted_boxes.index(block["block_bbox"]) + 1
-                block["sub_index"] = (
-                    num_sub_index + sorted_boxes.index(block["block_bbox"]) + 1
-                )
-
-        def nearest_match_(input_blocks, distance_type="manhattan", is_add_index=True):
-            for block in input_blocks:
-                bbox = block["block_bbox"]
-                min_distance = float("inf")
-                min_distance_config = [
-                    [float("inf"), float("inf")],
-                    float("inf"),
-                    float("inf"),
-                ]  # for double text
-                nearest_gt_index = 0
-                for match_block in parsing_res_by_pre_cuts:
-                    match_bbox = match_block["block_bbox"]
-                    if distance_type == "nearest_iou_edge_distance":
-                        distance, min_distance_config = _nearest_iou_edge_distance(
-                            bbox,
-                            match_bbox,
-                            block["sub_label"],
-                            vision_labels=vision_labels,
-                            no_mask_labels=no_mask_labels,
-                            median_width=median_width,
-                            title_labels=title_labels,
-                            title_text=block["title_text"],
-                            sub_title=block["sub_title"],
-                            min_distance_config=min_distance_config,
-                            tolerance_len=10,
-                        )
-                    elif distance_type == "title_text":
-                        if (
-                            match_block["block_label"] in title_labels + ["abstract"]
-                            and match_block["title_text"] != []
-                        ):
-                            iou_left_up = _calculate_overlap_area_div_minbox_area_ratio(
-                                bbox,
-                                match_block["title_text"][0][1],
-                            )
-                            iou_right_down = (
-                                _calculate_overlap_area_div_minbox_area_ratio(
-                                    bbox,
-                                    match_block["title_text"][-1][1],
-                                )
-                            )
-                            iou = 1 - max(iou_left_up, iou_right_down)
-                            distance = _manhattan_distance(bbox, match_bbox) * iou
-                        else:
-                            distance = float("inf")
-                    elif distance_type == "manhattan":
-                        distance = _manhattan_distance(bbox, match_bbox)
-                    elif distance_type == "vision_footnote":
-                        if (
-                            match_block["block_label"] in vision_labels
-                            and match_block["vision_footnote"] != []
-                        ):
-                            iou_left_up = _calculate_overlap_area_div_minbox_area_ratio(
-                                bbox,
-                                match_block["vision_footnote"][0],
-                            )
-                            iou_right_down = (
-                                _calculate_overlap_area_div_minbox_area_ratio(
-                                    bbox,
-                                    match_block["vision_footnote"][-1],
-                                )
-                            )
-                            iou = 1 - max(iou_left_up, iou_right_down)
-                            distance = _manhattan_distance(bbox, match_bbox) * iou
-                        else:
-                            distance = float("inf")
-                    elif distance_type == "vision_body":
-                        if (
-                            match_block["block_label"] in vision_title_labels
-                            and block["vision_footnote"] != []
-                        ):
-                            iou_left_up = _calculate_overlap_area_div_minbox_area_ratio(
-                                match_bbox,
-                                block["vision_footnote"][0],
-                            )
-                            iou_right_down = (
-                                _calculate_overlap_area_div_minbox_area_ratio(
-                                    match_bbox,
-                                    block["vision_footnote"][-1],
-                                )
-                            )
-                            iou = 1 - max(iou_left_up, iou_right_down)
-                            distance = _manhattan_distance(bbox, match_bbox) * iou
-                        else:
-                            distance = float("inf")
-                    # when reference block cross mulitple columns, its order should be after the blocks above it.
-                    elif distance_type == "append":
-                        if match_bbox[3] <= bbox[1]:
-                            distance = -(match_bbox[2] * 10 + match_bbox[3])
-                        else:
-                            distance = float("inf")
-                    else:
-                        raise NotImplementedError
-
-                    if distance < min_distance:
-                        min_distance = distance
-                        if is_add_index:
-                            nearest_gt_index = match_block.get("index", 999)
-                        else:
-                            nearest_gt_index = match_block.get("sub_index", 999)
-
-                if is_add_index:
-                    block["index"] = nearest_gt_index
-                else:
-                    block["sub_index"] = nearest_gt_index
-
-                parsing_res_by_pre_cuts.append(block)
-
-        # double text label
-        double_text_blocks.sort(
-            key=lambda x: (
-                x["block_bbox"][1] // 10,
-                x["block_bbox"][0] // median_width,
-                x["block_bbox"][1] ** 2 + x["block_bbox"][0] ** 2,
-            ),
-        )
-        # filter the reference blocks from all blocks that cross mulitple columns.
