pyxllib 0.3.96__py3-none-any.whl → 0.3.197__py3-none-any.whl

pyxlpr/data/removeline.py

@@ -1,179 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-# @Author : 陈坤泽
-# @Email  : 877362867@qq.com
-# @Date   : 2020/09/15 21:00
-
-
-from pyxllib.xlcv import *
-
-
-class RemoveLine:
-    def __init__(self, filename):
-        self.bgr_img = xlcv.read(filename)
-        self.gray_img = cv2.cvtColor(self.bgr_img, cv2.COLOR_BGR2GRAY)
-        self.binary_img = cv2.adaptiveThreshold(self.gray_img, 255, 0, 1, 11, 3)
-
-    def debug(self):
-        """ 同run，只是debug模式，会输出中间结果
-        """
-        bgr_img = self.bgr_img
-        xlcv.show(bgr_img, '0 src')
-        dprint(bgr_img.shape)
-
-        lines = self.detect_lines()
-        xlcv.show(xlcv.lines(bgr_img, lines, [0, 0, 255]), '1 hough')
-        lines = self.refine_lines(lines)
-        xlcv.show(xlcv.lines(bgr_img, lines, [0, 0, 255]), '2 expand')
-        dprint(lines.shape)
-
-        dst = self.remove_lines(lines)
-        xlcv.show(dst, 'result')
-
-    def run(self):
-        """
-        :return: 获得去掉线后的图片
-        """
-        lines = self.detect_lines()
-        if lines.any():
-            lines = self.refine_lines2(lines)
-            return self.remove_lines(lines)
-        else:
-            return self.bgr_img
-
-    def detect_lines(self, rho=1, theta=np.pi / 180, threshold=80, min_line_length=50, max_line_gap=30):
-        lines = cv2.HoughLinesP(self.binary_img, rho, theta, threshold, min_line_length, max_line_gap)
-        # 不知道为什么返回值第2维会多一个1，把它删了~
-        lines = np.array([]) if lines is None else lines.reshape(-1, 4)
-        return lines
-
-    def refine_lines(self, lines):
-        im = self.binary_img
-        n, m = im.shape
-
-        def f(v):
-            """ 辅助函数：四舍五入取整 """
-            return int(round(v))
-
-        def expand(x, y, dx, dy):
-            """ 从(x,y)开始，按dx,dy逐步遍历，直到遇到im二值图为0的位置 """
-            while True:
-                j, i = f(x + dx), f(y + dy)
-                if 0 <= j < m and 0 <= i < n and im[i][j]:
-                    x, y = x + dx, y + dy
-                else:
-                    return f(x), f(y)
-
-        new_lines = []
-        for line in lines:
-            # 1 确保x1是左边，x2是右边
-            x1, y1, x2, y2 = line
-            if x2 < x1:
-                x1, y1, x2, y2 = x2, y2, x1, y1
-            # 2 计算斜率，分两类处理
-            if abs(x1 - x2) > abs(y1 - y2):
-                dx, dy = 1, (y1 - y2) / (x1 - x2)
-            else:
-                dx, dy = (x1 - x2) / (y1 - y2), 1
-            # 3 向左右各自延展
-            x1, y1 = expand(x1, y1, -dx, -dy)
-            x2, y2 = expand(x1, y1, dx, dy)
-            new_lines.append((x1, y1, x2, y2))
-        return np.array(new_lines)
-
-    def refine_lines2(self, lines):
-        """ 标准的延展会有些问题，会误删很多斜的笔划
-
-        这个版本的删除，会只删除比较水平的情况，宁愿少删，但不误删
-        """
-        im = self.binary_img
-        n, m = im.shape
-
-        def f(v):
-            """ 辅助函数：四舍五入取整 """
-            return int(round(v))
-
-        def expand(x, y, dx, dy):
-            """ 从(x,y)开始，按dx,dy逐步遍历，直到遇到im二值图为0的位置 """
-            while True:
-                j, i = f(x + dx), f(y + dy)
-                if 0 <= j < m and 0 <= i < n and im[i][j]:
-                    x, y = x + dx, y + dy
-                else:
-                    return f(x), f(y)
-
-        new_lines = []
-        for line in lines:
-            # 1 确保x1是左边，x2是右边
-            x1, y1, x2, y2 = line
-            if x2 < x1:
-                x1, y1, x2, y2 = x2, y2, x1, y1
-            # 2 计算斜率，分两类处理
-            dx, dy = 1, (y1 - y2) / (x1 - x2 + sys.float_info.epsilon)
-            if abs(dy) > math.tan(math.pi / 180 * 10):
-                # 倾斜度必须在10度以内（偏水平的线）
-                continue
-            # 3 向左右各自延展
-            x1, y1 = expand(x1, y1, -dx, -dy)
-            x2, y2 = expand(x1, y1, dx, dy)
-            new_lines.append((x1, y1, x2, y2))
-        return np.array(new_lines)
-
-    def remove_lines(self, lines, box_size=(7, 3), num=5):
-        """
-        :param lines:
-        :param box_size: 所有尺寸，默认是7*3的矩阵
-            7*3矩阵中，如果有超出4个前景点，则恢复该背景点为原始像素值
-        :param num: 矩阵至少需要的点数量
-        :return:
-        """
-        # 1 计算填充用的默认背景色
-        src = self.bgr_img
-        if not lines.any(): return src
-        bg_color = xlcv.bg_color(src, binary_img=self.binary_img)
-        # dprint(bg_color)
-
-        # 2 删笔画的mask矩阵
-        mask = xlcv.lines(np.zeros(src.shape[:2]), lines, [255], 2)
-        tmp = xlcv.lines(src, lines, bg_color, 2)  # 在原图画上背景颜色的线，产生抹除效果
-        dst = np.array(tmp)
-        tmp = cv2.cvtColor(tmp, cv2.COLOR_BGR2GRAY)
-
-        # 3 恢复断开的笔划
-        top, left = box_size[0] // 2, box_size[1] // 2
-        bottom, right = box_size[0] - top, box_size[1] - left
-        for i in range(src.shape[0]):
-            for j in range(src.shape[1]):
-                if not mask[i, j]: continue
-                arr = tmp[max(i - top, 0):i + bottom, max(j - left, 0):j + right].reshape(-1)
-                if sum(arr < 128) >= num:
-                    dst[i, j] = src[i, j]
-        # xlcv.show(tmp)
-        return dst
-
-
-def test_removeline(path):
-    """ 批量处理，检查去线功能效果 """
-
-    def func(p1, p2):
-        dst = RemoveLine(str(p1)).run()
-        cv2.imwrite(str(p2), dst)
-
-    d1, d2 = Dir(path), Dir(path + '+')
-    d2.ensure_dir()
-    d1_state = d1.select('*.png')
-    d1_state.subs = d1_state.subs
-    d1_state.procpaths(func, ref_dir=d2, pinterval=1000)
-
-
-if __name__ == '__main__':
-    TicToc.process_time(f'{dformat()}启动准备共用时')
-    tictoc = TicToc(__file__)
-    os.chdir(r'D:\RealEstate2020')
-
-    # RemoveLine(r'5_label+\agreement_label1_text2\handwriting\C100001245519-003.png').debug()
-    test_removeline(r'5_label+\agreement_label1_text2\handwriting')
-    test_removeline(r'5_label+\agreement_label1_text2\printed')  # 65680张图，704秒
-
-    # cv2.waitKey(0)
-    tictoc.toc()

