基于改进主动外观模型的人脸识别算法研究

ON FACE RECOGNITION BASED ON IMPROVED ACTIVE APPEARANCE MODEL

提出一种基于改进主动外观模型AAM(Active appearance model)方法的人脸识别方法。先采用平移不变小波分解图像获得低频系数,把低频系数作为图像的纹理表示,能更充分表征人脸面部的纹理特征。然后采用增量子空间学习算法更新训练样本的特征空间,通过实时对模型的更新和学习,实时更新特征空间更有效地描述样本图像间的相似性或差异性。最后把提取的面部特征点信息作为每个人脸的特征向量,根据最近邻分类器进行人脸识别验证。实验结果证明了该改进方法的有效性。

A face recognition method based on improved active appear model is proposed. In the method first the translation invariant wavelet is performed to decompose images to obtain low frequency coefficients,and then these coefficients are used as images＇ texture representation,which can express more enough information of texture features on human face. Next,the incremental subspace learning algorithm is used to update the feature space of training sample,which timely updates the feature space and more efficiently describes the similarity or difference among sample images through timely updating and learning the model. Finally,the extracted facial features are used as the eigenvector of each face,according to the nearest neighbour classifier the face recognition and verification are done. Experimental results demonstrate the effectiveness of the improved method.