convolution_experiment.py

import cv2
import numpy as np
from matplotlib import pyplot as plt


def slice_matrix(matrix, row_start, col_start, size):
    col_end = col_start + size
    row_end = row_start + size
    return [row[col_start:col_end] for row in matrix[row_start:row_end]]

def convolve2d(kernel, matrix):
    # Flip the kernel 180 degrees.

    # Horizontal flip
    horizontal_flip = [row[::-1] for row in kernel]

    # Vertical flip of the already horizontally flipped kernel
    flipped_kernel = horizontal_flip[::-1]

    output_matrix = np.zeros_like(matrix)

    for m, row in enumerate(matrix):
        for n, _ in enumerate(row):
            sum = 0

            for k, kernel_row in enumerate(flipped_kernel):
                for l, _ in enumerate(kernel_row):
                    sum += flipped_kernel[k][l] * matrix[m - k][n - l]

            output_matrix[m][n] = sum

    return output_matrix


image = cv2.imread('assets/house_pixel_art.png')
gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
print("Shape of the image:", gray_image.shape)
image_height, image_width = gray_image.shape[:2]

kernel = np.array(
    [[-1, 0, 1],
     [-2, 0, 2],
     [-1, 0, 1]]
)

output_image = np.zeros_like(gray_image)

for i in range(1, image_height - 2):
    for j in range(1, image_width - 2):
        sliced_matrix = slice_matrix(
            matrix=gray_image, row_start=i, col_start=j, size=3
        )

        sliced_matrix_as_np_array = np.array(sliced_matrix)
        result_of_convolution = convolve2d(kernel, sliced_matrix_as_np_array)
        output_image[i + 1][j + 1] = result_of_convolution[0, 0]

plt.imshow(output_image, cmap='gray')
plt.title('OutDönüşümü'), plt.xticks([]), plt.yticks([])
plt.show()