Research And Realization Of Face Detection And Recognition

With the rapid development of science and technology, the needs for security verification and information identifying of the people are becoming more and more widely. The detection and recognition of target image are important research contents. As one of the most important visual image, facial image has an important position in the research of machine vision, multimedia technology and information processing technology. Face detection is the first step and important procedure of facial information processing. It has become an independent and active research subject all over the world.As the important procedure of intelligent video surveillance system, face detection and face recognition method has achieved great development. Based on the analysis of previous work, this thesis uses AdaBoost algorithm in face detection. With the training samples of face image and non face image, this thesis builds AdaBoost classifier based on Haar-like features and integral graph. Meanwhile, this thesis builds cascade classifier and improves the performance of face detection so that this classifier can meet the requirements of actual application. Do face detecting for real images by using AdaBoost cascade classifier, this thesis has achieved good results.On the basis of AdaBoost algorithm, this thesis proposes a face detecting method combined AdaBoost with SVM. This method retains the AdaBoost cascade framework and uses the SVM classifier to replace the traditional AdaBoost classifier. In the process of training SVM classifier, extract several Haar-like features which have the best performance of classifying from face and non face samples. Then build the SVM classifier. With high fitting degree in the training sample, this method ensures good generalization ability by checking with a test set. And then this thesis makes the classifier pay more attention to the face samples by using different punishment factors. Meanwhile, this thesis proposes the improved method for the extraction of Haar-like features, which reduces the calculating quantity and improves the computing speed.After realizing the face detection, face recognition is needed as the next step to judge whether the detected face is in the target face library. This thesis uses the PCA method in face recognition process, projects higher dimensional face information to the lower dimensional eigenface subspace. Then judge whether two faces are similar through the distance of coefficients.Finally on the basis of actual demand, this thesis designs a human face detection and recognition system with the core of DM642. This system includes storage module, video encoding and decoding module, audio module, clock module and power module, etc.