Human face recognition method based on fuzzy two-dimensional kernel principal component analysis

Abstract

The invention discloses a human face recognition method based on fuzzy kernel-based two-dimensional principal component analysis (K2DPCA), which comprises the following steps: introducing a fuzzy concept to the K2DPCA, calculating the category of samples in a high-dimensional feature space (for short) and the subordination degree of each sample to the category by using the fuzzy K nearest neighbor algorithm, and defining fuzzy divergence matrixes of the samples according to the subordination degree information. Because the category of the samples and the distribution information are completely fused into the human face feature extraction, the marginal problem caused by the reason that the human face is easy to be affected by illumination, expression and other factors and the existing hard classification problem are solved. Due to the establishment of the category division data of the human face samples in the high-dimensional feature space, and the selection of the feature vector which can ensure that the divergence between the category is more than the divergence in the category after projecting as the optimal projection axis direction, the optimum projection axis selection accuracy is improved, and the capacity of representing the human face feature of the fuzzy K2DPCA is increased.

Description

technical field [0001] The present invention relates to a face recognition method, in particular to a face recognition method based on fuzzy two-dimensional kernel principal component analysis. classifier. Background technique [0002] Face recognition has become an important research direction in biometrics recognition, mainly by extracting effective face identification features and designing complex classifiers. Both Principal Component Analysis (PCA) and Two-Dimensional Principal Component Analysis (2DPCA) are linear feature extraction methods in the sense of least square error, and they cannot effectively extract the nonlinear structural features of faces and classify them. In response to this problem, Hui Kong et al. proposed a two-dimensional kernel principal component analysis (K2DPCA) method [Hui Kong, Lei Wang, Eam K T, et al. Generalized 2D Principal Component Analysis for face image representation and recognition[J]. Neural Networks , 2005, 18(5-6): 585-594.], i...

AI Technical Summary

Problems solved by technology

However, PCA, 2DPCA, and K2DPCA methods do not make full use of the category information of training samples [Chen Songcan, Sun Tingkai. Class Information- Incorporated principal Component Analysis[J]. Neurocomputing, 2005, 69(1-3): 216-223.] Therefore, Li Yongzhi and others proposed a method of nuclear principal component analysis of combined category information [Li Yongzhi, Yang Jingyu, Wu Songsong. A method of combined category information of nuclear principal component analysis [J]. Pattern Recognition and Artificial Intelligence, 2008, 21(3):410-416.], using the known category information of training samples for feature extraction, which partially solves the problem of using the category information of training samples, but there are still two problems. On the one hand, when the samples are illuminated When it is far away from the mean value of the sample (that is, the class center) due to the influence of , expression and accessories, etc., there will be marginal class problems [Zhuang Zhemin, Zhang Aniu, Li Finnish. Research on face recognition based on optimized LDA algorithm[J]. Journal of Electronics and Information Technology , 2007, 29(9): 2047-2049.], it is unscientific to simply classify samples into a certain class at this time, and it will cause hard classification problems [Yang Wankou, Wang Jianguo, Ren Mingwu, etc. Fuzzy Inverse Fisher Discriminant Analysis and Its application in face recognition[J]. Chinese Journal of Image and Graphics, 2009, 14(1): 88-93.]; on the other hand, only the eigenvector corresponding to the relatively large eigenvalue is selected as the optimal projection axis, discarding the subtle change information of the face [Ju Shenggen, Zhou Jiliu, Wang Chaobin, etc. Face recognition based on statistical feature fusion[J]. Sichuan University Journal (Natural Science Edition), 2009, 46(3): 618 -621.], such optimal projection axis selection is not accurate

Images