How does trainCasca​deObjectDe​tector work?

I was looking around to find the "algorithm" or detailed description behind this function in particular and anything associate to this toolbox. Can someone please direct me to the place where I can get a detailed description of why and how does this thing work? I'm working on a project about Facial Recognition using SVD and I'm using trainCascadeObjectDetector as the facial detector, and so I need to know the details behind its "magic" so I can explain it in my report. Thank you. I hope this is not a secret Matlab keep it to themselves.

Answer 1

[Dima Lisin]

Hello Dai,

Do you need to explain how the face detector itself works, or how the training of the face detector works?

vision.CascadeObjectDetector implements the deletection algorithm by Viola and Jones. It is a "sliding-window" approach, which slides a window across an image, and tries to determine whether or not there is a face in each location using a cascade of boosted classifiers. The original algorithm uses Haar-like features. vision.CacadeObjectDetector also gives you the option to use LBP or HOG features.

The Algorithms section of the vision.CascadeObjectDetector documentation gives a good high-level overview of what it is doing.

As far as the training process, there is a tutorial in the documentation explaining some of the details.

[Dai Tran]

Thank you for your answer. I need to know the mechanism behind the training process and how it's implement. As far as I know, the tutorial link you gave me only tell me the surface of the process, but it doesn't tell me what algorithm is use for each stage of training or what's the program is doing to the positive and negative part. Since this is a semester long project and I'm going to give a demonstration and presentation at the end of the semester. This will definitely be ask by our professor so I just want to be able to answer him in detail. Thank you again for your help