-        # they should be ordered using "append".
-        double_text_reference_blocks = []
-        i = 0
-        while i < len(double_text_blocks):
-            if double_text_blocks[i]["block_label"] == "reference":
-                double_text_reference_blocks.append(double_text_blocks.pop(i))
-            else:
-                i += 1
-        nearest_match_(
-            double_text_blocks,
-            distance_type="nearest_iou_edge_distance",
-        )
-        nearest_match_(
-            double_text_reference_blocks,
-            distance_type="append",
-        )
-        parsing_res_by_pre_cuts.sort(
-            key=lambda x: (x["index"], x["block_bbox"][1], x["block_bbox"][0]),
-        )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["index"] = num_index + idx + 1
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # title label
-        title_blocks.sort(
-            key=lambda x: (
-                x["block_bbox"][1] // 10,
-                x["block_bbox"][0] // median_width,
-                x["block_bbox"][1] ** 2 + x["block_bbox"][0] ** 2,
-            ),
-        )
-        nearest_match_(title_blocks, distance_type="nearest_iou_edge_distance")
-
-        if doc_flag:
-            text_sort_labels = ["doc_title"]
-            text_label_priority = {
-                label: priority for priority, label in enumerate(text_sort_labels)
-            }
-            doc_titles = []
-            for i, block in enumerate(parsing_res_by_pre_cuts):
-                if block["block_label"] == "doc_title":
-                    doc_titles.append(
-                        (i, block["block_bbox"][1], block["block_bbox"][0]),
+                split_block_idxes.append(block_idx)
+
+        if len(iner_block_idxes) > 0:
+            if len(split_block_idxes) > 0:
+                for split_block_idx in split_block_idxes:
+                    split_block_bbox = block_bboxes[split_block_idx]
+                    x1, y1, x2, y2 = tmp_region_bbox
+                    x1_prime, y1_prime, x2_prime, y2_prime = split_block_bbox
+                    edge_distance_list = [
+                        (x1_prime - x1) / image_width,
+                        (y1_prime - y1) / image_height,
+                        (x2 - x2_prime) / image_width,
+                        (y2 - y2_prime) / image_height,
+                    ]
+                    max_distance = max(edge_distance_list)
+                    src_index = edge_distance_list.index(max_distance)
+                    dst_index = index_conversion_map[src_index]
+                    tmp_region_bbox[dst_index] = split_block_bbox[src_index]
+                    tmp_region_bbox, iner_idxes = shrink_supplement_region_bbox(
+                        tmp_region_bbox,
+                        ref_region_bbox,
+                        image_width,
+                        image_height,
+                        iner_block_idxes,
+                        block_bboxes,
                     )
-            doc_titles.sort(key=lambda x: (x[1], x[2]))
-            first_doc_title_index = doc_titles[0][0]
-            parsing_res_by_pre_cuts[first_doc_title_index]["index"] = 1
-            parsing_res_by_pre_cuts.sort(
-                key=lambda x: (
-                    x["index"],
-                    text_label_priority.get(x["block_label"], 9999),
-                    x["block_bbox"][1],
-                    x["block_bbox"][0],
-                ),
-            )
+                    if len(iner_idxes) == 0:
+                        continue
+            matched_bboxes = [block_bboxes[idx] for idx in iner_block_idxes]
+            supplement_region_bbox = calculate_minimum_enclosing_bbox(matched_bboxes)
+            break
         else:
-            parsing_res_by_pre_cuts.sort(
-                key=lambda x: (
-                    x["index"],
-                    x["block_bbox"][1],
-                    x["block_bbox"][0],
-                ),
-            )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["index"] = num_index + idx + 1
