Research On Multi-pose Face Recognition Via RS-SIFT Algorithm

Automatic face recognition, which has involved image processing, pattern recognition,machine vision, and psychology and so on, is always one of the most popular themes inartificial intelligence. With the rapid development of artificial intelligence, face recognitiontechnology has been engaged in some areas such as monitoring system, door security, crimeidentification and passport verification successfully. The current face recognition systemscan achieve a satisfied result under controlled environment. However, in the real life, thistechnology is far from perfect. Age variation, illumination variation, and pose variation arethree major challenges encountered by current face recognition systems. In addition, areal-time face recognition system is also limited by the computing time when the amount ofdata is very large. This paper will focus on feature extraction and dimensional reduction tosolve the multi-pose face recognition. The main jobs are indicated as below:1) Using the SIFT algorithm to extract the feature point, then find out the robust SIFTfeatures to create the face feature database based on the frequency of each feature pointappearing in variant images. The other feature points with redundant information will beabandoned. By this way, the accuracy can be increased and the store space and CPU timecan be reduced.2) In order to reduce the dimension of feature extracted by SIFT, rough sets theory isintroduced. This algorithm utilizes attribute reduction to decrease the dimension of featureextracted by SIFT, then a new RS-SIFT feature descriptor which can save the store spaceand to improve the real time performance is created to present the original image.3) Build a multi-pose face recognition model based on the proposed algorithm, andthen verify the performance and robust of this method in Matlab.