Global texture constrained active shape models for image interpretation

In this paper, we propose to improve active shape models (ASMs) by global texture constraint for reliable image interpretation. In the proposed method, ASM search strategy is firstly used, then, in order to evaluate the fitting degree of the current shape to the interpreted image, warped global texture subspace reconstruction error is exploited. When the current shape is more fitted than last iteration ASM search strategy will continue. Otherwise, global texture is used to predict the shape model parameters to get more fitted shape then the previous iteration, a strategy similar with active appearance model. By such an interleave iteration, our method takes the advantages of ASMs while fully utilizing the global texture information for accurate image interpretation. Experiments on our database containing 300 labeled face images significantly show the effectiveness of our method.