An efficient landmark localization for face occlusion

Landmark localization for facial representation is usually affected by the variation of illumination or occlusion, in order to reduce these factors, in this paper, we propose an efficient landmark localization method for face images based on modified active shape model (MASM) and Gabor filters to reduce the drawbacks of intensity contrast and occlusion. The approach mainly consists of two phases. First, face images are preprocessed by the proposed illumination normalization method using Gabor wavelets. Then, upgrading the performance of Gabor kernels reduction based on the similarity of Gabor features vector, the location of facial features can fit more efficient and fast by the proposed feature-based weighted warping. The advantages of this proposed method not only obtain the better face alignment, but also overcome the active shape model (ASM) method which caused the failure for aligned target. The experimental results verify that this approach can achieve in AR face database with face occlusion cases and Yale face database_B with varied illumination conditions.