Modeling And Matching Of Active Appearance Model

With the development of the information technology, people have no choice but to confront the explosion of knowledge. Therefore, the automatical information processing is gradually becoming the necessary requirement of our life. However, this is a difficult problem for computer vision and machine learning techniques. It is the active appearance model (AAM) that can help release the trouble. This model can be utilized to extract the shape and the texture of deformable objects in images and recover the image structure to understand the images. As a result, AAM has been widely applied to image and video processing.Theoretic analyses and solutions to several key problems existing in the AAM are given in this paper. These analyses mainly focus on the shape modeling, the texture modeling and the model fitting procedure. Novel modeling and fitting algorithms are proposed to improve the accuracy and robustness of AAM. The main achievements of this paper are as following:(1) Since intensity values used in standard AAM cannot provide enough information for image alignment, a texture representation method based on a bank of Gabor ?lters is proposed. Local structures of an image in different scales and directions can be extracted by Gabor filters. Given the problem of the excessive storage and computational complexity of the Gabor filters, three different Gabor-based image representations are used in AAM:? GaborD is the sum of Gabor ?lter responses over directions,? GaborS is the sum of Gabor ?lter responses over scales, and? GaborSD is the sum of Gabor ?lter responses over scales and directions. Through a large number of experiments, it is shown that the proposed Gabor representations lead to more accurate and robust matching between model and images.(2) To strengthen the robustness of AAM to illumination variations, a new texture representation, which combines Gabor phase filters and local binary patterns (LBP) operator, is presented. On the one hand, Gabor phase filters are robustness to illumination variations. On the other hand, LBP is able to efficiently encode local information and compress the redundancy in the Gabor filtered images. Experimental results on various datasets demonstrate the effectiveness of the proposed texture representation, which results in a more accurate and reliable matching. (3) To deal with the under-sample problem (USP) in AAM, this paper develops a tensor-based multivariate linear regression (TMLR) model to perform the regression task. The image texture is considered as a 2-d tensor, together with a tensor based regression model to model the model fitting procedure. In addition, an alternative iterative projection is deduced here to approximate the solution for TMLR. Experimental results show that the proposed TMLR based AAM performs better than traditional AAM in terms of accuracy and efficiency.(4) Since the high-dimensional texture has too much nonlinearity, a shape refinement method is proposed based on a complexity feedback framework. This method firstly defines the texture complexity to evaluate the shape. Then, a complexity feedback framework is introduced to refine the shape and reduce the complexity of the texture. Furthermore, a set of statistical measures are utilized to model the texture based on the Gaussian assuption. As a result, both the complexity and the computational cost of AAM are reduced. Experimental results show that the proposed method greatly improves the efficiency of model as well as the accuracy.(5) To improve the accuracy of the age estimation system, AAM is combined with multiple principal angle analysis (MPA) to fulfill the feature extraction. This paper first proposes MPA and the alternative solution to approximate the optimum. It is based on the idea to maximize the mutual information among multiple subspaces of the same class while minimize that of the different classes. Finally, the system utilizes AAM to model the facial feature and MPA to extract discriminant information for age estimation. Experimental results show that the proposed framework can effectively improve the accuracy of the age estimation system.