The Research And Improvement Based On The Active Appearance Model

The object positioning and feature extraction of images are important tasks in computer vision, and they play important roles in the image analysis and image processing. In all sorts of image positioning methods, the active appearance model (the Active the Appearance Model, referred to as the AAM) is an accurate and efficient model. This model was approached in1998by F.T.Cootes, from Carnegie Mellon University, who proposed and successfully applied to the human facial feature points location. Because of good scalability, fast processing and accurate feature location, AAM is considered as a prominent representative of facial feature points location, especially in the face image processing has been widely used feature point positioning. This article describes the process of AAM modeling, then presents an improved method for the loss of image information in the modeling process, finally uses AAM to the face pose estimation.First, the article follows up the study of the AAM fitting algorithm. The algorithm combines AAM with the Lucas-Kanade algorithm. On the one hand, by calculating some parameters used in the fitting process in advance, the program can reduce the amount of computation in the iterations. On the other hand, the effective use of AAM’s modeling capability and fast fitting of the L-K algorithm can help the program to achieve the target of facial feature points location and face recognition. Through experiments it can be found that the program will be more precise to the face image location.Secondly, with the consideration of preserving image information, the article comes up with the AAM fitting algorithm based on kernel methods. Put the kernel method of pattern recognition into the AAM modeling can still retain the high-dimensional information of the original images while doing the reductions on the shape and texture. After a series of experiments, it can be found that AAM fitting algorithm based on kernel method is more accurate on the face image location.Finally, in this part the AAM fitting algorithm applies to face pose estimation, that is ridge regression based on AAM. During the face pose estimation, the accuracy of the final experimental results will be affected by lighting or background, therefore the estimate of the target needs some image pre-processing. This algorithm can use the AAM on the normalized sample face images to eliminate some of the information has nothing to do with the attitude, such as the face of some rigid transformations, thereby improving the accuracy of head pose estimation.