A Research On Algorithms For Face Recognition Technology Based On Independent Component Analysis

Face recognition technology is a recognition method based on biologicalfeatures. Compared with fingerprint recognition and other traditional means ofidentification, it is easier to be accepted by users because of its accuracy,concealment and non-intrusive features. Just for this reason, face recognitiontechnology has a wide range of applications in many fields. In recent years, facialfeature extraction in face recognition technology has become one of hot spot basedon biological characteristics.Based on summary of the content and method of face recognition technology,this paper discusses in detail three types of feature extraction method based onsubspace analysis: PCA (principal component analysis) method, LDA (LinearDiscriminate Analysis) method and ICA (Independent Component Analysis)method. Then we elaborate the main idea of the three methods in detail andintroduce their algorithm process and implementation. Meanwhile, we conductexperiment analysis of the three methods on the OR L face database and discuss themain influencing factors of recognition rate of the three algorithms. Theexperimental results show that ICA method has a higher recognition rate than theother two. In this paper, we present two dimensional Gabor wavelet tr ansform andits application in extraction of facial features. Gabor wavelet has a big advantageon expression of local features of face. By making use of this advantage, a facerecognition method is proposed combining two dimensional Gabor wavelettransform with ICA, which improves the algorithm. Face image is represented withthe amplitude and phase of Gabor wavelet, and the amplitude and information ofthe phase are considered as the basis of recognition. Separating matrix is evaluatedby ICA method, and recognition according to their types is realized by nearestneighbor classifier. This method combines both the advantages of Gabor wavelettransform and ICA method. Experimental results show that the improved algorithm has a high recognition rate, and it still maintains high recognition rate, more than90%, especially in cases with small training samples, which is favorable forpractical application.