I have a doubt regarding the histogram on the LBP algorithm.
Let's say I have an image that is 1000x1000 and I split this image into 4 sub-images of 250x250 each so as far as I know I should compute each sub-image histogram generating a histogram vector of 59 positions (Uniform LBP version).
My doubt is, the Wikipedia article says I have to concatenate each histogram vector to generate the feature vector so the feature vector of my image is a vector with 236 positions ? 59 positions * 4 sub-images.
Is it wrong to compute the whole image and generate just one 59 position histogram vector as my feature vector?
With more splitting you get more information about the image: what are the textures on different parts of the image. Without splitting, you get only an average texture. Just imagine you want to recognize a cat from a histogram, or from an image with 100*100 pixels.