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # title-text label
-        nearest_match_(title_text_blocks, distance_type="title_text")
-
-        def hor_tb_and_ver_lr(x):
-            input_bbox = x["block_bbox"]
-            is_horizontal = _get_bbox_direction(input_bbox)
-            if is_horizontal:
-                return input_bbox[1]
-            else:
-                return input_bbox[0]
-
-        parsing_res_by_pre_cuts.sort(
-            key=lambda x: (x["index"], hor_tb_and_ver_lr(x)),
-        )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["index"] = num_index + idx + 1
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # image,figure,chart,seal label
-        nearest_match_(
-            vision_blocks,
-            distance_type="nearest_iou_edge_distance",
-            is_add_index=False,
-        )
-        parsing_res_by_pre_cuts.sort(
-            key=lambda x: (
-                x["sub_index"],
-                x["block_bbox"][1],
-                x["block_bbox"][0],
-            ),
-        )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # image,figure,chart,seal title label
-        nearest_match_(
-            vision_title_blocks,
-            distance_type="nearest_iou_edge_distance",
-            is_add_index=False,
-        )
-        parsing_res_by_pre_cuts.sort(
-            key=lambda x: (
-                x["sub_index"],
-                x["block_bbox"][1],
-                x["block_bbox"][0],
-            ),
-        )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # vision footnote label
-        nearest_match_(
-            vision_footnote_blocks,
-            distance_type="vision_footnote",
-            is_add_index=False,
-        )
-        text_label_priority = {"vision_footnote": 9999}
-        parsing_res_by_pre_cuts.sort(
-            key=lambda x: (
-                x["sub_index"],
-                text_label_priority.get(x["sub_label"], 0),
-                x["block_bbox"][1],
-                x["block_bbox"][0],
-            ),
-        )
-
-        for idx, block in enumerate(parsing_res_by_pre_cuts):
-            block["sub_index"] = num_sub_index + idx + 1
-
-        # header、footnote、header_image... label
-        nearest_match_(other_blocks, distance_type="manhattan", is_add_index=False)
-
-        # add all parsing result
-        final_parsing_res_list.extend(parsing_res_by_pre_cuts)
-
-        # update num index
-        num_sub_index += len(parsing_res_by_pre_cuts)
-        for parsing_res in parsing_res_by_pre_cuts:
-            if parsing_res.get("index"):
-                num_index += 1
-
-    parsing_res_list = [
-        {
-            "block_label": parsing_res["block_label"],
-            "block_content": parsing_res["block_content"],
-            "block_bbox": parsing_res["block_bbox"],
-            "block_image": parsing_res.get("block_image", None),
-            "sub_label": parsing_res["sub_label"],
-            "sub_index": parsing_res["sub_index"],
-            "index": parsing_res.get("index", None),
-            "seg_start_coordinate": parsing_res.get(
-                "seg_start_coordinate", float("inf")
-            ),
-            "seg_end_coordinate": parsing_res.get("seg_end_coordinate", float("-inf")),
-            "num_of_lines": parsing_res.get("num_of_lines", 1),
-        }
-        for parsing_res in final_parsing_res_list
-    ]
-
-    return parsing_res_list
-
-
-def _manhattan_distance(
-    point1: Tuple[float, float],
-    point2: Tuple[float, float],
-    weight_x: float = 1.0,
-    weight_y: float = 1.0,
-) -> float:
-    """
-    Calculate the weighted Manhattan distance between two points.
-
-    Args:
-        point1 (Tuple[float, float]): The first point as (x, y).
-        point2 (Tuple[float, float]): The second point as (x, y).
-        weight_x (float): The weight for the x-axis distance. Default is 1.0.
-        weight_y (float): The weight for the y-axis distance. Default is 1.0.
-
-    Returns:
-        float: The weighted Manhattan distance between the two points.
