Face Detection Algorithm Based On The Combination Of Linear Asymmetric Cascade Classifier And Color Verification

With the development of society, high-speed effective status authentication is needed urgently in interactive technology, access control, criminal detection done by Police, and identity recognition etc. Compared with other biometric characteristics recognition, face recognition is direct, concealed, convenient, friendly, secure, and stable, it is studied by a lot of people.Linear asymmetric cascade classifier based on feature fuzzy classification algorithm and linear asymmetric cascade classifier based on feature space's double threshold algorithm are studied in this paper. Each of them is combined with color verification, then two real-time algorithms are designed. Software system of the two algorithms is developed. The work is as follows:Linear asymmetric cascade classifier real-time face detection algorithm based on feature fuzzy classification and it's combination with color verification algorithm are studied in this paper. First, fuzzy classification technology is used in harr-like feature in training, and harr-like feature weak classifier is designed. Second Fast Feature Selection(FFS) algorithm is used to select harr-like weak classification for every node. Third, Linear Asymmetric Classifier(LAC) algorithm is used to design harr-like weak classification selected for every node so as to form strong classification. While detecting, a node' strong classification can be chosen to detect face.To remove non-skin and find out face, the detecting result is dealt by color verification. Experiments show that the algorithm is timesaving in train, high efficient detecting, and timesaving in detection, and is of high detection rate.Linear asymmetric cascade classifier real-time face detection algorithm based on feature space's dual-threshold and it's combination with color verification algorithm are studied. While training, harr-like feature curve of face and non-face are drawn in the feature space, and two thresholds are find out. Then Fast Feature Selection algorithm is used to select harr-like weak classification for every node. Third, Linear Asymmetric Classifier is used to design harr-like weak classification selected for every node so as to form strong classification. Last, these strong classifications are cascaded to form cascade classification. While detecting, picture must be detected by every node in turn .Then the detecting result is dealt by color verification to remove non-skin and find out face. Experiments show that the algorithm is more accurate than traditional single threshold and the train time is less than them.Face detection software system is made in Visual C++.net 2003 with OpenCV1.0 Machine Vision Library. The software is of friendly interface, convenient use, free input picture format, and extensive use etc.