pyxlpr/data/specialist.py

@@ -1,57 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-# @Author : 陈坤泽
-# @Email  : 877362867@qq.com
-# @Date   : 2021/06/06 11:16
-
-
-import numpy as np
-
-from pyxllib.file.pupil import struct_unpack
-from pyxllib.file.specialist import XlBytesIO
-
-
-def read_from_dgrl(dgrl):
-    """ 解析中科院的DGRL格式数据
-
-    Database Home: http://www.nlpr.ia.ac.cn/databases/handwriting/Home.html
-    CASIA 在线和离线中文手写数据库的一些数据读取功能
-
-    参考代码：https://blog.csdn.net/DaGongJiGuoMaLu09/article/details/107050519
-        有做了大量简化、工程封装
-
-    TODO 可以考虑做一个返回类似labelme格式的接口，会更通用
-        因为有时候会需要取整张原图
-        而且如果有整个原图，那么每个文本行用shape形状标记即可，不需要取出子图
-
-    :param dgrl: dgrl 格式的文件，或者对应的二进制数据流
-    :return: [(img0, label0), (img1, label1), ...]
-    """
-    # 输入参数可以是bytes，也可以是文件
-    f = XlBytesIO(dgrl)
-    # 表头尺寸
-    header_size = f.unpack('I')
-    # 表头剩下内容，提取 code_length
-    header = f.read(header_size - 4)
-    code_length = struct_unpack(header[-4:-2], 'H')  # 每个字符存储的字节数，一般都是用gbk编码，2个字节
-    # 读取图像尺寸信息，文本行数量
-    height, width, line_num = f.unpack('I' * 3)
-
-    # 读取每一行的信息
-    res = []
-    for k in range(line_num):
-        # 读取该行的字符数量
-        char_num = f.unpack('I')
-        label = f.readtext(char_num, code_length=code_length)
-        label = label.replace('\x00', '')  # 去掉不可见字符 \x00，这一步不加的话后面保存的内容会出现看不见的问题
-
-        # 读取该行的位置和尺寸
-        y, x, h, w = f.unpack('I' * 4)
-
-        # 读取该行的图片
-        bitmap = f.unpack('B' * (h * w))
-        bitmap = np.array(bitmap).reshape(h, w)
-
-        res.append((bitmap, label))
-
-    return res