-    """
-    return weight_x * abs(point1[0] - point2[0]) + weight_y * abs(point1[1] - point2[1])
-
-
-def _calculate_horizontal_distance(
-    input_bbox: List[int],
-    match_bbox: List[int],
-    height: int,
-    disperse: int,
-    title_text: List[Tuple[int, List[int]]],
-) -> float:
-    """
-    Calculate the horizontal distance between two bounding boxes, considering title text adjustments.
-
-    Args:
-        input_bbox (List[int]): The bounding box coordinates [x1, y1, x2, y2] of the input object.
-        match_bbox (List[int]): The bounding box coordinates [x1', y1', x2', y2'] of the object to match against.
-        height (int): The height of the input bounding box used for normalization.
-        disperse (int): The dispersion factor used to normalize the horizontal distance.
-        title_text (List[Tuple[int, List[int]]]): A list of tuples containing title text information and their bounding box coordinates.
-                                                  Format: [(position_indicator, [x1, y1, x2, y2]), ...].
-
-    Returns:
-        float: The calculated horizontal distance taking into account the title text adjustments.
-    """
-    x1, y1, x2, y2 = input_bbox
-    x1_prime, y1_prime, x2_prime, y2_prime = match_bbox
-
-    # Determine vertical distance adjustment based on title text
-    if y2 < y1_prime:
-        if title_text and title_text[-1][0] == 2:
-            y2 += title_text[-1][1][3] - title_text[-1][1][1]
-        vertical_adjustment = (y1_prime - y2) * 0.5
-    else:
-        if title_text and title_text[0][0] == 1:
-            y1 -= title_text[0][1][3] - title_text[0][1][1]
-        vertical_adjustment = y1 - y2_prime
-
-    # Calculate horizontal distance with adjustments
-    horizontal_distance = (
-        abs(x2_prime - x1) // disperse
-        + vertical_adjustment // height
-        + vertical_adjustment / 5000
-    )
-
-    return horizontal_distance
-
+            edge_distance_list_tmp = [
+                x for x in edge_distance_list_tmp if x != min_distance
+            ]
+            min_distance = min(edge_distance_list_tmp)
+            src_index = index_conversion_map[edge_distance_list.index(min_distance)]
+    return supplement_region_bbox, iner_block_idxes
 
-def _calculate_vertical_distance(
-    input_bbox: List[int],
-    match_bbox: List[int],
-    width: int,
-    disperse: int,
-    title_text: List[Tuple[int, List[int]]],
-) -> float:
-    """
-    Calculate the vertical distance between two bounding boxes, considering title text adjustments.
 
-    Args:
-        input_bbox (List[int]): The bounding box coordinates [x1, y1, x2, y2] of the input object.
-        match_bbox (List[int]): The bounding box coordinates [x1', y1', x2', y2'] of the object to match against.
-        width (int): The width of the input bounding box used for normalization.
-        disperse (int): The dispersion factor used to normalize the vertical distance.
-        title_text (List[Tuple[int, List[int]]]): A list of tuples containing title text information and their bounding box coordinates.
-                                                  Format: [(position_indicator, [x1, y1, x2, y2]), ...].
+def update_region_box(bbox, region_box):
+    if region_box is None:
+        return bbox
 
-    Returns:
-        float: The calculated vertical distance taking into account the title text adjustments.
-    """
-    x1, y1, x2, y2 = input_bbox
-    x1_prime, y1_prime, x2_prime, y2_prime = match_bbox
-
-    # Determine horizontal distance adjustment based on title text
-    if x1 > x2_prime:
-        if title_text and title_text[0][0] == 3:
-            x1 -= title_text[0][1][2] - title_text[0][1][0]
-        horizontal_adjustment = (x1 - x2_prime) * 0.5
-    else:
-        if title_text and title_text[-1][0] == 4:
-            x2 += title_text[-1][1][2] - title_text[-1][1][0]
-        horizontal_adjustment = x1_prime - x2
-
-    # Calculate vertical distance with adjustments
-    vertical_distance = (
-        abs(y2_prime - y1) // disperse
-        + horizontal_adjustment // width
-        + horizontal_adjustment / 5000
-    )
+    x1, y1, x2, y2 = bbox
+    x1_region, y1_region, x2_region, y2_region = region_box
 
-    return vertical_distance
+    x1_region = int(min(x1, x1_region))
+    y1_region = int(min(y1, y1_region))
+    x2_region = int(max(x2, x2_region))
+    y2_region = int(max(y2, y2_region))
 
+    region_box = [x1_region, y1_region, x2_region, y2_region]
 
-def _nearest_edge_distance(
-    input_bbox: List[int],
-    match_bbox: List[int],
-    weight: List[float] = [1.0, 1.0, 1.0, 1.0],
-    label: str = "text",
-    no_mask_labels: List[str] = [],
-    min_edge_distance_config: List[float] = [],
-    tolerance_len: float = 10.0,
-) -> Tuple[float, List[float]]:
-    """
-    Calculate the nearest edge distance between two bounding boxes, considering directional weights.
