Face Detection Based On Haar-Feature Probability Distribution And SVM

The purpose of face detection is to indicate the face area in a defined image. The face detection demands a lot of calculating work which is the most important pre-processing of face recognition. How to detect the faces rapidly and effectively is an important project for the face detection.This thesis emphasizes on a unique face detection method which is based on Haar feature probability distribution, which is the revolution and improvement of the old research results and the summary of the current face detection methods. The algorithm includes two parts: first is the rough face detection; second is the exactness face validation.At the stage of rough face detection, the thesis introduces a new method based on Haar feature probability distribution. This method records the selection correction rate of every rectangle feature as zero is threshold value in the face sample data. Then it will select the best performed feature to detect faces by using accordant distribution application strategy. By the two processed above, it will results in a wonderful effect of rough face detection. The other outstand advantage of this method is that it does not need any un-face sample, which avoid the trouble of selecting these samples. Compared with the other rough detection methods like template matching, skin-color model or Mathematical Morphology, the new method has a lot of advantages.At the stage of exactness validation, the research used the combination strategy of the linear nose SVM classification and unlined SVM classification, in which the PCA method is used to degrade the date in the nonlinear classification. The eigenfaces are required through the PCA processing. The advantages of fast speed of SVM detection and the high correction rate of nonlinear SVM combine together, which influences the practice of classification and detection speed faster and result in a better effect of exactness validation.The thesis made a detailed introduction of rough face detection and exactness validation. It testes the whole process and proves the validity of this method.