Implementing Sobel Gradient in Octave

Question

So I am supposed to apply a Sobel filter on an image in Octave, but I am not allowed to use any functions from the image package. I wrote the code but my output is just a black image. Here is what I have so far:

%Sobel Gradient

kx= [1 ,0 ,-1; 2,0,-2; 1, 0 ,-1];
ky= [1,2,1; 0,0, 0; -1, -2 ,-1];

H = conv2(kx,im2double(my_img),'same');
V = conv2(ky,im2double(my_img),'same');
E = sqrtm(H.*H + V.*V);
figure 4
imshow(E, [])

Thank you for the help!

Try normalising the image prior to display - I am no good on `Octave` but something like `E = max(0, E ./ (max(max(E)) / 255));` — Mark Setchell, Apr 06 '18 at 15:09
@MarkSetchell the square brakets on `imshow` shoudl be doing that job — Ander Biguri, Apr 06 '18 at 15:15
@AnderBiguri Ok, thank you! My assertion that I am no good on Octave was correct :-) — Mark Setchell, Apr 06 '18 at 15:17
@ray this is not a duplicate, the reason it does not work is very different! — Ander Biguri, Apr 06 '18 at 15:20
@AnderBiguri I flagged it as a duplicate because it contains the correct implementation of the Sobel filter. — rayryeng, Apr 06 '18 at 15:21
@rayryeng I see! I still think its not a duplicate, but if you think so feel free to dupe it again ;) — Ander Biguri, Apr 06 '18 at 15:23
@AnderBiguri I can't lol. You've already reopened it. It's OK. Have a vote from me. — rayryeng, Apr 06 '18 at 15:23

Ander Biguri · Accepted Answer · 2018-04-06T15:24:33.577

2

Few minor mistakes, and your problem is fixed:

1) you convolve your image by the kernel, not the opposite. Mainly because you are using 'same' and that will output the size of the first input, 3x3.

2) you dont want a matrix square root, just the normal elementwise square root.

my_img=imread('cameraman.tif') % available in MATLAB, not sure octave
kx= [1 ,0 ,-1; 2,0,-2; 1, 0 ,-1];
ky= [1,2,1; 0,0, 0; -1, -2 ,-1];

H = conv2(im2double(my_img),kx,'same');
V = conv2(im2double(my_img),ky,'same');
E = sqrt(H.*H + V.*V); % or sqrt(H.^2+V.^2)
figure
imshow(E, [])

edited Apr 06 '18 at 15:24

answered Apr 06 '18 at 15:18

Ander Biguri

35,140
11
74
120

1

Have my vote too, for being 8,342,602 times better at Octave than me, although multiplying that by my big, fat zero doesn't come to much ;-) – Mark Setchell Apr 06 '18 at 15:36

Implementing Sobel Gradient in Octave

1 Answers1