Face Recognition under Varying Lighting Conditions: A Combination of Weber-face and Local Directional Pattern for Feature Extraction and Support Vector Machines for Classification

In the last two decades, an increasing number of illumination pretreatment methods and local feature descriptors have been proposed to address the issue of face recognition under different illumination conditions. Although these have achieved impressive results, the problem of how to maximize the reduction of the effect of variable lighting on captured images remains open. We assume that face images for training are captured under good lighting environments, and face images for testing are captured under various lighting environments, and propose a new approach as follows: (i) normalize the illumination components of face images using the Weber–face method; (ii) extract the features of the obtained images using a local directional pattern descriptor; and (iii) use support vector machines (SVM) for classification. The potential of the proposed approach is demonstrated by comparing a combination of illumination normalization methods (histogram equalization, Gradientfaces, and Weber–face), local descriptors (center–symmetric local binary pattern, local binary pattern, local phase quantization, local ternary pattern, and rotated local binary pattern) and the PCA method using nearest–neighbor and SVM classifiers. Experimental results for the extended Yale B face database indicate that the proposed approach achieves an accuracy of 0.12% to 4.26% higher than other methods using the same approach.