Study On Some Key Technologies In Face Recognition System

Face recognition is a biological identification technology that based on face feature information, and it combine machine learning, image processing, artificial intelligence and so on many specialized technology. In order to improve the recognition rate and shorten time of the recognition, the academic circle has invested major efforts in the study of algorithm. Recent years have witnessed an positive experiment results by researchers represneted by H. Cevikalp who applied the Image Set to identify a person by modeling a simple affine hull or convex hull on the basis of a set of images of a person from a clip of video. This paper aims to bridge theory and reality by applying facial recognition of image set to real time system. Generally speaking, the main work is just as follow:(1) The work includes a thorough study on facial recognition algorithm of image set and its processing, and a thorough study on how to model image set and how to get distance. This paper compared LPQ and LBP with combining the algorithm of image set and SRC to study its effectiveness in facial identification. With experiment on AR and Yale face data base, the results showed that the combination of image set and LPQ is of better recognition results and better robustness against light noise. Meanwhile, it is also concluded that facial recognition based on image set can better exploit the continuity of video images for better recognition results.(2) Get rid of images with too much poise deflection by applying SVM and LBP to categorizing. The recognition rate is 94.73% in our own facial poise data base and the running time of procedure is quick. The results showed that it is possible to lower the parameter g(g is gamma coefficient of radial basis kernel function)without sacrificing the recognition rate and that the poise is subject to the parts below nose. We also carried out face landmarks with the dlib open-source dababase. With 68 face landmarks, we can emulate the proportion of the left-side and right-side face to get an approximate angle for further modeling of the facial image.(3) We make research on IP camera and some functions of Software Development Kit(SDK). By making research on the IP camera, its data formats and the process of get frame, we rewrite the streaming data formats. So we can use Open CV to deal with the image.(4) We design a real-time face recognition system based on IP camera and Image Set algorithm by way of Open CV and C++ programming development, and the interact through MFC interface implementation. First, Ada Boost algorithm is used to detect face in each frame and join the nose detection for secondary screening, at the same time by estimating pose filter out the images of posture deflection, and then LPQ is used to extract the features of texture. Finally, we obtain the results by Image Set algorithm. Experimental results show that the system can deal with real-time video and has robustness to the illumination and the change of posture.