convolute2d_brute_force.py

def convolution(self, matrix, kernel):
	# Matrix is a 3d array with shape of [width, height, channel]
	# Kernel is a 2d array with shape of [width, height]
	width = matrix.shape[0]
	height = matrix.shape[1]
	output_matrix = np.zeros((width, height, 3))
	
	# a simple helper method to sum up the values of the kernel
	kernel_sum = self._kernelSum(kernel)
	
	offset = len(kernel) // 2
	for row in range(width):
		for col in range(height):
			for ch in range(3):
				# the convolution value for a kernel window on the image
				accumulator = 0.0
				
				for j in range(len(kernel)):
					for k in range(len(kernel)):
						xcoord = row + j - offset
						ycoord = col + k - offset
						value = 0

						# if (x,y) coordinate is out of bounds, pick the kernel center instead 
						if xcoord < 0 or xcoord >= width or ycoord < 0 or ycoord >= height:
							value = matrix[row][col][ch]
						else:
							value = matrix[xcoord][ycoord][ch]
						accumulator += (value * kernel[j][k])
                    
				# An online guide told me that you should clamp convoluted values to a range of [0,255] and also
				# normalize them by dividing by the kernel_sum.
				average = self._clamp(accumulator) / kernel_sum
				output_matrix[row][col][ch] = int(average)
	return output_matrix