|
| 1 | +### METADATA |
| 2 | +# author: Marco Dalla Vecchia @marcodallavecchia |
| 3 | +# description: Simple blurring animation of simple image |
| 4 | +# data-source: letterA.tif was created using ImageJ (https://imagej.net/ij/) |
| 5 | +### |
| 6 | + |
| 7 | +### INFO |
| 8 | +# This script creates the animated illustration of blurring in episode 6 |
| 9 | +### |
| 10 | + |
| 11 | +### USAGE |
| 12 | +# The script was written in Python 3.12 and required the following Python packages: |
| 13 | +# - numpy==2.2.3 |
| 14 | +# - scipy==1.15.2 |
| 15 | +# - matplotlib==3.10.1 |
| 16 | +# - tqdm==4.67.1 |
| 17 | +# |
| 18 | +# The script can be executed with |
| 19 | +# $ python create_blur_animation.py |
| 20 | +# The output animation will be saved directly in the fig folder where the markdown lesson file will pick it up |
| 21 | +### |
| 22 | + |
| 23 | +### POTENTIAL IMPROVEMENTS |
| 24 | +# - Change colors for rectangular patches in animation |
| 25 | +# - Ask for image input instead of hard-coding it |
| 26 | +# - Ask for FPS as input |
| 27 | +# - Ask for animation format output |
| 28 | + |
| 29 | +# Import packages |
| 30 | +import numpy as np |
| 31 | +from scipy.ndimage import convolve |
| 32 | +from matplotlib import pyplot as plt |
| 33 | +from matplotlib import patches as p |
| 34 | +from matplotlib.animation import FuncAnimation |
| 35 | +from tqdm import tqdm |
| 36 | + |
| 37 | +# Path to input and output images |
| 38 | +data_path = "../../../data/" |
| 39 | +fig_path = "../../../fig/" |
| 40 | +input_file = data_path + "letterA.tif" |
| 41 | +output_file = fig_path + "blur-demo.gif" |
| 42 | + |
| 43 | +# Change here colors to improve accessibility |
| 44 | +kernel_color = "tab:red" |
| 45 | +center_color = "tab:olive" |
| 46 | +kernel_size = 3 |
| 47 | + |
| 48 | +### ANIMATION FUNCTIONS |
| 49 | +def init(): |
| 50 | + """ |
| 51 | + Initialization function |
| 52 | + - Set image array data |
| 53 | + - Autoscale image display |
| 54 | + - Set XY coordinates of rectangular patches |
| 55 | + """ |
| 56 | + im.set_array(img_convolved) |
| 57 | + im.autoscale() |
| 58 | + k_rect.set_xy((-0.5, -0.5)) |
| 59 | + c_rect1.set_xy((kernel_size / 2 - 1, kernel_size / 2 - 1)) |
| 60 | + return [im, k_rect, c_rect1] |
| 61 | + |
| 62 | +def update(frame): |
| 63 | + """ |
| 64 | + Animation update function. For every frame do the following: |
| 65 | + - Update X and Y coordinates of rectangular patch for kernel |
| 66 | + - Update X and Y coordinates of rectangular patch for central pixel |
| 67 | + - Update blurred image frame |
| 68 | + """ |
| 69 | + pbar.update(1) |
| 70 | + row = (frame % total_frames) // (img_pad.shape[0] - kernel_size + 1) |
| 71 | + col = (frame % total_frames) % (img_pad.shape[1] - kernel_size + 1) |
| 72 | + |
| 73 | + k_rect.set_x(col - 0.5) |
| 74 | + c_rect1.set_x(col + (kernel_size/2 - 1)) |
| 75 | + k_rect.set_y(row - 0.5) |
| 76 | + c_rect1.set_y(row + (kernel_size/2 - 1)) |
| 77 | + |
| 78 | + im.set_array(all_frames[frame]) |
| 79 | + im.autoscale() |
| 80 | + |
| 81 | + return [im, k_rect, c_rect1] |
| 82 | + |
| 83 | +# MAIN PROGRAM |
| 84 | +if __name__ == "__main__": |
| 85 | + |
| 86 | + print(f"Creating blurring animation with kernel size: {kernel_size}") |
| 87 | + |
| 88 | + # Load image |
| 89 | + img = plt.imread(input_file) |
| 90 | + |
| 91 | + ### HERE WE USE THE CONVOLVE FUNCTION TO GET THE FINAL BLURRED IMAGE |
| 92 | + # I chose a simple mean filter (equal kernel weights) |
| 93 | + kernel = np.ones(shape=(kernel_size, kernel_size)) / kernel_size ** 2 # create kernel |