pyxlpr/eval/__init__.py

@@ -1,85 +0,0 @@
-#!/usr/bin/env python3
-# -*- coding: utf-8 -*-
-# @Author : 陈坤泽
-# @Email  : 877362867@qq.com
-# @Date   : 2021/03/03 09:35
-
-""" 主要含有功能
-
-结果分数测评
-结果调试、可视化分析
-"""
-
-import pandas as pd
-
-from pyxlpr.data.coco import *
-
-
-class KieResParser:
-    """ Key Information Extraction Result Parser
-
-    关键信息提取结果的分析工具
-    """
-
-    def __init__(self, df, keys):
-        """
-        :param df: 必须要有一列file文件标识，用来进行分组判断，每一组的识别效果
-        :param keys: 信息种类，有哪些键
-            for k in keys，都有对应名称的 dt_{k},  gt_{k}  属性列，测试集场合，可以没有 gt 列
-        """
-        assert 'file' in df.columns
-        self.df = df
-        self.keys = keys
-        self.add_diff_columns()
-
-    def add_diff_columns(self):
-        """ 使用该功能，必须要有对应的gt列 """
-        for k in self.keys:
-            diff_col = f'diff_{k}'
-            if diff_col not in self.df.columns:
-                dt_col, gt_col = f'dt_{k}', f'gt_{k}'
-                self.df[diff_col] = [StrDiffType.difftype(row[dt_col], row[gt_col]) for idx, row in self.df.iterrows()]
-
-    def check_file(self):
-        """ 按每份文档为单位查看效果
-
-        TODO
-            1、支持查看单文件
-            2、支持筛选key，只对比查看部分key
-            3、支持显示key名称
-        """
-        pc = PairContent(Dir.TEMP / 'dt.txt', Dir.TEMP / 'gt.txt')
-        for i, (idx, item) in enumerate(self.df.iterrows(), start=1):
-            pc.add(f'{i}、{item.file}')
-            for k in self.keys:
-                pc.add(item[f'dt_{k}'], item[f'gt_{k}'])
-            pc.add('')
-        pc.bcompare(wait=False)
-
-    def stat_difftype(self):
-        """ 分析dt各种情况下检测效果、质量 """
-        items = []
-        index = []
-        for i in StrDiffType.typename.keys():
-            vals = []
-            for k in self.keys:
-                vals.append(sum(self.df[f'diff_{k}'] == i))
-            if sum(vals):
-                items.append(vals)
-                index.append(i)
-        df = pd.DataFrame.from_records(items, columns=self.keys, index=index)
-        return df
-
-    def check_key(self, key):
-        """ 检查某个特定key的识别效果 """
-        n = len(self.df)  # 总条目数
-        pc = PairContent(Dir.TEMP / f'dt-{key}.txt', Dir.TEMP / f'gt-{key}.txt')
-        dt_col, gt_col, diff_col = f'dt_{key}', f'gt_{key}', f'diff_{key}'
-        for difftype, items in self.df.groupby(diff_col):
-            m = len(items)
-            pc.add(f'{difftype}、{StrDiffType.typename[difftype]}（{m}/{n}≈{m / n:.2%}）')
-            for _, x in items.iterrows():
-                f = x['file'] + ', '
-                pc.add(f + x[dt_col], f + x[gt_col])
-            pc.add('')
-        pc.bcompare(wait=False)