# Why conv2 gives opposite sign

1 visualización (últimos 30 días)
xiaojuezi el 23 de Ag. de 2020
Comentada: Image Analyst el 24 de Ag. de 2020
Hi, I have a simple matrix I:
I = [1,2,3,4,5;
6,7,8,9,10;
11,12,13,14,15;
16,17,18,19,20;
21,22,23,24,25]
My kernel is:
k = [0,0,0;
1,0,-1;
0,0,0]
I would like to compute for the non-boundary elements. For each element, I want to compute its left neighbour - its right neighbour,i.e.:
dIx = [~,~,~,~,~;
~,6-8,7-9,8-10,~;
~,11-13,12-14,13-15,~;
~,16-18,17-19,18-20,~;
~,~,~,~,~]
I used this line to achieve this:
dIx = conv2(I,k,'valid');
This gives me the result:
dIx = [0,0,0,0;
0,2,2,2;
0,2,2,2;
0,2,2,2]
which is the opposite sign of what I expected. Why is the case?
Thank you very much.
##### 0 comentariosMostrar -2 comentarios más antiguosOcultar -2 comentarios más antiguos

Iniciar sesión para comentar.

Matt J el 23 de Ag. de 2020
Editada: Matt J el 23 de Ag. de 2020
Remember that a convolution does not slide k over I, but rather a pre-flipped version of k. You can compensate for this by doing,
k = [0,0,0;
1,0,-1;
0,0,0]
k=fliplr(flipud(k));
dIx = conv2(I,k,'valid');
although for the kernel you've shown this pre-flip happens to be equivalent to k=-k.
##### 1 comentarioMostrar -1 comentarios más antiguosOcultar -1 comentarios más antiguos
xiaojuezi el 24 de Ag. de 2020
Thank you very much!

Iniciar sesión para comentar.

### Más respuestas (1)

Image Analyst el 23 de Ag. de 2020
Convolution flips the kernel - that's just how it works, and there are theoretical/mathematical reasons for it. Perhaps you can use imfilter() instead if you don't want to flip the kernel to be flipped.
##### 3 comentariosMostrar 1 comentario más antiguoOcultar 1 comentario más antiguo
Matt J el 24 de Ag. de 2020
filter2 does have that control, however. You could use that to avoid implementing your own kernel flip.
Image Analyst el 24 de Ag. de 2020
Here are the boundary options for conv2:
Here are the options for imfilter():
It looks to me like imfilter() has more options than conv2(). What control did you want to have over the boundary that conv2() has but imfilter() does not? 'Valid'? You could always just use indexing to crop out that part yet still avoid the flipping of the kernel, which you don't want to do.
outputImage = outputImage(2:end-1, 2:end-1);

Iniciar sesión para comentar.

### Categorías

Más información sobre Neighborhood and Block Processing en Help Center y File Exchange.

### Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by