Design Of Video Face Recognition System Based On SURF

Currently, the main use of biometric technology with the body's own characteristics,including physical characteristics and behavioral characteristics to come on identity authentication and identification, and physiological characteristics(fingerprint, iris and face)is commonly used. In recognition of the identity of the process, the biometric technology IT and physiological characteristics of an effective combination of people, has bee slowly replacing traditional authentication methods. Compared to fingerprint and iris recognition is a non-contact identification technology, through the whole facial feature recognition, has a higher natural, acceptable and uniqueness. And the human face is the most common mode of communication, is of great significance in the interaction between people, and it has become a hot research field of biometric identification technology.In order to detect the face in the video.Face detection and recognition are two key issues are sensitive in the video face images Face Detection for multi-gesture and multi-light conditions resulting face image and other interference factors; and because the face recognition process calculation the problems led to large real-time performance is relatively poor, often can not meet the actual demand.To solve these problems, respectively Viola-Jones and SURF algorithm. In order to facilitate the use of algorithm for the realization of the system, this paper chooses VS2010 and OpenCV to complete the system.Since the SURF feature vectors are high-dimensional vector, the conventional algorithm based on nearest neighbor matching algorithm, the matching process is equivalent to a high-dimensional space nearest neighbor search problem, too computationally intensive. By some relevant comparison algorithm for high-dimensional space nearest neighbor search problem, stratified K-mean trees and multiple random KD tree has better performance.Since this paper the feature point matching stage face image matching algorithm using two-way FLANN due FLANN feature matching speed, two-way FLANN matching algorithm at the cost of a small amount of time to enhance the exchange rate of the match. Analysis of thedemand for recognition experimental platform during the face recognition system uses OpenCV library functions related to integration of the corresponding image processing functions.Finally, the experimental platform to achieve face recognition, face recognition and laboratory at Georgia Tech face database collection on the experiment, each experimental results show an image of the face region out are properly detected, and correct rate of 81.7%.Average each took 0.685 seconds. Basically reached the expected functions proposed real-time video face recognition system.