how to compute percentage of similarity between histograms (feature vector)

Question

I need to compare and compute similarity between local binary pattern(LBP) histograms of grey scale images for face recognition.

Computed histogram is in array form and used Accord.net to compute it.

(Histogram is int [,][] form and when I get a histogram using break point, i obtain int [3,3][256] so they are combination of [0,0], [0,1], [0,2], [1,0] ...... [2,1], [2,2] and inside of [,] array, there are 256 values from 0 to 255)

My code for obtaining histogram is this;

//LBP setting is default which are;
//Cell size of the histogram is 6 pixcel
//Block size of the histogram is 3 cells
 private int [,][] histo (BitmapSource croppedImage)
 {
     var converted = BitmapFromSource(croppedImage);
     var output = LBP.ProcessImage(converted);
     var qwert = LBP.Histograms;
     { LBP.ProcessImage(converted); };

     return qwert;
 }

How can I get the percentage of similarity between histograms?

Could you just provide me some hint to compare them?

Additionally, I have total 48 histograms under 1 image. Is there any effective way to combine 48 histograms together for comparison with another set of 48 histograms? and any recommendation of effective library then Accord.Net to compute LBP histograms? Information and documents for Accord.Net is very few and it makes me harder to learn about it as I am a beginner.

Thank you for reading my post and I really appreciate your help.

Muhammet Ali Asan Muhammet Ali Asan · Accepted Answer · 2016-07-03T07:03:36

OpenCV is leading image processing library which provides many of widely known algorithms.It is available in CSharp .

For histogram comparison, Chi-squared distance is a good method. It is available in OpenCV.

For your 3rd question,concatenating histograms,you have to normalize concatenated histogram into [0-1] range for better comparison.

how to compute percentage of similarity between histograms (feature vector)

1 Answers