The Research Of Face Recognition Based On Video Surveillance

In the field of computer vision and pattern recognition, face recognition technology research started earlier. After decades of development, the research of face recognition technology has been more and more mature. Currently there have been many commercial face recognition systems, which under ideal conditions have been recognized by the people. However, in practical applications, due to its face image and various external conditions, such as the expression of the quality characteristics of the method, the dimension of the feature vector size, face recognition technology is not widely spread, so looking for a simple and effective identification method of small amount of calculation, the popularization and application of face recognition can be widely.Face recognition is through face image feature extraction, and then compared with the characteristics of the training in the knowledge base to classify and recognize. Finally, the recognition results are given, which is the test face image belongs to the category of the database.This paper made a research on the face detection, image preprocessing, feature extraction, classification and recognition problems, the main work and achievements include:1. Introduces the basic principle of Adaboost algorithm for face detection.Due to the requirements of face recognition in video monitoring of real-time, this paper adopts the fast Adaboost algorithm for face detection, the detection rate and the computing speed can meet the requirement of real-time system.2. Summarizes the advantages and disadvantages of LBP operator and study the effects of LBP face recognition factor.In this part, we summarize the advantages and disadvantages of LBP operator based on the detailed description of LBP operator. By experiment shows the effect of traditional LBP face recognition factor:pretreatment and block the way. The results showed that:LBP pretreatment significantly improved recognition rate of recognition; block size has a great influence on recognition rate, the higher the block number the more the recognition rate, but blocks too much lead to a fall in the recognition rate, while too much will make the block feature vector dimension is too large, excessive computation time and storage space is not conducive to the practical application of face recognition.3. Study an improved LBP operator based on LBP operator and Adopt City Block Distance as LBP histogram similarity measure method.Experimental results on ORL face database show that the improved LBP operator is better than the traditional LBP operator and still keeps the advantages of LBP operator and using City Block Distance as LBP histogram similarity measure method is superior to using Chi square of the distance.4. A face recognition system is designed and developed.Through the research of face detection and face recognition method, we design and implement a simple face recognition system based on matlab.