napari-tmidas 0.2.2__py3-none-any.whl → 0.2.4__py3-none-any.whl

napari_tmidas/processing_functions/scipy_filters.py

@@ -16,10 +16,319 @@ from napari_tmidas._registry import BatchProcessingRegistry
 
 if SCIPY_AVAILABLE:
 
+    @BatchProcessingRegistry.register(
+        name="Resize Labels (Nearest, SciPy)",
+        suffix="_scaled",
+        description="Resize a label mask or label image by a scale factor using nearest-neighbor interpolation (scipy.ndimage.zoom, grid_mode=True) to preserve label integrity without shifting position.",
+        parameters={
+            "scale_factor": {
+                "type": "float",
+                "default": 1.0,
+                "min": 0.01,
+                "max": 10.0,
+                "description": "Factor by which to resize the label image (e.g., 0.8 for 80% size, 1.2 for 120% size). 1.0 means no resizing.",
+            },
+        },
+    )
+    def resize_labels(
+        label_image: np.ndarray, scale_factor: float = 1.0
+    ) -> np.ndarray:
+        """
+        Resize labeled objects while maintaining original array dimensions.
+
+        Objects are scaled isotropically and centered within the original
+        coordinate system, preserving spatial relationships with other data.
+
+        Parameters
+        ----------
+        label_image : np.ndarray
+            3D label image where each unique value represents a distinct object
+        scale_factor : float
+            Scaling factor (e.g., 0.8 = 80% size, 1.2 = 120% size)
+
+        Returns
+        -------
+        np.ndarray
+            Resized label image with same dimensions as input
+        """
+        import numpy as np
+        from scipy.ndimage import zoom
+
+        scale_factor = float(scale_factor)
+        if scale_factor == 1.0:
+            return label_image.copy()
+
+        original_shape = np.array(label_image.shape)
+
+        # Resize the labeled objects
+        scaled = zoom(
+            label_image,
+            zoom=scale_factor,
+            order=0,  # Preserve label values
+            grid_mode=True,  # Consistent coordinate system
+            mode="grid-constant",
+            cval=0,
+        ).astype(label_image.dtype)
+
+        new_shape = np.array(scaled.shape)
+        result = np.zeros(original_shape, dtype=label_image.dtype)
+
+        # Center the resized objects in the original array
+        offset = ((original_shape - new_shape) / 2).astype(int)
+
+        if scale_factor < 1.0:
+            # Place smaller objects in center
+            slices = tuple(slice(o, o + s) for o, s in zip(offset, new_shape))
+            result[slices] = scaled
+        else:
+            # Extract center region from larger objects
+            slices = tuple(
+                slice(-o if o < 0 else 0, s - o if o < 0 else s)
+                for o, s in zip(offset, original_shape)
+            )
+            result = scaled[slices]
+
+        return result
+
+    @BatchProcessingRegistry.register(
+        name="Subdivide Labels into 3 Layers",
+        suffix="_layers",
+        description="Subdivide each labeled object into 3 concentric layers and return a single label image where each layer receives a unique ID offset.",
+        parameters={
+            "is_half_body": {
+                "type": bool,
+                "default": False,
+                "description": "Enable this if the object is cut in half (e.g., half-spheroid). This will create layers as if it were a full body, so the cut surface shows inner/middle/outer layers.",
+            },
+            "cut_axis": {
+                "type": int,
+                "default": 0,
+                "min": 0,
+                "max": 2,
+                "description": "For half-bodies: which axis the object is cut along (0=Z, 1=Y, 2=X). Only the cut axis will be scaled 2x, not all dimensions.",
+            },
+        },
+    )
+    def subdivide_labels_3layers(
+        label_image: np.ndarray, is_half_body: bool = False, cut_axis: int = 0
+    ) -> np.ndarray:
+        """Subdivide labeled objects into three concentric layers.
+
+        Each object is partitioned into inner, middle, and outer shells of approximately
+        equal thickness. The layers are combined into a single label image using
+        non-overlapping ID ranges so they remain distinguishable.
+
+        Parameters
+        ----------
+        label_image : np.ndarray
+            Label image where each unique value represents a distinct object.
+        is_half_body : bool, optional
+            If True, treats the object as a half-body (e.g., half-spheroid cut at a plane).
