采用ASPCM模型进行姿势鲁棒性人脸识别

Pose robust face recognition based on ASPCM model

提出ASPCM模型,并将其用于不同姿势下的人脸识别。对人脸图像的形状表示和纹理表示进行主成分分析,建立形状模型和纹理模型;以形状参数、纹理参数和姿势参数间的转换确定人脸图像与头部角度间的映射关系;使用精确性和概括性两个标准衡量ASPCM模型的分解性能和合成性能;根据平均纹理相似度判断输入图像与模型视图间的相似程度。实验表明,该模型分解性能的精确性误差和概括性误差均在1.85°以内;合成性能的这两种误差均在1.1个像素以内;精确性和概括性的平均纹理相似度均在95.8%以上;当头部转动角度在25°以内时,该模型的识别率达到100%。

A novel model named Analysis Synthesis Principal proposed and applied to face recognition with various poses. Component Mapping（ASPCM） is Shape representation and texture representation are subjected to principal component analysis, resulting in shape model and texture model respectively. Relationship between facial image and 3D head angles is obtained from the transformation rules of shape, texture and pose parameters. Accuracy and generalization areused to gauge analysis and synthesis abilities of ASPCM model. Average texture similarity is constructed to gauge degree of similarity between model views and target images. The experiments show that accuracy and generalization of analysis in average angular error are both below 1.85°. Accuracy and generalization of synthesis in average position error are both below 1.1 pixels. The two average texture similarities are both beyond 95.8%. Recognition rate of ASPCM model reaches 100 % while head pose range is below 25°.