Applying Sobel operator in numpy to simple binary image

Question

I'm trying to apply the Sobel operator to a simple binary image, but the resulting gradient is flipped (when comparing to the output of scipy's signal.convolve2d function.

from scipy import signal
import numpy as np
import matplotlib.pyplot as plt


def sobelx_homemade(arr, kx):
    offset = 1
    sx = np.zeros(arr.shape)
    for y in range(offset, arr.shape[0] - offset):
        for x in range(offset, arr.shape[1] - offset):
            rstart, rend = y-offset, y+offset+1
            cstart, cend = x-offset, x+offset+1
            w = arr[rstart:rend, cstart:cend]
            Ix = kx * w
            sx[y, x] = Ix.sum()
    return sx

A = np.array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                   [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                   [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0],
                   [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                   [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
                   [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]])


kx = np.array([
    [-1, 0, 1],
    [-2, 0, 2],
    [-1, 0, 1]
], dtype=np.float)

ky = np.array([
    [1, 2, 1], 
    [0, 0, 0], 
    [-1, -2, -1]
], dtype = np.float)

Gx = signal.convolve2d(A, kx, boundary='symm', mode='same')
Gy = signal.convolve2d(A, ky, boundary='symm', mode='same')

# calculate homemade sobel x gradient
myGx = sobelx_homemade(A, kx)

plt.subplot(131)
plt.title("Original")
plt.imshow(A, cmap="gray")
plt.subplot(132)
plt.title("Gx")
plt.imshow(Gx, cmap="gray")
plt.subplot(133)
plt.title("My Gx")
plt.imshow(myGx, cmap="gray")

I expect the images labeled "Gx" and "My Gx" to be identical.

lifebythedrop · Accepted Answer · 2019-03-26 03:36:43Z

1

So, it turns out that a true convolution flips the kernel/filter matrix, which explains the flipped result.

This video from Andrew Ng, explains the differences between textbook convolution and cross-correlation commonly used in machine learning style image processing: https://youtu.be/tQYZaDn_kSg?t=308

answered Mar 26, 2019 at 3:36

lifebythedrop

4417 silver badges18 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Applying Sobel operator in numpy to simple binary image

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related