+            Only the specified cut_axis will be scaled by 2x (not all dimensions) to avoid
+            excessive memory usage. The cut surface will then show all three layers
+            (inner, middle, outer) as if it were the interior of a complete object.
+            Default is False.
+        cut_axis : int, optional
+            For half-bodies: specifies which axis the object is cut along (0=Z, 1=Y, 2=X).
+            Only this axis will be scaled 2x to virtually complete the object. For a
+            hemisphere cut horizontally, use axis 0 (Z). Default is 0.
+
+        Returns
+        -------
+        numpy.ndarray
+            Single label image containing all three layers with unique label IDs.
+        """
+        # Define scale factors for the three boundaries
+        # To get equal thickness, we need to think about the "radius" reduction
+        # For 3 equal layers, we want boundaries at 1.0, ~0.67, ~0.33
+        scale_middle = 0.67  # ~67% size
+        scale_inner = 0.33  # ~33% size
+
+        original_shape = np.array(label_image.shape)
+
+        # Store information for mapping back to original coordinates
+        half_body_offset = None
+        cut_at_beginning = None
+
+        # If it's a half-body, we need to virtually "complete" the object first
+        # by mirroring it along the cut axis to create a full object
+        if is_half_body:
+            # Validate cut_axis
+            if cut_axis < 0 or cut_axis >= label_image.ndim:
+                raise ValueError(
+                    f"cut_axis must be between 0 and {label_image.ndim - 1}, got {cut_axis}"
+                )
+
+            # Find bounding box of the object along cut_axis
+            axes_to_sum = tuple(
+                i for i in range(label_image.ndim) if i != cut_axis
+            )
+            projection = np.sum(label_image > 0, axis=axes_to_sum)
+            nonzero_indices = np.where(projection > 0)[0]
+
+            if len(nonzero_indices) == 0:
+                # Empty image
+                return np.zeros_like(label_image)
+
+            # Get bounding box along cut axis
+            bbox_start = nonzero_indices[0]
+            bbox_end = nonzero_indices[-1] + 1
+
+            # Extract just the object portion along cut_axis
+            extract_slices = [slice(None)] * label_image.ndim
+            extract_slices[cut_axis] = slice(bbox_start, bbox_end)
+            object_portion = label_image[tuple(extract_slices)]
+
+            # Determine which end has the cut surface (max area)
+            # by checking areas at both ends of the object portion
+            object_projection = np.sum(object_portion > 0, axis=axes_to_sum)
+            first_slice_area = object_projection[0]
+            last_slice_area = object_projection[-1]
+
+            # Mirror this portion to create a complete object
+            flipped = np.flip(object_portion, axis=cut_axis)
+
+            if first_slice_area >= last_slice_area:
+                # Cut surface is at the beginning, so concatenate [flipped, original]
+                # This places the cut surface (the first slice) in the middle
+                work_image = np.concatenate(
+                    [flipped, object_portion], axis=cut_axis
+                )
+                cut_at_beginning = True
+            else:
+                # Cut surface is at the end, so concatenate [original, flipped]
+                # This places the cut surface (the last slice) in the middle
+                work_image = np.concatenate(
+                    [object_portion, flipped], axis=cut_axis
+                )
+                cut_at_beginning = False
+
+            work_shape = np.array(work_image.shape)
+
+            # Remember the offset for mapping back
+            half_body_offset = bbox_start
+        else:
+            work_image = label_image
+            work_shape = original_shape
+
+        # Helper function to create a scaled version centered in working space
+        def create_scaled_labels(scale_factor):
+            if scale_factor == 1.0:
+                return work_image.copy()
+
+            scaled = ndimage.zoom(
+                work_image,
+                zoom=scale_factor,
+                order=0,  # Preserve label values
+                grid_mode=True,  # Consistent coordinate system
+                mode="grid-constant",
+                cval=0,
+            ).astype(work_image.dtype)
+
+            new_shape = np.array(scaled.shape)
+            result = np.zeros(work_shape, dtype=work_image.dtype)
+
+            # Center the resized objects in the working array
