The Study Of Facial Feature Extraction Method Based On Gabor Wavelet Transform

Face recognition technology is a key technique in the field of biometric identification technology, which is widely applied in identification, information security, human-computer interaction and so on. Facial feature extraction is the key part of facial recognition technology and it directly affects the recognition rate. Gabor wavelet can simulate simple visual neurons in the mammalian visual system, and describe images’ local information in different scales and directions, which can be effectively used in facial feature extraction algorithm. Therefore, this paper mainly studies the facial feature extraction methods based on Gabor wavelet transform.Gabor wavelet can capture detail information of face images in different angles. On the one hand, images transformed by Gabor wavelet can produce40children sub-images in five dimensions and eight directions. If the number of training samples is very big, the computation and storage capacity will be pretty large. On the other hand, there is some negative information which affects the recognition rate included in the40children images. Therefore, how to reduce the high dimension of feature sets transformed by Gabor wavelet is a hot spot in this article.The first part of this paper introduce the research background, research significance, research status, face databases and the main content of face recognition technology. Then, the paper studies the facial feature extraction method based on HG2DPCA.Since the recognition rate of HPCA method is not very high, this paper introduce the HGPCA method to extract facial feature and reduce dimension. Since the heavy computation caused by matrix tensile according to the column of PCA method, the paper put forward the HG2DPCA method for facial feature extraction. Experimental results show the validity of the improved algorithm and confirmed the Gabor wavelet is very important for face feature extraction. The improved method increases recognition but causes time overhead. Although HG2DPCA algorithm chooses the sampling technique and2DPCA method to reduce the feature dimension, the computation amount is still a little high. Thus more research needs to be done. Finally, the paper discusses facial feature extraction technology based on the Gabor amplitude statistical property. For the feature information imperfection, the paper put forward a new facial feature extraction method based on Gabor amplitude phase statistical properties. The phase response of images transformed by Gabor wavelet can exactly reflect the local information in the images and has strong robustness to the environmental light intensity. So the phase feature is indispensable for feature extraction. Design and implementation of algorithm in Matlab experimental platform and experimental results reflect the superiority of the improved algorithm in recognition performance.