The Research On Face Recognition Based On High Discriminative SIFT And LGQP

With the progress of science and technology and the rapid development of multimedia technology,face recognition technology has been widely applied in various applications including the field of security management,which illustrates the importance of face recognition technology.The key of face recognition technology is to extract effective information from face image which is one of the important factors in the task of face recognition.Therefore,how to extract the most effective features from a large amount of face images has become a hot spot in face recognition technology,this paper improves the traditional face feature extraction methods based on high discriminability of SIFT(Scale invariant feature transform)and LGQP(Local Gabor Quaternary Pattern).This paper mainly consists of the following three aspects:1.This paper proposes a novel adaptive sub-block strategy to improve the effectiveness of recognition algorithm.The face image is partitioned into sub-blocks according to the distribution of SIFT feature points,which leads to result that each facial feature lie in the same sub-block and reduce the impact on recognition results due to the pose or facial expression variation.The experimental results on ORL,YALE and JAFFE show that the effectiveness of the proposed method.2.This paper proposes a local quaternary model(LQP)operator which calculates the intensity variation with mean and standard deviation of local region based on the traditional LBP(Local Binary Pattern)operator.Furthermore,LQP operator is combined with Gabor filter and exploited to extract LGQP features from face images.The experimental results show that LGQP features is more stable and reliable than LGBP and LGTP.3.Inspired by the Fisher linear discriminant analysis theory,this paper proposes an algorithm for calculating the discriminability of SIFT key point based on the inter-class and intra-class relationship.After using this algorithm to calculate the discriminability of SIFT point and then calculate the LGQP features of the SIFT point in different orientation and scales.Finally,LGQP features with different orientation and scales and SIFT features are combined together to complete the face recognition.