Face Recognition Algorithm Based On Gabor And Adaboost

In recent years, with the development of biometric identification technology, face recognition technology has more and more important theoretical value and extensive application prospect. It has become a research hot topic in the fields of pattern recognition, image processing and computer vision, etc.Because feeding of animal brain cortical simple cells have the same characteristics with Gabor wavelet kernel function in two-dimensional reflection area, that is the strong spatial position and orientation selectivity, and can capture corresponds to the spatial and the frequency of the local structure information, Gabor wavelet can well describe the texture features of the image, is widely used in digital image processing. Recently, more and more researchers apply Gabor wavelet in face recognition, and achieved very good results. Based on the use of Gabor wavelet for face feature extraction, a face recognition algorithm which combines Gabor wavelet and Adaboost is proposed in this paper. The main works are accomplished in the paper as following:1. Theoretical knowledge of two-dimensional Gabor filters was discussed, and elaborated on the existing face recognition method based on Gabor features.2. Based on the8orientations Gabor filters for facial feature extraction, this paper uses16orientations Gabor filters and24orientations Gabor filters to extract face feature. Increase the filtering orientation of the Gabor filters, can better extract face image texture information, so as to enhance the recognition effect.3. This paper construct Gabor feature points based on the grid structure, it reduce the amount of extracted Gabor features data. Then Adaboost is used to select the feature points which have the best ability of face image classification and combine them together to form a cascade classifier for face recognition.In this paper, the CAS-PEAL face database is used for Adaboost training. The experiments carried out on the FERET database is used the16orientations Gabor feature training result, the results on the FERET database are98.2%,99.5%,77.8%and75.6%recognition rates on the subsets Fb, Fc, Dupl and Dup2, respectively. Experimental results show that, increasing Gabor filter orientation can better extract facial texture feature. At the same time the structure of the cascade classifier of the Adaboost algorithm can greatly improve the recognition speed of the algorithm, it makes the proposed algorithm have more practical applications.