Blur Animation source code and simplified output

episodes/06-blurring.md

@@ -247,8 +247,7 @@ in [the scikit-image documentation](https://scikit-image.org/docs/dev/user_guide
 
 ::::::::::::::::::::::::::::::::::::::::::::::::::
 
-This animation shows how the blur kernel moves along in the original image in
-order to calculate the colour channel values for the blurred image.
+Let's consider a very simple image to see blurring in action. The animation below shows how the blur kernel (large red square) moves along the image on the left in order to calculate the corresponding values for the blurred image (yellow central square) on the right. In this simple case, the original image is single-channel, but blurring would work likewise on a multi-channel image.
 
 ![](fig/blur-demo.gif){alt='Blur demo animation'}
 

episodes/fig/source/06-blurring/create_blur_animation.py

@@ -0,0 +1,162 @@
+### METADATA
+# author: Marco Dalla Vecchia @marcodallavecchia
+# description: Simple blurring animation of simple image
+# data-source: letterA.tif was created using ImageJ (https://imagej.net/ij/)
+###
+
+### INFO
+# This script creates the animated illustration of blurring in episode 6
+###
+
+### USAGE
+# The script requires the following Python packages:
+# - numpy
+# - scipy
+# - matplotlib
+# - tqdm
+# Install them with
+# $ conda install numpy scipy matplotlib tqdm
+#
+# The script can be executed with
+# $ python create_blur_animation.py
+# The output animation will be saved directly in the fig folder where the markdown lesson file will pick it up
+###
+
+### POTENTIAL IMPROVEMENTS
+# - Change colors for rectangular patches in animation
+# - Ask for image input instead of hard-coding it
+# - Ask for FPS as input
+# - Ask for animation format output
+
+# Import packages
+import numpy as np
+from scipy.ndimage import convolve
+from matplotlib import pyplot as plt
+from matplotlib import patches as p
+from matplotlib.animation import FuncAnimation
+from tqdm import tqdm
+
+# Path to input and output images
+data_path = "../../../data/"
+fig_path = "../../../fig/"
+input_file = data_path + "letterA.tif"
+output_file = fig_path + "blur-demo.gif"
+
+# Change here colors to improve accessibility
+kernel_color = "tab:red"
+center_color = "tab:olive"
+kernel_size = 3
+
+### ANIMATION FUNCTIONS
+def init():
+    """
+    Initialization function
+    - Set image array data
+    - Autoscale image display
+    - Set XY coordinates of rectangular patches
+    """
+    im.set_array(img_convolved)
+    im.autoscale()
+    k_rect.set_xy((-0.5, -0.5))
+    c_rect1.set_xy((kernel_size / 2 - 1, kernel_size / 2 - 1))
+    return [im, k_rect, c_rect1]
+
+def update(frame):
+    """
+    Animation update function. For every frame do the following:
+    - Update X and Y coordinates of rectangular patch for kernel
+    - Update X and Y coordinates of rectangular patch for central pixel
+    - Update blurred image frame
+    """
+    pbar.update(1)
+    row = (frame % total_frames) // (img_pad.shape[0] - kernel_size + 1)
+    col = (frame % total_frames) % (img_pad.shape[1] - kernel_size + 1)
+
+    k_rect.set_x(col - 0.5)
+    c_rect1.set_x(col + (kernel_size/2 - 1))
+    k_rect.set_y(row - 0.5)
+    c_rect1.set_y(row + (kernel_size/2 - 1))
+
+    im.set_array(all_frames[frame])
+    im.autoscale()
+
+    return [im, k_rect, c_rect1]
+
+# MAIN PROGRAM
+if __name__ == "__main__":
+
+    print("Creating blurred animation with kernel size:", kernel_size)
+
+    # Load image
+    img = plt.imread(input_file)
+
+    ### HERE WE USE THE CONVOLVE FUNCTION TO GET THE FINAL BLURRED IMAGE
+    # I chose a simple mean filter (equal kernel weights)
+    kernel = np.ones(shape=(kernel_size, kernel_size)) / kernel_size ** 2 # create kernel
+    # convolve the image i.e. apply mean filter
+    img_convolved = convolve(img, kernel, mode='constant', cval=0) # pad borders with zero like below for consistency
+
+
+    ### HERE WE CONVOLVE MANUALLY STEP-BY-STEP TO CREATE ANIMATION
+    img_pad = np.pad(img, (int(np.ceil(kernel_size/2) - 1), int(np.ceil(kernel_size/2) - 1))) # Pad image to deal with borders
+    new_img = np.zeros(img.shape, dtype=np.uint16) # this will be the blurred final image
+
+    # add first frame with complete blurred image for print version of GIF
+    all_frames = [img_convolved]
+
+    # precompute animation frames and append to the list
+    total_frames = (img_pad.shape[0] - kernel_size + 1) * (img_pad.shape[1] - kernel_size + 1) # total frames if by change image is not squared
+    for frame in range(total_frames):
+        row = (frame % total_frames) // (img_pad.shape[0] - kernel_size + 1) # row index
+        col = (frame % total_frames) % (img_pad.shape[1] - kernel_size + 1) # col index
+        img_chunk = img_pad[row : row + kernel_size, col : col + kernel_size] # get current image chunk inside the kernel
+        new_img[row, col] = np.mean(img_chunk).astype(np.uint16) # calculate its mean -> mean filter
+        all_frames.append(new_img.copy()) # append to animation frames list
+
+    # We now have an extra frame
+    total_frames += 1 
+
+    ### FROM HERE WE START CREATING THE ANIMATION
+    # Initialize canvas
+    f, (ax1, ax2) = plt.subplots(1,2, figsize=(10,5))
+
+    # Display the padded image -> this one won't change during the animation
+    ax1.imshow(img_pad, cmap='gray')
+    # Initialize the blurred image -> this is the first frame with already the final result
+    im = ax2.imshow(img_convolved, animated=True, cmap='gray')
+
+    # Define rectangular patches to identify moving kernel
+    k_rect = p.Rectangle((-0.5,-0.5), kernel_size, kernel_size, linewidth=2, edgecolor=kernel_color, facecolor='none', alpha=0.8) # kernel rectangle
+    c_rect1 = p.Rectangle(((kernel_size/2 - 1), (kernel_size/2 - 1)), 1, 1, linewidth=2, edgecolor=center_color, facecolor='none') # central pixel rectangle
+    # Add them to the figure
+    ax1.add_patch(k_rect)
+    ax1.add_patch(c_rect1)
+
+    # Fix limits to the right image (without padding) is the same size as the left image (with padding)
+    ax2.set(
+        ylim=((img_pad.shape[0] - kernel_size / 2), -kernel_size / 2), 
+        xlim=(-kernel_size / 2, (img_pad.shape[1] - kernel_size / 2))
+    )
+
+    # We don't need to see the ticks
+    ax1.axis("off")
+    ax2.axis("off")
+
+    # Create progress bar to visualize animation progress
+    pbar = tqdm(total=total_frames)
+
+    ### HERE WE CREATE THE ANIMATION
+    # Use FuncAnimation to create the animation
+    ani = FuncAnimation(
+        f, update, 
+        frames=range(total_frames), 
+        interval=50, # we could change the animation speed
+        init_func=init,
+        blit=True
+    )
+
+    # Export animation
+    plt.tight_layout()
+    ani.save(output_file)
+    pbar.close()
+    print("Animation exported")