LBP built in function returning decimal answers

I need a pixelwise LBP for a project and I am trying to compare results to the built in extractLBPFeatures(...).

I tested the algorithm on an MNIST image due to the small size of 28x28, which gives 784 pixels. When I ran the following code:

LBP_Hist = extractLBPFeatures(MNIST_grey,'Upright',false,'Normalization','None');

I recieved an array of singles, which the values range from 1.3935 to 282.8942, however shouldn't the magnitude of the bins be equal to the number of pixels with that LBP value? The total sum of the LBP_Hist is 398.5727, which does not match the number of pixels, 784. I took off normalization but it seems to still be scaled. How does MATLAB calculate the LBP_Hist?

Answer 1

[Akash]

Hi Ryan,

I understand that you are comparing the results obtained from “extractLBPFeatures()” and have noticed some discrepancies in the values.

To address your concern, the sum of the “LBP_Hist” array does not necessarily need to be equal to the total number of pixels in the image. The “extractLBPFeatures()” function is designed to return extracted uniform local binary patterns from a grayscale image. These features encode local texture information that can be utilized in tasks such as classification, detection, and recognition.

To gain a better understanding of the specific calculations performed by MATLAB for “LBP_Hist”, I recommend referring to the MATLAB documentation on “extractLBPFeatures()”. This documentation will provide you with detailed information on the method and its underlying calculations.

https://in.mathworks.com/help/vision/ref/extractlbpfeatures.html

For a more comprehensive understanding of how LBP works and its applications, I suggest exploring the links provided below. They contain valuable information and discussions on this topic.

https://in.mathworks.com/matlabcentral/answers/351837-lbp-for-face-recognition?s_tid=answers_rc1-2_p2_MLT

https://in.mathworks.com/matlabcentral/answers/822600-what-is-the-output-value-of-extractlbpfeatures-function-represent