Fast Face Recognition Algorithm Based On Facial Special Description

At present, face recognition technology for video surveillance has aroused widespread attention. Thus, lots of human and material resources have been put in this field by many national research institutions and commercial organizations, which have achieved a large number of research results. For face recognition based on video, it not only needs a high accuracy, but also needs a fast real-time recognition speed. Howerver, speed and accuracy is a pair of mutually restraining factors in the actual application. Therefore, it is of great importance to find a solution to meet with the two indicators. Furether, it is significiant for the extensive application of the product.Facial point description based SIFT algorithm has a good recognition effect, a strong robustness, and the ablility of recognizing a person under single gallery. However, low speed is its fital drawback, which results to the algorithm can not statisfy application requirement of real-time recognition. In the view of video face recognition application requirements, in this thesis, we study the algorithm of SIFT deeply, and improved the algorithm with key point extraction, descriptor creation, strategy matching. Compared the original algorithm with the improved algorithm, the recognition speed has been improved greatly under the expected recognition rate and robustness.When using DoG to extract key points, the establishment of gaussian pyramids needs a series of image convolution and operation, which are waste time greatly. Because of the defect about DoG extract key point slowly. In this thesis, using FAST key point extraction algorithm for the extraction of face image key point. For the key pints are too centralized in FAST algorithm, the maximum inhibition principle is used to select local optimal point.In the stage of establishing a key descriptor, gradient and the direction are used to build descriptors in SIFT algorithm, which is relatively complicated. And in the key points matching stage, for computer, relative to the binary operations, the floating point arithmetic spend more time and takes up more space. The Brief algorithm is fast when creating a descriptor.Aiming at this problem, Brief algorithm is applied into face recognition. Meanwhile, we combined Brief with FAST detection algorithm and partial matching strategy to propose a new facial key point description algorithm named FBrief.Under the comparing experiments of diverse gallery recognition rate, single gallery recognition rate, constraint recognition rate, and recognition time, the simulation results show that the algorithm we proposed obtains a faster recognition speed compared, with other face recognition algorithms with the same recognition rate. More important, when the image were obscure largely, the proposed algorithm also shows a better robustness than other algorithms.