Face Recognition Based On Gabor Pixel-Pattern-Based Texture Feature (GPPBTF)

Face recognition has been one of the hottest research topics in computer vision for recent two decades due to wide potential applications in security,finance,law enforcement and military.Its research achievements promote the development of correlative disciplines.With the continuous development of face recognition technology,the larger image library,the higher recognition rate,the faster the rate of identification has become the new direction of face recognition technology.At present,there are already quite mature face recognition technologies using in real life.Since Pixel-pattern-based texture feature(PPBTF) has been applied to real-time facial expression recognition system successfully,we use it in face recognition and propose a novel face representation approach,Gabor Pixel-Pattern-Based Texture Feature(GPPBTF).Gabor wavelets can extract image's local information exactly and is robustness to displacement, deformation,rotation,scaling and illumination variances.PPBTF can describe image's texture information and is insensitive to illumination and time saving.This paper combines Gabor wavelets and PPBTF's advantages to represent face image.The combining operator is termed as Gabor Pixel-Pattern-Based Texture Feature(GPPBTF).Null Space-based Kernel Fisher Discriminant Analysis(NKFDA) has been used in face recognition successfully.NKFDA can map the feature vector into null space of kernel function to get more effective discriminant features and reduce the dimension of feature vector.This paper proposed a method based on multiple NKFDAs to classify the face images which solve the problem of losing useful information when mapping the high dimension vector into low dimension space.This paper proposes a method which combines GPPBTF and multiple NKFDAs to classify the face images on FERET database.This method achieves higher recognition accuracy.