| 94 | + # convolve the image, i.e., apply mean filter |
| 95 | + img_convolved = convolve(img, kernel, mode='constant', cval=0) # pad borders with zero like below for consistency |
| 96 | + |
| 97 | + |
| 98 | + ### HERE WE CONVOLVE MANUALLY STEP-BY-STEP TO CREATE ANIMATION |
| 99 | + 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 |
| 100 | + new_img = np.zeros(img.shape, dtype=np.uint16) # this will be the blurred final image |
| 101 | + |
| 102 | + # add first frame with complete blurred image for print version of GIF |
| 103 | + all_frames = [img_convolved] |
| 104 | + |
| 105 | + # precompute animation frames and append to the list |
| 106 | + total_frames = (img_pad.shape[0] - kernel_size + 1) * (img_pad.shape[1] - kernel_size + 1) # total frames if by chance image is not squared |
| 107 | + for frame in range(total_frames): |
| 108 | + row = (frame % total_frames) // (img_pad.shape[0] - kernel_size + 1) # row index |
| 109 | + col = (frame % total_frames) % (img_pad.shape[1] - kernel_size + 1) # col index |
| 110 | + img_chunk = img_pad[row:row + kernel_size, col:col + kernel_size] # get current image chunk inside the kernel |
| 111 | + new_img[row, col] = np.mean(img_chunk).astype(np.uint16) # calculate its mean -> mean filter |
| 112 | + all_frames.append(new_img.copy()) # append to animation frames list |
| 113 | + |
| 114 | + # We now have an extra frame |
| 115 | + total_frames += 1 |
| 116 | + |
| 117 | + ### FROM HERE WE START CREATING THE ANIMATION |
| 118 | + # Initialize canvas |
| 119 | + f, (ax1, ax2) = plt.subplots(ncols=2, figsize=(10, 5)) |
| 120 | + |
| 121 | + # Display the padded image -> this one won't change during the animation |
| 122 | + ax1.imshow(img_pad, cmap="gray") |
| 123 | + # Initialize the blurred image -> this is the first frame with already the final result |
| 124 | + im = ax2.imshow(img_convolved, animated=True, cmap="gray") |
| 125 | + |
| 126 | + # Define rectangular patches to identify moving kernel |
| 127 | + k_rect = p.Rectangle((-0.5, -0.5), kernel_size, kernel_size, linewidth=2, edgecolor=kernel_color, facecolor="none", alpha=0.8) # kernel rectangle |
| 128 | + c_rect1 = p.Rectangle(((kernel_size/2 - 1), (kernel_size/2 - 1)), 1, 1, linewidth=2, edgecolor=center_color, facecolor="none") # central pixel rectangle |
| 129 | + # Add them to the figure |
| 130 | + ax1.add_patch(k_rect) |
| 131 | + ax1.add_patch(c_rect1) |
| 132 | + |
| 133 | + # Fix limits of the image on the right (without padding) so that it is the same size as the image on the left (with padding) |
| 134 | + ax2.set( |
| 135 | + ylim=((img_pad.shape[0] - kernel_size / 2), -kernel_size / 2), |
| 136 | + xlim=(-kernel_size / 2, (img_pad.shape[1] - kernel_size / 2)) |
| 137 | + ) |
| 138 | + |
| 139 | + # We don't need to see the ticks |
| 140 | + ax1.axis("off") |
| 141 | + ax2.axis("off") |
| 142 | + |
| 143 | + # Create progress bar to visualize animation progress |
| 144 | + pbar = tqdm(total=total_frames) |
| 145 | + |
| 146 | + ### HERE WE CREATE THE ANIMATION |
| 147 | + # Use FuncAnimation to create the animation |
| 148 | + ani = FuncAnimation( |
| 149 | + f, update, |
| 150 | + frames=range(total_frames), |
| 151 | + interval=50, # we could change the animation speed |
| 152 | + init_func=init, |
| 153 | + blit=True |
| 154 | + ) |
| 155 | + |
| 156 | + # Export animation |
| 157 | + plt.tight_layout() |
| 158 | + ani.save(output_file) |
| 159 | + pbar.close() |
| 160 | + print("Animation exported") |
0 commit comments