Add computer_vision\otsu_threshold.py

computer_vision/otsu_threshold.py

@@ -0,0 +1,77 @@
+import numpy as np
+from PIL import Image
+
+"""
+Otsu thresholding algorithm for image processing
+https://en.wikipedia.org/wiki/Otsu%27s_method
+"""
+
+
+def otsu_threshold(image: Image) -> Image:
+    """
+    Applies Otsu's thresholding method to a grayscale image.
+
+    Parameters:
+    image (PIL.Image.Image): A grayscale PIL image object.
+
+    Returns:
+    PIL.Image.Image: A binary image after applying Otsu's thresholding.
+
+    Example:
+    >>> from PIL import Image
+    >>> import numpy as np
+    >>> image_array = np.array(
+    ...     [[0, 0, 0, 0], [255, 255, 255, 255], [0, 0, 0, 0], [255, 255, 255, 255]],
+    ...     dtype=np.uint8
+    ... )
+    >>> image = Image.fromarray(image_array)
+    >>> binary_image = otsu_threshold(image)
+    >>> np.array(binary_image)
+    array([[  0,   0,   0,   0],
+           [255, 255, 255, 255],
+           [  0,   0,   0,   0],
+           [255, 255, 255, 255]], dtype=uint8)
+    """
+    # Convert the image to numpy array
+    pixel_array = np.array(image)
+
+    # Compute histogram
+    hist, _ = np.histogram(pixel_array, bins=256, range=(0, 256))
+
+    # Compute between class variance
+    total_pixels = pixel_array.size
+    current_max, threshold = 0.0, 0  # Ensure current_max is a float
+    sum_total, sum_foreground = 0.0, 0.0  # Ensure these are floats
+    weight_background, weight_foreground = 0.0, 0.0  # Ensure these are floats
+
+    for i in range(256):
+        sum_total += i * hist[i]
+
+    for i in range(256):
+        weight_background += hist[i]
+        if weight_background == 0:
+            continue
+        weight_foreground = total_pixels - weight_background
+        if weight_foreground == 0:
+            break
+        sum_foreground += i * hist[i]
+
+        mean_background = sum_foreground / weight_background
+        mean_foreground = (sum_total - sum_foreground) / weight_foreground
+
+        between_class_variance = (
+            weight_background
+            * weight_foreground
+            * (mean_background - mean_foreground) ** 2
+        )
+
+        if between_class_variance > current_max:
+            current_max = between_class_variance
+            threshold = i
+
+    # Apply threshold to the image
+    binary_image = pixel_array > threshold
+    binary_image = binary_image.astype(np.uint8) * 255
+
+    # Convert numpy array back to PIL image
+    return Image.fromarray(binary_image)