A function that applies the convolution operation to a given image with a given kernel with or without padding.

convulation.py

# -*- coding: utf-8 -*-
import cv2
import numpy as np


def convulate(image, kernel, padding=False, paddingType=cv2.BORDER_CONSTANT):
    '''
    @param image: 2D array of type <numpy.ndarray> for given image.
    @param kernel: 2D array of type <numpy.ndarray> for given kernel.
    @param padding: Type <bool> to indicate if padding required.
    @param borderType: Type of padding from cv2 border types.

    returns: Image after convolution operation of kernel
             (Upto 2 Decimal Places)
    '''
    image = cv2.copyMakeBorder(image, 1, 1, 1, 1, paddingType)
    result = cv2.filter2D(image, -1, kernel, borderType=cv2.BORDER_CONSTANT)

    if padding:
        return np.round(result[1:-1, 1:-1], 2)
    else:
        return np.round(result[2:-2, 2:-2], 2)


# Example Code
if __name__ == '__main__':
    # Simple 2D matrix
    image = [[17, 14, 13, 9, 17],
             [21, 64, 62, 41, 19],
             [42, 54, 61, 52, 40],
             [41, 30, 31, 34, 38],
             [20, 24, 40, 58, 55]]
    image = np.array(image).astype('float64')

    # Simple averaging kernel for blurring
    kernel = [[1, 1, 1],
              [1, 1, 1],
              [1, 1, 1]]
    kernel = np.array(kernel)
    kernel = kernel/9

    # Convolution with no padding
    print('No Padding')
    print(convulate(image, kernel))

    # Convultion after symmetric padding
    print('\nWith Padding')
    print(convulate(image, kernel, True, cv2.BORDER_REFLECT))