Detecting 8 connected neighborhood of a object in image

Hello

I need to find robust method for detecting the 8-connected neighbor of a rectangle-like object in an image.

I was considering something like dilation on the image but this does not seems to detect all the adjacent pixels in every case (considering change of shape) so know I am looking for another approach.

If you have any idea please let me know !

Regards

Answer 1

[Matt J]

You mean you have a binary image and you want to detect pixels with 8 neighbours that are "1"? If so,

 kernel=[1 1 1; 1 0 1; 1 1 1]/8;

 idx = conv(image,kernel,'same')>=.998;  %.998 is a tolerance close to 1

and then if you want to, you can convert idx to subscript indices

 [I,J]=find(idx);

[Dimitris M]

Hello

I tried your algorithm but it doesn't seem to work. From what I have seen the conv() function requires vectors as input and the "image" and "kernel" are 2-D arrays !

Is there a way to work this out ?

Thanks again for the feedback

[Matt J]

Sorry, I meant conv2.

idx = conv2(image,kernel,'same')>=.998;

Answer 2

[Image Analyst]

Dmitris:

This is done with bwhitmiss() in the Image Processing Toolbox. Go here to Steve's blog to see examples: http://blogs.mathworks.com/steve/2011/07/08/binary-image-hit-miss-operator/. You just need to look at 4 cases, where the center pixel is true and a single corner is true. All of the other possible 252 cases will be 4 connected or not connected at all.