Face Feature AAM-based Positioning And Identification

Active Appearance Model (AAM), as an well-known and important feature extraction algorithm,was proposed by F. T. Cootes et al. in1998and has been employed to localization of facial feature. Due to its flexible structure and excellent performance, AAM has been extensively applied to face image processing, such as face detection, face recognition,face tracking and face animation.1)The paper studies a facial tracking and recognition algorithm based on AAM.Firstly, AAM’s modeling process and the matching process is described in details and make a really modeling process and obtain the matching results which confirms that the AAM feature extraction and matching calculation is accuracy Due to consuming too long by the gradient climb problems in AAM original matching algorithm,AAM matching process has been improved by Lucas-Kanade Inverse Compositional algorithm which make calculation simplified. Secondly,the tracking and positioning which combines AAM modeling and LK algorithm and applies to human faces completes the face modeling by capturing the first two frames of the tracking video as training set. Then, we update error parameters until convergence in the AAM matching calculation part.We follow the whole process recycled to the last frame of the video. The combined algorithm achieved good results in the tracking of the face in the video.2) We studies the combination of the AAM feature extraction and Support Vector Machine (SVM) which applies to face recognition. From the view of classification, We applied SVM to the AAM face matching and explore its practicability. Support Vector Machine proposed in the1990s is a highly effective data classification algorithm,and has been a rapid development because of its good fault tolerance and generalization ability of samples. First training image by the AAM feature extraction, and then use the support vector machine classifier. The experiments show that the combining face recognition of support vector machine and the AAM feature extraction is superior to the original AAM match.