Local Phase-context for Face Recognition under Varying Conditions

In this paper, we address the problem of face recognition of low-resolution images under varying light, illumination and blur using local texture based face representation. The main contribution is the texture representation using Phase-Context which is based on four-quadrant mask of the Fourier transform phase in local neighborhoods.The contextual phase generates a more discriminative code filtering responses, and a more effective feature set than the Local Phase Quantization (LPQ) descriptor which is suffering from the influence of the noisy filter responses, the order relation breakdown of the generated codes, and the discretization effect of the quantization.The experimental results on CMU-PIE, extended YALE-B and CAS-PEAL-R1 databases show that the Phase-Context methodology is more descriptive than LPQ, outperforming the widely used Local Binary Pattern (LBP), and Histogram of oriented Gradients (HOG).