LBP algorithm of OpenCV

edit

HI Everyone,

I am working on LBP based face detection on openCV. I wanted to understand the algorithm clearly but I could not find any clear documents on these algorithm or their implementation. I have some very specific questions. Here they are:

1. In lbpCascade_frontal_face.xml: what is maxCatCount, maxWeakCount, minHitRate, maxFalseAlarm, weightTrimRate, maxDepth, internalNodes, leafValues, tree etc? I know they are used in the algorithm. But I do not understand the meaning of each one of them. For example, why we take a particular feature and not the other for a particular stage? how we are deciding which feature/ node to choose?

2.What is feature values in the LBP_frontal_face_classifier.xml? I know they are a vector of 4 integers. But how should I use this features? I though stage 0 access to the first feature but access is not in this pattern. What is the access pattern to this features?

1. All the papers in literature give a high level overview only. Their descriptions mainly consist of LBP calculation from neighborhood pixels. But how this LBP values is used against those elements in the classifier?

2. How does intergral image help in calculating LBP value of a pixel? Please do not give reference/answer using HAAR features. I know how HAAR is used. I need to understand LBP.

I read some papers, articles. But none clearly describes how LBP based face detection works or the algorithm in details. If someone wants to develop a face detection program on his own,what are the steps he should follow- no document describes that.

If anyone can help me with the answers it would be a great help for me. -you can answer the questions or you can direct me to someone who knows the answers or you can give me some link of articles/documents/papers that would help me to understand this algorithm.

Thanks a lot for reading. Eagerly waiting for your answers. I am badly in need of those.

-Hassan