+    return region_box
 
-    Args:
-        input_bbox (list): The bounding box coordinates [x1, y1, x2, y2] of the input object.
-        match_bbox (list): The bounding box coordinates [x1', y1', x2', y2'] of the object to match against.
-        weight (list, optional): Directional weights for the edge distances [left, right, up, down]. Defaults to [1, 1, 1, 1].
-        label (str, optional): The label/type of the object in the bounding box (e.g., 'text'). Defaults to 'text'.
-        no_mask_labels (list, optional): Labels for which no masking is applied when calculating edge distances. Defaults to an empty list.
-        min_edge_distance_config (list, optional): Configuration for minimum edge distances [min_edge_distance_x, min_edge_distance_y].
-        Defaults to [float('inf'), float('inf')].
-        tolerance_len (float, optional): The tolerance length for adjusting edge distances. Defaults to 10.
 
-    Returns:
-        Tuple[float, List[float]]: A tuple containing:
-            - The calculated minimum edge distance between the bounding boxes.
-            - A list with the minimum edge distances in the x and y directions.
-    """
-    match_bbox_iou = _calculate_overlap_area_div_minbox_area_ratio(
-        input_bbox,
-        match_bbox,
-    )
-    if match_bbox_iou > 0 and label not in no_mask_labels:
-        return 0, [0, 0]
-
-    if not min_edge_distance_config:
-        min_edge_distance_config = [float("inf"), float("inf")]
-    min_edge_distance_x, min_edge_distance_y = min_edge_distance_config
-
-    x1, y1, x2, y2 = input_bbox
-    x1_prime, y1_prime, x2_prime, y2_prime = match_bbox
-
-    direction_num = 0
-    distance_x = float("inf")
-    distance_y = float("inf")
-    distance = [float("inf")] * 4
-
-    # input_bbox is to the left of match_bbox
-    if x2 < x1_prime:
-        direction_num += 1
-        distance[0] = x1_prime - x2
-        if abs(distance[0] - min_edge_distance_x) <= tolerance_len:
-            distance_x = min_edge_distance_x * weight[0]
-        else:
-            distance_x = distance[0] * weight[0]
-    # input_bbox is to the right of match_bbox
-    elif x1 > x2_prime:
-        direction_num += 1
-        distance[1] = x1 - x2_prime
-        if abs(distance[1] - min_edge_distance_x) <= tolerance_len:
-            distance_x = min_edge_distance_x * weight[1]
-        else:
-            distance_x = distance[1] * weight[1]
-    elif match_bbox_iou > 0:
-        distance[0] = 0
-        distance_x = 0
-
-    # input_bbox is above match_bbox
-    if y2 < y1_prime:
-        direction_num += 1
-        distance[2] = y1_prime - y2
-        if abs(distance[2] - min_edge_distance_y) <= tolerance_len:
-            distance_y = min_edge_distance_y * weight[2]
-        else:
-            distance_y = distance[2] * weight[2]
-        if label in no_mask_labels:
-            distance_y = max(0.1, distance_y) * 10  # for abstract
-    # input_bbox is below match_bbox
-    elif y1 > y2_prime:
-        direction_num += 1
-        distance[3] = y1 - y2_prime
-        if abs(distance[3] - min_edge_distance_y) <= tolerance_len:
-            distance_y = min_edge_distance_y * weight[3]
-        else:
-            distance_y = distance[3] * weight[3]
-    elif match_bbox_iou > 0:
-        distance[2] = 0
-        distance_y = 0
-
-    if direction_num == 2:
-        return (distance_x + distance_y), [
-            min(distance[0], distance[1]),
-            min(distance[2], distance[3]),
+def convert_formula_res_to_ocr_format(formula_res_list: List, ocr_res: dict):
+    for formula_res in formula_res_list:
+        x_min, y_min, x_max, y_max = list(map(int, formula_res["dt_polys"]))
+        poly_points = [
+            (x_min, y_min),
+            (x_max, y_min),
+            (x_max, y_max),
+            (x_min, y_max),
         ]
-    else:
-        return min(distance_x, distance_y), [
-            min(distance[0], distance[1]),
-            min(distance[2], distance[3]),
-        ]
-
-
-def _get_weights(label, horizontal):
-    """Define weights based on the label and orientation."""
