Face recognition using variance weightiness LBP based on single training image per person

A novel and efficient method for face recognition with only one training image is proposed in this paper. It combines both texture feature and global topological information, and merges different features with proper weightiness. Firstly, we partition facial images according to “Three courtyards and five eyes” theory. Secondly, LBP is used to exact texture feature, and then we choose variance for weightiness to merge features, here we make use of variance to estimate the scattered degree between classes. Finally, the Chi-square is used to measure similarity, and we send them to the nearest classifier for further recognition. On ORL facial database, experiments show a better effect and higher recognition accuracy of this method.