texture detection

[vekkuli]

I put it together so that it compares the histograms of two images. I'll attach it here tomorrow when I'm back on that computer. However, if I got it right, if you want to match objects in an image you'll want to compare the histograms of different areas in the image, not the whole image. For this you might want think about an approach.

The study in the pdf is about two separate methods: color histogram matching and the LBP. They both evaluate using the histogram intersection but I've made the assumption we're only interested in the LBP solution for now.

[Madhubalan]

Hi vekkuli,

              In the pdf ,they are using histogram bin reduction,intergal histogram and histogram intersection for histogram matching,if you can provide source code for histogram matching using these steps along with LBP,it will be very helpful for me.

Thank you

[vekkuli]

I suppose you're on a good track there. I finally realised what they are actually doing in the paper as well. Overall, they use the integral color histogram matching to simply produce the candidate areas for the LBP method to check. But like you said, I don't see why we couldn't apply the integral histogram to the LBP map as well and leave out the color histogram part altogether. If you google integral histogram you'll find at least this paper, http://www.merl.com/papers/docs/TR2005-057.pdf, describing the technique. Give it a look and try to, for instance, mimick the pseudo-code (Algorithm 3.1.) on LabVIEW yourself. The topic is interesting but I'm afraid I don't have all the time in the world to help you at this very moment.

[vekkuli]

Here are the VI's I talked about earlier.

[pawhan11]

Long time ago I had to implement something called local binary patttern for crack detection. 

I don't remember much of this code and implementation itself but it was working.

Hope it helps 🙂

[Madhubalan]

Hi Vekkuli,

               In the LBP file that you had given,in which i could able to understand everything except the values of +X and +Y,how you are calculating and why we are giving this value as input?

Thank you

[vekkuli]

Those values give the coordinates of the 3x3 neighbourhood pixels relative (+x, +y) to the center pixel (x, y). I chose to assemble the bits from the neighbourhood by starting from the top-left corner (+x = -1, +y = 1) and advancing clockwise. You can, of course, use an arbitrary order but you need to stick to one in the LBP calculations.