A Face Recognition Method Based on Wavelet Transform and Sparse Representation

Abstract

The invention discloses that each face image to be trained is decomposed by wavelet transform to obtain corresponding low-frequency components and three high-frequency components corresponding to different details, and then the three high-frequency components are fused by principal component analysis method, Obtain the corresponding final high-frequency fusion training image, construct a low-frequency dictionary and a high-frequency dictionary respectively, and finally perform sparse representation of the face images to be tested in the test sample in the low-frequency dictionary and high-frequency dictionary respectively to realize face recognition, And introduce the concept of cross-correlation coefficient to further increase the accuracy of face recognition; the advantage is that by setting the low-frequency dictionary and the high-frequency dictionary, the preliminary category of the face image set that the face image to be tested belongs to in the standard face database can be delineated, And by introducing the cross-correlation coefficient to further determine the category of the face image set that the face image to be tested finally belongs to in the standard face database, thereby improving the recognition rate and reliability of face recognition, illumination robustness and expression The robustness is better.

Description

Technical field [0001] The invention relates to a face recognition method, in particular to a face recognition method based on wavelet transform and sparse representation. Background technique [0002] Face recognition technology has become an increasingly popular research direction in the field of pattern recognition and artificial intelligence because of its huge application prospects in system security verification, identity management, credit verification, video conferencing, human-computer interaction, and smart home. . In the past, many face recognition methods have low recognition rates under the conditions of illumination changes and expression changes. Therefore, the research of highly robust face recognition methods is still one of the challenges and important contents in the current face recognition research. [0003] Traditional global feature extraction algorithms such as Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), etc., have low recognitio...

AI Technical Summary

Problems solved by technology

Yang et al. analyzed Gabor features combined with SRC, and proposed a Gabor feature-based sparse representation face recognition (GSRC) method. Since the Gabor features extracted from the local area can avoid the influence of many unfavorable factors, this method improves human face recognition. Face recognition rate, and has certain robustness, but Gabor transformation needs to be carried out in different scales and directions, so the calculation complexity of this method is high, and the recognition time is long; LDA considers the category information of samples, and is suitable for Classification, Zhang Yong introduced the LDA method into SRC, and realized a face recognition method based on linear discriminant analysis and sparse representation classification (LDA-SRC). The SRC method cannot be carried out smoothly on many face recognition problems, and the method itself does not consider the influence of factors such as illumination, expression, and occlusion; Liu Zi et al. introduced a Greedy Search (GS) idea into SRC method, proposed a face classification method based on sparse representation and greedy search (SRC-GS), the recognition rate of this method is ideal in the case of face occlusion, but it did not verify the robustness of other aspects of the algorithm, and The computational complexity is high

Tang et al. proposed a weighted group (Weighted Group) sparse representation classification method (WGSRC), the method for mixed l 1，2 The norm is weighted, and finally the sample is recognized by regularizing the reconstruction error. This method has good robustness to the pose change of the face, but the weight required for recognition is set according to experience, and Its recognition rate is low on face databases with large illumination changes

Images