Human Face Detection Based On Support Vector Machine

Human face detection is one of the most important research fields on pattern recognition. To build fully automated systems that analyze the information contained in face images, robust and efficient face detection algorithms are required. This paper mainly make research on classify methods based on statistical theory, support vector machine(SVM), and feature extraction method—wavelet transform, and using them in human face detection.Support Vector Machines is a newly established theory about machine learning . It can train polynomial function, neural networks, or radial basis function (RBF) classifiers. It operates on a principle, called structural risk minimization, which aims to minimize the upper bound on the expected generalization error. Its flaw is that it needs a lot of time training, and when in case of coming up with more stylebooks, the final detection precision and rate will be affected without preprocession if trained directly. Through studying boosting arithmetic, This paper combines SVM and boost, and presents SVM+boost methods to construct classifier, so the final result can be improved.About wavelet transform theory, no doubt it's perfect now, but it's a big problem that how to apply it into human face feature extraction. So long as how to extract feature vectors that can represent human face features, this paper presents a reasonable way to extract wavelet coefficient through matlab experiment and testing results, that is ,sorting and projection which plus low frequency vector, and gives analysis on the experiment result affected by sorting and projectin methods. Thus we can apply SVM into human face detection in wavelet domain more reasonably. This paper also applies a way called "bootstraps" to choose non-face library. the coarse-to-fine theory is applied, too.This paper establishes a human face detection system based on SVM, and give corresponding result analysis after abundant experiments. The result shows that it can get better performance than tranditional ways about human face detection based on SVM.