+            offset = ((work_shape - new_shape) / 2).astype(int)
+            slices = tuple(slice(o, o + s) for o, s in zip(offset, new_shape))
+            result[slices] = scaled
+
+            return result
+
+        # Create the three scaled versions
+        full_labels = work_image.copy()  # Outer boundary (100%)
+        middle_labels = create_scaled_labels(
+            scale_middle
+        )  # Middle boundary (~67%)
+        inner_labels = create_scaled_labels(
+            scale_inner
+        )  # Inner boundary (~33%)
+
+        # Layer 3 (outermost shell): Full - Middle
+        layer3 = full_labels.copy()
+        layer3[middle_labels > 0] = 0
+
+        # Layer 2 (middle shell): Middle - Inner
+        layer2 = middle_labels.copy()
+        layer2[inner_labels > 0] = 0
+
+        # Layer 1 (innermost core): Inner
+        layer1 = inner_labels.copy()
+
+        max_label = int(label_image.max()) if label_image.size else 0
+        if max_label == 0:
+            return np.zeros_like(label_image)
+
+        if np.issubdtype(label_image.dtype, np.integer):
+            max_needed = max_label * 3
+            dtype_choices = [label_image.dtype, np.uint32, np.uint64]
+            for dtype in dtype_choices:
+                try:
+                    info = np.iinfo(dtype)
+                except ValueError:
+                    continue
+                if max_needed <= info.max:
+                    result_dtype = dtype
+                    break
+            else:
+                result_dtype = np.uint64
+        else:
+            result_dtype = np.uint32
+
+        result = np.zeros(work_shape, dtype=result_dtype)
+
+        layer1_mask = layer1 > 0
+        if np.any(layer1_mask):
+            result[layer1_mask] = layer1[layer1_mask].astype(
+                result_dtype, copy=False
+            )
+
+        layer2_mask = layer2 > 0
+        if np.any(layer2_mask):
+            result[layer2_mask] = (
+                layer2[layer2_mask].astype(result_dtype, copy=False)
+                + max_label
+            )
+
+        layer3_mask = layer3 > 0
+        if np.any(layer3_mask):
+            result[layer3_mask] = layer3[layer3_mask].astype(
+                result_dtype, copy=False
+            ) + (2 * max_label)
+
+        # If half-body mode, extract back the original half and place in original coordinates
+        if is_half_body:
+            # Extract the appropriate half depending on where the cut surface was
+            slices = [slice(None)] * result.ndim
+            mid_point = work_shape[cut_axis] // 2
+
+            if cut_at_beginning:
+                # Cut surface was at the beginning, we concatenated [flipped, original]
+                # So extract the second half to get back the original
+                slices[cut_axis] = slice(mid_point, work_shape[cut_axis])
+            else:
+                # Cut surface was at the end, we concatenated [original, flipped]
+                # So extract the first half to get back the original
+                slices[cut_axis] = slice(0, mid_point)
+
+            result_object = result[tuple(slices)]
+
+            # Place back into original volume at original position
+            final_result = np.zeros(original_shape, dtype=result.dtype)
+            place_slices = [slice(None)] * result.ndim
+            place_slices[cut_axis] = slice(
+                half_body_offset,
+                half_body_offset + result_object.shape[cut_axis],
+            )
+            final_result[tuple(place_slices)] = result_object
+            result = final_result
+
+        return result
+
     @BatchProcessingRegistry.register(
         name="Gaussian Blur",
         suffix="_blurred",
-        description="Apply Gaussian blur to the image",
+        description="Apply Gaussian blur to the image. Supports dimension_order hint (TYX, ZYX, etc.) to process frame-by-frame or apply 3D blur.",
         parameters={
             "sigma": {
                 "type": float,
@@ -30,16 +339,53 @@ if SCIPY_AVAILABLE:
             }
         },
     )
-    def gaussian_blur(image: np.ndarray, sigma: float = 1.0) -> np.ndarray:
+    def gaussian_blur(
+        image: np.ndarray, sigma: float = 1.0, dimension_order: str = "Auto"
+    ) -> np.ndarray:
         """
-        Apply Gaussian blur to the image
+        Apply Gaussian blur to the image.
+
+        Args:
+            image: Input image (YX, TYX, ZYX, CYX, TCYX, TZYX, etc.)
+            sigma: Standard deviation for Gaussian kernel
+            dimension_order: Dimension interpretation hint (Auto, YX, TYX, ZYX, CYX, TCYX, etc.)