-    if label == "doc_title":
-        return (
-            [1, 0.1, 0.1, 1] if horizontal else [0.2, 0.1, 1, 1]
-        )  # left-down ,  right-left
-    elif label in [
-        "paragraph_title",
-        "table_title",
-        "abstract",
-        "image",
-        "seal",
-        "chart",
-        "figure",
-    ]:
-        return [1, 1, 0.1, 1]  # down
-    else:
-        return [1, 1, 1, 0.1]  # up
-
-
-def _nearest_iou_edge_distance(
-    input_bbox: List[int],
-    match_bbox: List[int],
-    label: str,
-    vision_labels: List[str],
-    no_mask_labels: List[str],
-    median_width: int = -1,
-    title_labels: List[str] = [],
-    title_text: List[Tuple[int, List[int]]] = [],
-    sub_title: List[List[int]] = [],
-    min_distance_config: List[float] = [],
-    tolerance_len: float = 10.0,
-) -> Tuple[float, List[float]]:
-    """
-    Calculate the nearest IOU edge distance between two bounding boxes, considering label types, title adjustments, and minimum distance configurations.
-    This function computes the edge distance between two bounding boxes while considering their overlap (IOU) and various adjustments based on label types,
-    title text, and subtitle information. It also applies minimum distance configurations and tolerance adjustments.
-
-    Args:
-        input_bbox (List[int]): The bounding box coordinates [x1, y1, x2, y2] of the input object.
-        match_bbox (List[int]): The bounding box coordinates [x1', y1', x2', y2'] of the object to match against.
-        label (str): The label/type of the object in the bounding box (e.g., 'image', 'text', etc.).
-        vision_labels (List[str]): List of labels for vision-related objects (e.g., images, icons).
-        no_mask_labels (List[str]): Labels for which no masking is applied when calculating edge distances.
-        median_width (int, optional): The median width for title dispersion calculation. Defaults to -1.
-        title_labels (List[str], optional): Labels that indicate the object is a title. Defaults to an empty list.
-        title_text (List[Tuple[int, List[int]]], optional): Text content associated with title labels, in the format [(position_indicator, [x1, y1, x2, y2]), ...].
-        sub_title (List[List[int]], optional): List of subtitle bounding boxes to adjust the input_bbox. Defaults to an empty list.
-        min_distance_config (List[float], optional): Configuration for minimum distances [min_edge_distance_config, up_edge_distances_config, total_distance].
-        tolerance_len (float, optional): The tolerance length for adjusting edge distances. Defaults to 10.0.
-
-    Returns:
-        Tuple[float, List[float]]: A tuple containing:
-            - The calculated distance considering IOU and adjustments.
-            - The updated minimum distance configuration.
