Illumination Removal via Gaussian Difference L0 Norm Model for Facial Experssion Recognition

Face images in the logarithmic space can be considered as a sum of texture component and lighting map according to Lambert Reflection. However, it is still not easy separate these two parts, because face contour boundaries change are difficult distinguish. In order enhance separation quality this paper proposes an illumination standardization algorithm based on extreme L0 Gaussian difference regularization constraints, assuming that massively spread all over image but simple, regular, sparse enough. The proposed uses iterative smoothing method, which achieves more accurate estimation by reserving fewest boundaries. Thus, original restored better simply subtracting estimated. experiments organized with steps: first step observe restoration, second test effectiveness our for complex classification tasks. We choose facial expression step. experimental results show effectively recover details from extremely dark or light regions. experiment, we use CNN classifier emotion accuracy, making comparison removal state-of-the-art preprocessing methods. works best at about 5 7 percent accuracy higher than other algorithms. Therefore, proven provide effective processing technical support problems require high degree preservation texture. contribution is, first, enhanced TV model boundary constraint estimation. Second, response formulated difference, strongly responds Third, emphasizes necessity images.