Intra-facial-feature canonical correlation analysis for face recognition

In the real-world face recognition tasks, the limited size of facial images and the lack amount of training data would lead to fail the work or degrade the recognition accuracy dramatically. Face recognition approaches such as principal component analysis (PCA), linear discriminant analysis (LDA) and linear regression classification (LRC) are limited by the number of collected facial features. The traditional canonical correlation analysis (CCA) can address the relationship between two multivariate data sets that could be miss portion of important messages. In order to handle the insufficient collected facial information and improve the CCA, we propose intra-facial-features canonical correlation analysis (ICCA). The ICCA involves multiple sets of multivariate of different facial features that could be eyes, nose and mouth. Moreover, the proposed approach can also calculate the relationship among the intra facial features. Experimental results show that the proposed approach achieves better recognition rate than the traditional statistical analysis on the AR face database.