Illumination Compensation for Face Recognition under Uncontrolled Lighting Conditions using Wavelet Tensor-based Reconstruction

Uncontrolled illumination conditions for capture of human face images significantly affect the overall performance of face recognition systems in many real-world applications. To overcome this challenge, we propose a novel algorithm called Wavelet Tensor-based Illumination Compensation (WTIC) to enhance the quality of face images captured under uncontrolled lighting conditions. Firstly, we establish a third-order tensor using twelve wavelet subbands obtained from the face image. We propose using high-order singular value decomposition to adjust the core tensor to compensate for the illumination intensity of the face image with adaptive weight coefficients corresponding to twelve frontal slices of the core tensor. The skin color of the image is then enhanced by fixing the obtained third inverse factor with a proposed designated inverse factor. Experiments conducted on four widely used databases (CMU-PIE, FERET, and FEI) reveal that the proposed algorithm not only yields clearer, smoother, and more natural face images but also considerably enhances the accuracy of face recognition systems.