Group of training pixel selection in one image

Hello, I'm working on a detection of wildfire in an image using machine learning. I'm confused on how can I get the training area in the image. The image is the result of a PCA differencing between a before wildfire image and a post-wildfire image. In the resulting image, I want to select a group of pixels to be the training area for a future supervised classification machine learning. My question is, How can I select the pixels in the image? Is there any algorithm from Matlab which, when I select 1 pixel, will check the surrounding pixels to see if there is any resemblance, delivering me a matrix grouping all the pixels? Or if there any other suggestion on how to perform a pixels selection training, they are most welcome.

I've been stuck with this too much time and I would really appreciate your help.

Thank you!

Answer 1

[Image Analyst]

You can use imfreehand() to lasso an irregular pixel region.

You can use imfilter(), conv2(), and nlfilter() to do operations on a pixels and some or all of its neighbors. For example to get the mean difference between a pixel and its 8 neighbors, do this

kernel = [-1,-1,-1;-1,8,-1;-1,-1,-1]/8;
meanNeighborDiff = conv2(double(grayImage), kernel, 'same');

[ben dp]

Thank you! I will try as you writed