Based on the multi - template matching with complete KFDA Face Recognition Technology
|School||Nanjing University of Technology and Engineering|
|Keywords||Face Detection Feature Abstraction Skin Segmentation Multiple Related Templates Matching Complete Kernel Fisher Linear Discriminant Analysis (CKFD)|
Compared to other biology feature recognition, face recognition has more advantages. It takes on wide and particular applications in national important departments and society security and has been an important subject in the fields of computer vision, pattern recognition and artificial intelligence in recent years.This paper respectively makes researches on face detection and feature abstraction algorithms which are two parts of automatic face recognition system. The relevant software is also empoldered.In the face detection part, an algorithm based on skin segmentation plus multiple related templates matching is designed, which firstly constructs a skin model and then two-eyes-in-whole template is used for determining the potential location for face roughly after pretreatment. The whole face templates with different width-height ratios are used for searching and localization. The experimental results show that the method is relatively fast and has good detection result even in complex background. Furthermore, not only could it correctly detect the full faces, but also certain side faces.In the feature abstraction part, a new kernel Fisher discriminant analysis algorithm called complete KFD (CKFD) is studied out. This method constructs a new framework in the case of small sample size problem, which makes KFDA algorithm more transparent and simpler. Besides, CKFD algorithm is capable of making use of two categories of discriminant information for feature abstraction, i.e., the discriminant information within and outside of the null space of the within-class scatter matrix, which makes it more powerful. The experimental results in YALE face image database demonstrate that recognition capability in this method is not only much better than linear projection analysis but also has certain-advantage contrasted with KPCA and KFDA.