Enhanced Active Shape Models with Global Texture Constraints for Image Analysis

Shiguang Shan, Wen Gao, Wei Wang, Debin Zhao, Baocai Yin

Abstract: Active Shape Model (ASM) has been widely recognized as one of the best methods for image understanding. In this paper, we propose to enhance ASMs by introducing global texture constraints expressed by its reconstruction residual in the texture subspace. In the proposed method, each landmark is firstly matched by its local profile in its current neighborhood, and the overall configure of all the landmarks is re-shaped by the statistical shape constraint as in the ASMs. Then, the global texture is warped out from the original image according to the current shape model, and its reconstruction residual from the pre-trained texture subspace is further exploited to evaluate the fitting degree of the current shape model to the novel image. Also, the texture is exploited to predict and update the shape model parameters before we turn to the next iterative local matching for each landmark. Our experiments on the facial feature analysis have shown the effectiveness of the proposed method.