Abstract:

This paper presented a linear and non-linear face recognition method based on appearance. The linear algorithms used include principal component analysis (PCA) and linear discriminant analysis (LDA). Two non-linear methods were kernel principal component analysis (KPCA) and nuclear fisher analysis (KFA).Linear dimensionality reduction projection method was based on second-order dependency coding mode information, and non-linear method was used to deal with the relationship between three or more pixels. In the last stage,the Mahcos measure was used to define the similarity measure of two images after the corresponding dimension reduction technology.Experiments show that when used together with Gabor wavelet,LDA and KFA had the highest performance, 93.33% of CMC and ROC results, respectively. Through the comprehensive analysis of 400 images in AT&T database, it was found that the performance of linear and non-linear algorithms was affected by the number of image classification, image preprocessing and the number of face images in recognition test set.

Key words: principal component analysis, linear discriminant analysis, kernel principal component analysis, kernel fisher analysis, face recognition, dimension reduction technology