+                            If TYX/CYX: processes each frame/channel independently (2D blur per slice)
+                            If ZYX: applies 3D blur to spatial volume
+                            If YX or Auto: processes as-is
+
+        Returns:
+            Blurred image with same shape as input
         """
-        return ndimage.gaussian_filter(image, sigma=sigma)
+        # Handle different dimension orders
+        if dimension_order in ["TYX", "CYX"] and len(image.shape) == 3:
+            # Process frame-by-frame or channel-by-channel (2D blur)
+            result = np.zeros_like(image)
+            for i in range(image.shape[0]):
+                result[i] = ndimage.gaussian_filter(image[i], sigma=sigma)
+            return result
+        elif (
+            dimension_order in ["TCYX", "TZYX", "ZCYX"]
+            and len(image.shape) == 4
+        ):
+            # Process each T/Z and C slice independently (2D blur)
+            result = np.zeros_like(image)
+            for i in range(image.shape[0]):
+                for j in range(image.shape[1]):
+                    result[i, j] = ndimage.gaussian_filter(
+                        image[i, j], sigma=sigma
+                    )
+            return result
+        elif dimension_order == "ZYX" and len(image.shape) == 3:
+            # Apply 3D blur to spatial volume
+            return ndimage.gaussian_filter(image, sigma=sigma)
+        else:
+            # YX, Auto, or other: process as-is
+            return ndimage.gaussian_filter(image, sigma=sigma)
 
     @BatchProcessingRegistry.register(
         name="Median Filter",
         suffix="_median",
-        description="Apply median filter for noise reduction",
+        description="Apply median filter for noise reduction. Supports dimension_order hint (TYX, ZYX, etc.) to process frame-by-frame or apply 3D filter.",
         parameters={
             "size": {
                 "type": int,
@@ -50,8 +396,57 @@ if SCIPY_AVAILABLE:
             }
         },
     )
-    def median_filter(image: np.ndarray, size: int = 3) -> np.ndarray:
+    def median_filter(
+        image: np.ndarray, size: int = 3, dimension_order: str = "Auto"
+    ) -> np.ndarray:
         """
-        Apply median filter for noise reduction
+        Apply median filter for noise reduction.
+
+        Args:
+            image: Input image (YX, TYX, ZYX, CYX, TCYX, TZYX, etc.)
+            size: Size of the median filter window
+            dimension_order: Dimension interpretation hint (Auto, YX, TYX, ZYX, CYX, TCYX, etc.)
+                            If TYX/CYX: processes each frame/channel independently (2D filter per slice)
+                            If ZYX: applies 3D filter to spatial volume
+                            If YX or Auto: processes as-is
+
+        Returns:
+            Filtered image with same shape as input
         """
-        return ndimage.median_filter(image, size=size)
+        # Handle different dimension orders
+        if dimension_order in ["TYX", "CYX"] and len(image.shape) == 3:
+            # Process frame-by-frame or channel-by-channel (2D filter)
+            result = np.zeros_like(image)
+            for i in range(image.shape[0]):
+                result[i] = ndimage.median_filter(image[i], size=size)
+            return result
+        elif (
+            dimension_order in ["TCYX", "TZYX", "ZCYX"]
+            and len(image.shape) == 4
+        ):
+            # Process each T/Z and C slice independently (2D filter)
+            result = np.zeros_like(image)
+            for i in range(image.shape[0]):
+                for j in range(image.shape[1]):
+                    result[i, j] = ndimage.median_filter(
+                        image[i, j], size=size
+                    )
+            return result
+        elif dimension_order == "ZYX" and len(image.shape) == 3:
+            # Apply 3D filter to spatial volume
+            return ndimage.median_filter(image, size=size)
+        else:
+            # YX, Auto, or other: process as-is
+            return ndimage.median_filter(image, size=size)
+
+else:
+    # Export stub functions that raise ImportError when called
+    def gaussian_blur(*args, **kwargs):
+        raise ImportError(
+            "SciPy is not available. Please install scipy to use this function."
+        )
+
+    def median_filter(*args, **kwargs):
+        raise ImportError(
+            "SciPy is not available. Please install scipy to use this function."
+        )