Add a more powerful postprocessor script

postprocess_example_1.py

@@ -0,0 +1,37 @@
+#!/usr/bin/env python3
+import sys
+
+from PIL import Image, ImageOps
+import numpy as np
+
+
+if __name__ == '__main__':
+    assert len(sys.argv) == 2, "Image path must be passed!"
+    path = sys.argv[1]
+
+    # load image, discard alpha (if present)
+    img = Image.open(path).convert("RGB")
+
+    # remove menu and indicators
+    data = np.array(img)
+    menu_is_open = (data[52:58, 52:58] == 0).all()
+    if menu_is_open:
+        # remove the entire menu, and the x in the top right corner
+        data[:, :120, :] = 255
+        data[40:81, 1324:1364, :] = 255
+    else:
+        # remove only the menu indicator circle
+        data[40:81, 40:81, :] = 255
+
+    # crop to the bounding box
+    img = Image.fromarray(data).convert("RGB")
+    bbox = ImageOps.invert(img).getbbox()
+    img = img.crop(bbox)
+
+    # set alpha channel
+    data = np.array(img.convert("RGBA"))
+    # copy inverted red channel to alpha channel, so that the background is transparent
+    # (could have also used blue or green here, doesn't matter)
+    data[..., -1] = 255 - data[..., 0]
+    img = Image.fromarray(data)
+    img.save(path)

postprocess_example_2.py

@@ -0,0 +1,127 @@
+import sys
+
+import numpy as np
+from PIL import Image, ImageEnhance, ImageFilter, ImageOps
+
+
+def apply_curve(img):
+    # Convert image to numpy array
+    img_array = np.array(img)
+
+    # Create lookup table. The x axis is all possible 16 bit gray colors, and the
+    # y axis is what they map to.
+    x = np.linspace(0, 65535, 65536)
+    # for example, 2949 -> 0
+    y = np.interp(x, [0, 5700, 7680, 65535], [0, 59000, 65535, 65535])
+    lut = np.clip(y, 0, 65535).astype(np.uint16)
+
+    # Apply lookup table
+    img_curved = np.interp(img_array, x, lut).astype(np.uint16)
+
+    # Convert back to PIL Image
+    return Image.fromarray(img_curved, mode="I;16")
+
+
+def add_border_to_image(img):
+    """Add a black border to the input image"""
+    border_size = 4
+    border_color = (0, 0, 0)
+    return ImageOps.expand(img, border=border_size, fill=border_color)
+
+
+def expand_bounding_box(bbox, padding, max_size):
+    """Expand the bounding box of the image by padding pixels"""
+    # Expand the bounding box by padding pixels
+    # left, upper, right, lower
+    bbox = (
+        max(bbox[0] - padding, 0),
+        max(bbox[1] - padding, 0),
+        min(bbox[2] + padding, max_size[0]),
+        min(bbox[3] + padding, max_size[1]),
+    )
+
+    return bbox
+
+
+def remove_dotted_background(img, blur_radius=10, block_size=20, padding=100):
+    """Get boundry box discounting dots and lines and marks"""
+    # Apply a blur to the image
+    filtered_image = img.filter(ImageFilter.GaussianBlur(blur_radius))
+
+    # Resize the image to be small to get pixel blocks, then resize back
+    original_img_size = filtered_image.size
+    filtered_image = filtered_image.resize(
+        (filtered_image.size[0] // block_size, filtered_image.size[1] // block_size),
+        Image.Resampling.NEAREST,
+    )
+    filtered_image = filtered_image.resize(original_img_size, Image.Resampling.NEAREST)
+
+    # Brighten image slightly and reduce contrast
+    # Since some pixels might be like, (255, 255, 254)
+    # enhancer = ImageEnhance.Contrast(img)
+    # img = enhancer.enhance(0.9)
+    enhancer = ImageEnhance.Brightness(filtered_image)
+    filtered_image = enhancer.enhance(1.05)
+
+    # Crop the image
+    bbox = ImageOps.invert(filtered_image).getbbox()
+    bbox = expand_bounding_box(bbox, padding, original_img_size)
+
+    return img.crop(bbox)
+
+
+def cleanup_image(path):
+    """Clean up a remarkable screenshot. Remove artifacts like menu and compass"""
+    # Load image, discard alpha (if present)
+    img = Image.open(path)
+    img = apply_curve(img)
+    original_mode = img.mode
+    print(f"Original mode: {original_mode}")
+    img_array = np.array(img)
+    img_normalized = (img_array / 256).astype("uint8")
+    img = Image.fromarray(np.stack((img_normalized,) * 3, axis=-1))
+    img = img.convert("RGB")
+
+    # Remove menu and indicators
+    data = np.array(img)
+    print(data.shape)
+    if np.all(data[1840, 37] == [0, 0, 0]):
+        print(f"The menu is open")
+        # Remove the entire menu, and the x in the top right corner
+        data[:, :120, :] = 255
+        data[40:81, 1324:1364, :] = 255
+    else:
+        print("The menu is closed")
+        # Remove only the menu indicator circle
+        data[40:81, 40:81, :] = 255
+
+    # Remove the compass from the top left
+    data[25:79, 30:79] = [255, 255, 255]
+
+    # Remove the close button from the top right
+    data[33:73, 1332:1372] = [255, 255, 255]
+
+    # Remove page range
+    print("Removing page range")
+    data[1820:1851, 590:806] = [255, 255, 255]
+
+    # Crop to the bounding box
+    img = Image.fromarray(data).convert("RGB")
+    img = remove_dotted_background(img)
+
+    # Set alpha channel
+    data = np.array(img.convert("RGBA"))
+
+    # Copy inverted red channel to alpha channel, so that the background is transparent
+    # (could have also used blue or green here, doesn't matter)
+    # data[..., -1] = 255 - data[..., 0]
+    return Image.fromarray(data)
+
+
+if __name__ == "__main__":
+    if not len(sys.argv) == 2:
+        raise Exception("Image path must be passed!")
+    path = sys.argv[1]
+    img = cleanup_image(path)
+    img.save(path)
+