Joint Feature Learning With Robust Local Ternary Pattern for Face Recognition

ABSTRACT In the age of rising crime, there is a critical need for high security. Biometrics has now gained a lot of attention. Facial biometrics is a method that can identify a specific individual in a facial image by analyzing and comparing patterns. Hence position-specific discriminative information can be exploited for face representation using Joint feature learning method. Having learned these feature projections for different face regions, spatial pooling is performed for face patches within each region to enhance the representative power of the learned features. Moreover, JFL model is stacked into a deep architecture to exploit hierarchical information for feature extraction and it further improves the recognition performance. In this paper, the extension of LBP algorithm is proposed. LBP is normally sensitive to noise and Local Ternary pattern partially solves this problem by encoding the minimum pixel difference into a separate state. The minimal pixel difference may be easily overwhelmed by noise. Thus, it is difficult to accurately determine its sign and magnitude. In this paper, concept of uncertain state is introduced to encode the small pixel difference. Robust Local Ternary Pattern is combined with JFL for better feature extraction and to improve the robustness to image noise.