-    """
-
-    x1, y1, x2, y2 = input_bbox
-    x1_prime, y1_prime, x2_prime, y2_prime = match_bbox
-
-    min_edge_distance_config, up_edge_distances_config, total_distance = (
-        min_distance_config
-    )
-
-    iou_distance = 0
-
-    if label in vision_labels:
-        horizontal1 = horizontal2 = True
-    else:
-        horizontal1 = _get_bbox_direction(input_bbox)
-        horizontal2 = _get_bbox_direction(match_bbox, 3)
-
-    if (
-        horizontal1 != horizontal2
-        or _get_projection_iou(input_bbox, match_bbox, horizontal1) < 0.01
-    ):
-        iou_distance = 1
-
-    if label == "doc_title":
-        # Calculate distance for titles
-        disperse = max(1, median_width)
-        tolerance_len = max(tolerance_len, disperse)
-
-    # Adjust input_bbox based on sub_title
-    if sub_title:
-        for sub in sub_title:
-            x1_, y1_, x2_, y2_ = sub
-            x1, y1, x2, y2 = (
-                min(x1, x1_),
-                min(y1, y1_),
-                min(x2, x2_),
-                max(y2, y2_),
+        ocr_res["dt_polys"].append(poly_points)
+        formula_res_text: str = formula_res["rec_formula"]
+        ocr_res["rec_texts"].append(formula_res_text)
+        if ocr_res["rec_boxes"].size == 0:
+            ocr_res["rec_boxes"] = np.array(formula_res["dt_polys"])
+        else:
+            ocr_res["rec_boxes"] = np.vstack(
+                (ocr_res["rec_boxes"], [formula_res["dt_polys"]])
             )
-        input_bbox = [x1, y1, x2, y2]
-
-    if title_text:
-        for sub in title_text:
-            x1_, y1_, x2_, y2_ = sub[1]
-            if horizontal1:
-                x1, y1, x2, y2 = (
-                    min(x1, x1_),
-                    min(y1, y1_),
-                    min(x2, x2_),
-                    max(y2, y2_),
-                )
-            else:
-                x1, y1, x2, y2 = (
-                    min(x1, x1_),
-                    min(y1, y1_),
-                    max(x2, x2_),
-                    min(y2, y2_),
-                )
-        input_bbox = [x1, y1, x2, y2]
-
-    # Calculate edge distance
-    weight = _get_weights(label, horizontal1)
-    if label == "abstract":
-        tolerance_len *= 2
-
-    edge_distance, edge_distance_config = _nearest_edge_distance(
-        input_bbox,
-        match_bbox,
-        weight,
-        label=label,
-        no_mask_labels=no_mask_labels,
-        min_edge_distance_config=min_edge_distance_config,
-        tolerance_len=tolerance_len,
-    )
-
-    # Weights for combining distances
-    iou_edge_weight = [10**8, 10**4, 1, 0.0001]
-
-    # Calculate up and left edge distances
-    up_edge_distance = y1_prime
-    left_edge_distance = x1_prime
-    if (
-        label in no_mask_labels or label in title_labels or label in vision_labels
-    ) and y1 > y2_prime:
-        up_edge_distance = -y2_prime
-        left_edge_distance = -x2_prime
-
-    min_up_edge_distance = up_edge_distances_config
-    if abs(min_up_edge_distance - up_edge_distance) <= tolerance_len:
-        up_edge_distance = min_up_edge_distance
-
-    # Calculate total distance
-    distance = (
-        iou_distance * iou_edge_weight[0]
-        + edge_distance * iou_edge_weight[1]
-        + up_edge_distance * iou_edge_weight[2]
-        + left_edge_distance * iou_edge_weight[3]
-    )
-
-    # Update minimum distance configuration if a smaller distance is found
-    if total_distance > distance:
-        edge_distance_config = [
-            edge_distance_config[0],
-            edge_distance_config[1],
-        ]
-        min_distance_config = [
-            edge_distance_config,
-            up_edge_distance,
-            distance,
-        ]
-
-    return distance, min_distance_config
-
-
-def get_show_color(label: str) -> Tuple:
-    label_colors = {
-        # Medium Blue (from 'titles_list')
-        "paragraph_title": (102, 102, 255, 100),
