结构化特征融合的主动表观模型

摘要

文中提出了一种基于结构化特征融合的主动表观模型(Active Appearance Model,AAM)匹配算法.为了增强主动表观模型匹配算法的泛化能力和精度,尤其是其对于人脸表观细节处的处理,本文将形状相关的局部特征描述子和全局纹理信息进行融合,并进一步将其应用到基于回归模型的主动表观模型匹配算法中.通过在XM2VTS和BioID人脸数据库上进行测试表明,本文算法比传统的主动表观模型匹配算法的误降低了15％左右.%The paper presents an improve Active Appearance Model (AAM) fitting algorithm based on structural feature fusion.In order to enhance the performance of AAM fitting algorithms in terms of generalization capacity and accuracy,especially for appearance details of human faces,the proposed fitting algorithm combines the shape-indexed local features and global texture information to form rich facial feature representations.Moreover,the new feature fusion method is used in regression-based AAM fitting.Experimental results obtain on the XM2VTS and BioID face datasets show that the proposed method reduces the fitting errors at around 15％ compared to the classical AAM fitting algorithm.