Research On Face Recongnition Based On The Mean And Variance Of The Local Texture Features

Compared with other biometrics, face feature as a special biometirc has several natural advantages, such as easy acceptance, low-cost, etc. After several years of development, face feature is widely used in various fields for system identification, which is more mature in theory. The extraction of facial features is a key factor to affect the result of recognition. How to extract face features and describe them effectively has been a focus, which is also the focus of this article.The local binary pattern (LBP) Operator is defined as a gray-scale invariant texture measure, derived from a general definition of texture in a local neighborhood. It was first introduced as a complementary measure for local image contrast. Recently, the LBP has been successfully applied to face recognition as texture descriptor and excellent result has achieved. Since edge gradient is more stable than pixel intensity, T. Jabid et al proposed Local Directional Pattern (LDP) which extract feature from the edge response. However the LDP operator is sensitive on uneven lighting conditions. In response to this shortcoming, the main contribution of the dissertation are as following:1. The dissertation studies based the image mean, variance,and mean-variance LDP feature in face recognition. Compared the features, under the same test conditions, shows that based variance LDP feature are unsatisfactory on the recognition results. Can be consid-ered apply image variance in other way; The LDP algorithm based on mean is optimal, the algorithm reduces the feature dimension while maintaining a high recognition rate.2. By analysising the lack of the variance LDP feature, the dissertation proposed image Variance Ternary Pattern (VTP). This algorithm can well be described as the image edge features, with simple computation, little calculation, etc., and compared with LDP based on variance has higher recognition rate. However VTP operator "ignore" describes a flat region in the image, so that the algorithm reduces the overall description of the characteristics of the image.3. The dissertation proposed a fusion feature (F-VTP) which combined with VTP features and features of the mean LDP. F-VTP combined VTP features and features of the mean LDP, Full use of the advantages of these two features, in the case of uneven illumination, greatly improve and enhance the accuracy of identification. Experiments show that the proposed F-VTP operator is superior to other operators.4. The dissertation proposed some improved k-Nearest Neighbor. In the face recogniti- on research, training face set contains multiple sub-categories and the value of k is fixed. For these two problems, the dissertation proposed some improved matching algorithms of k-NN which is more suitable for face recognition. Under the same conditions, the improved algorithm obtain a higher recognition rate.