-        "doc_title": (255, 248, 220, 100),  # Cornsilk
-        # Light Yellow (from 'tables_caption_list')
-        "table_title": (255, 255, 102, 100),
-        # Sky Blue (from 'imgs_caption_list')
-        "figure_title": (102, 178, 255, 100),
-        "chart_title": (221, 160, 221, 100),  # Plum
-        "vision_footnote": (144, 238, 144, 100),  # Light Green
-        # Deep Purple (from 'texts_list')
-        "text": (153, 0, 76, 100),
-        # Bright Green (from 'interequations_list')
-        "formula": (0, 255, 0, 100),
-        "abstract": (255, 239, 213, 100),  # Papaya Whip
-        # Medium Green (from 'lists_list' and 'indexs_list')
-        "content": (40, 169, 92, 100),
-        # Neutral Gray (from 'dropped_bbox_list')
-        "seal": (158, 158, 158, 100),
-        # Olive Yellow (from 'tables_body_list')
-        "table": (204, 204, 0, 100),
-        # Bright Green (from 'imgs_body_list')
-        "image": (153, 255, 51, 100),
-        # Bright Green (from 'imgs_body_list')
-        "figure": (153, 255, 51, 100),
-        "chart": (216, 191, 216, 100),  # Thistle
-        # Pale Yellow-Green (from 'tables_footnote_list')
-        "reference": (229, 255, 204, 100),
-        "algorithm": (255, 250, 240, 100),  # Floral White
-    }
+        ocr_res["rec_labels"].append("formula")
+        ocr_res["rec_polys"].append(poly_points)
+        ocr_res["rec_scores"].append(1)
+
+
+def caculate_bbox_area(bbox):
+    x1, y1, x2, y2 = map(float, bbox)
+    area = abs((x2 - x1) * (y2 - y1))
+    return area
+
+
+def get_show_color(label: str, order_label=False) -> Tuple:
+    if order_label:
+        label_colors = {
+            "doc_title": (255, 248, 220, 100),  # Cornsilk
+            "doc_title_text": (255, 239, 213, 100),
+            "paragraph_title": (102, 102, 255, 100),
+            "sub_paragraph_title": (102, 178, 255, 100),
+            "vision": (153, 255, 51, 100),
+            "vision_title": (144, 238, 144, 100),  # Light Green
+            "vision_footnote": (144, 238, 144, 100),  # Light Green
+            "normal_text": (153, 0, 76, 100),
+            "cross_layout": (53, 218, 207, 100),  # Thistle
+            "cross_reference": (221, 160, 221, 100),  # Floral White
+        }
+    else:
+        label_colors = {
+            # Medium Blue (from 'titles_list')
+            "paragraph_title": (102, 102, 255, 100),
+            "doc_title": (255, 248, 220, 100),  # Cornsilk
+            # Light Yellow (from 'tables_caption_list')
+            "table_title": (255, 255, 102, 100),
+            # Sky Blue (from 'imgs_caption_list')
+            "figure_title": (102, 178, 255, 100),
+            "chart_title": (221, 160, 221, 100),  # Plum
+            "vision_footnote": (144, 238, 144, 100),  # Light Green
+            # Deep Purple (from 'texts_list')
+            "text": (153, 0, 76, 100),
+            # Bright Green (from 'interequations_list')
+            "formula": (0, 255, 0, 100),
+            "abstract": (255, 239, 213, 100),  # Papaya Whip
+            # Medium Green (from 'lists_list' and 'indexs_list')
+            "content": (40, 169, 92, 100),
+            # Neutral Gray (from 'dropped_bbox_list')
+            "seal": (158, 158, 158, 100),
+            # Olive Yellow (from 'tables_body_list')
+            "table": (204, 204, 0, 100),
+            # Bright Green (from 'imgs_body_list')
+            "image": (153, 255, 51, 100),
+            # Bright Green (from 'imgs_body_list')
+            "figure": (153, 255, 51, 100),
+            "chart": (216, 191, 216, 100),  # Thistle
+            # Pale Yellow-Green (from 'tables_footnote_list')
+            "reference": (229, 255, 204, 100),
+            # "reference_content": (229, 255, 204, 100),
+            "algorithm": (255, 250, 240, 100),  # Floral White
+        }
     default_color = (158, 158, 158, 100)
     return label_colors.get(label, default_color)