Histalign/Transform: Add RGB to grayscale transformation

Author: olivierdelree

src/histalign/io/transform/__init__.py

@@ -216,6 +216,54 @@ def _transform_series(
     transform_index = 0
 
     transform_function = get_appropriate_transform_function(transform)
+
+    # TODO: Clean this up, had to add this extra transform in a hurry!
+    if transform == "rgb2gray":
+        order = source_file.dimension_order
+        if "Z" in order.value:
+            raise ValueError("Unsupported operation: RGB2gray on Z stack.")
+        elif "C" not in order.value:
+            raise ValueError("Image does not have a channel dimension.")
+
+        channel_index = order.value.index("C")
+        transformed_image = transform_function(
+            source_file.load(), channel_index, **kwargs
+        )
+
+        if isinstance(destination_file[0], Path) or seek_first:
+            order = source_file.dimension_order
+            transformed_shape = transformed_image.shape
+
+            updated_metadata = update_metadata(
+                source_file.metadata, transformed_shape
+            )
+
+            if seek_first:
+                # Writing the first image of a new series
+                destination_file[0].seek_next_series(
+                    shape=transformed_shape,
+                    dtype=source_file.dtype,
+                    metadata=updated_metadata,
+                )
+                seek_first = False
+            else:
+                destination_plugin_class = get_appropriate_plugin_class(
+                    destination_file[0], mode="w"
+                )
+                destination_file[0] = destination_plugin_class(
+                    destination_file[0],
+                    mode="w",
+                    dimension_order=order,
+                    shape=transformed_shape,
+                    dtype=source_file.dtype,
+                    metadata=updated_metadata,
+                )
+
+        destination_file[0].write_image(transformed_image, (slice(None),) * 2)
+
+        return
+
+
     for image in source_file.iterate_images(source_file.dimension_order):
         _module_logger.info(f"Transform {transform_index}/{transform_count}.")
 
@@ -264,6 +312,9 @@ def _transform_series(
 def update_metadata(metadata: OmeXml, transformed_shape: list[int]) -> OmeXml:
     updated_metadata = metadata.model_copy(deep=True)
 
+    # TODO: Fix this to support Z (and T?)
+    should_drop_c = metadata.SizeC > 1 and len(transformed_shape) < 3
+
     x_position = updated_metadata.DimensionOrder.index("X")
     y_position = updated_metadata.DimensionOrder.index("Y")
 
@@ -279,4 +330,11 @@ def update_metadata(metadata: OmeXml, transformed_shape: list[int]) -> OmeXml:
     updated_metadata.PhysicalSizeX = new_x_scaling
     updated_metadata.PhysicalSizeY = new_y_scaling
 
+    if should_drop_c:
+        updated_metadata.SizeC = 1
+        updated_metadata.Channel = []
+
+        order = updated_metadata.DimensionOrder
+        updated_metadata.DimensionOrder = order.replace("C", "")
+
     return updated_metadata

src/histalign/io/transform/transforms.py

@@ -3,18 +3,20 @@
 # SPDX-License-Identifier: MIT
 
 from collections.abc import Callable
-from typing import Literal
+from typing import Any, Literal
 
 import numpy as np
 from skimage.transform import rescale
 
-Transform = Literal["downscale"]
+Transform = Literal["downscale", "rgb2gray"]
 
 
 def get_appropriate_transform_function(transform: Transform) -> Callable:
     match transform:
         case "downscale":
             return downscaling_transform
+        case "rgb2gray":
+            return rgb_to_gray_transform
         case _:
             raise ValueError(f"Unknown transform '{transform}'.")
 
@@ -37,3 +39,40 @@ def downscaling_transform(
         anti_aliasing=True,
     )
     return array
+
+
+def rgb_to_gray_transform(
+    series: np.ndarray, channel_index: int, **kwargs: Any
+) -> np.ndarray:
+    """Convert an RGB image to grayscale using Luma conversion.
+
+    Note this is not perfect and does not reflect the transformation done by, e.g.,
+    ImageJ (they seem to do some fancier scaling).
+
+    Args:
+        image: Series containing images with RGB channels.
+        channel_index: Index of the channels.
+
+    Returns:
+        The image as grayscale. This will have a shape similar to the input image
+        without the channel.
+
+    References:
+        Factors: https://en.wikipedia.org/wiki/Grayscale#Luma_coding_in_video_systems
+    """
+    minimum = series.min()
+    maximum = series.max()
+    range_ = maximum - minimum
+
+    transformed = (
+        series.take(0, channel_index) * 0.299
+        + series.take(1, channel_index) * 0.587
+        + series.take(2, channel_index) * 0.114
+    )
+
+    transformed -= transformed.min()
+    transformed /= transformed.max()
+    transformed *= range_
+    transformed += minimum
+
+    return transformed.astype(series.dtype)