|
| 1 | +""" |
| 2 | +Implementation of gaussian filter algorithm |
| 3 | +""" |
| 4 | +from cv2 import imread, cvtColor, COLOR_BGR2GRAY, imshow, waitKey |
| 5 | +from numpy import pi, mgrid, exp, square, zeros, ravel, dot, uint8 |
| 6 | + |
| 7 | + |
| 8 | +def gen_gaussian_kernel(k_size, sigma): |
| 9 | + center = k_size // 2 |
| 10 | + x, y = mgrid[0-center:k_size-center, 0-center:k_size-center] |
| 11 | + g = 1/(2*pi*sigma) * exp(-(square(x) + square(y))/(2*square(sigma))) |
| 12 | + return g |
| 13 | + |
| 14 | + |
| 15 | +def gaussian_filter(image, k_size, sigma): |
| 16 | + height, width = image.shape[0], image.shape[1] |
| 17 | + # dst image height and width |
| 18 | + dst_height = height-k_size+1 |
| 19 | + dst_width = width-k_size+1 |
| 20 | + |
| 21 | + # im2col, turn the k_size*k_size pixels into a row and np.vstack all rows |
| 22 | + image_array = zeros((dst_height*dst_width, k_size*k_size)) |
| 23 | + row = 0 |
| 24 | + for i in range(0, dst_height): |
| 25 | + for j in range(0, dst_width): |
| 26 | + window = ravel(image[i:i + k_size, j:j + k_size]) |
| 27 | + image_array[row, :] = window |
| 28 | + row += 1 |
| 29 | + |
| 30 | + # turn the kernel into shape(k*k, 1) |
| 31 | + gaussian_kernel = gen_gaussian_kernel(k_size, sigma) |
| 32 | + filter_array = ravel(gaussian_kernel) |
| 33 | + |
| 34 | + # reshape and get the dst image |
| 35 | + dst = dot(image_array, filter_array).reshape(dst_height, dst_width).astype(uint8) |
| 36 | + |
| 37 | + return dst |
| 38 | + |
| 39 | + |
| 40 | +if __name__ == '__main__': |
| 41 | + # read original image |
| 42 | + img = imread(r'../image_data/lena.jpg') |
| 43 | + # turn image in gray scale value |
| 44 | + gray = cvtColor(img, COLOR_BGR2GRAY) |
| 45 | + |
| 46 | + # get values with two different mask size |
| 47 | + gaussian3x3 = gaussian_filter(gray, 3, sigma=1) |
| 48 | + gaussian5x5 = gaussian_filter(gray, 5, sigma=0.8) |
| 49 | + |
| 50 | + # show result images |
| 51 | + imshow('gaussian filter with 3x3 mask', gaussian3x3) |
| 52 | + imshow('gaussian filter with 5x5 mask', gaussian5x5) |
| 53 | + waitKey() |
0 commit comments