独立子空间下的草图人脸合成与识别

基于草图的人脸识别对于公安机关通过目击者描述出的犯罪嫌疑人草图来确定疑犯的身份具有重要的意义.为实现草图人脸识别,从人脸独立特征子空间和人脸形状结构特征出发,提出了一种新的草图人脸合成和识别方法.根据人脸图像在独立分量子空间上的重建,研究了基于独立成分分析(ICA)的草图人脸合成,实现了照片草图人脸转换;针对人脸结构信息在人脸识别中的重要作用,在分析主动形状模型(ASM)算法在提取人脸结构信息上所具有的局限性的基础上,提出了极坐标形状模型(PSM);最后,将合成的草图与待识别草图利用ICA/线性鉴别分别(LDA)和极坐标形状模型进行联合识别.实验结果表明,提出的算法可以有效地将照片转换为草图,并具有较好的识别率,1阶识别率达到94.7%,10阶达到99.1%,将子空间投影与极坐标形状模型联合起来识别,1阶识别率可提高5.3%,前3阶识别率平均提高约4.2%.算法较好的识别率基本上满足了自动草图识别系统的要求.

Face sketch synthesis and recognition based on independent subspace

Face sketch recognition plays an important role in determining the identity of the suspect through suspect sketch which is described by witnesses.In order to better realize face sketch recognition,a new face sketch synthesis and recognition algorithm was proposed based on face independent subspace and shape features.According to the face reconstruction based on independent component analysis(ICA) subspace,the face sketch synthesis based on ICA was studied,and the conversion between photo and sketch was realized.Considering the important role of the face structure information in face recognition and in analyzing the limitation of the structural information extracted by active shape model(ASM) algorithm,the polar shape model(PSM) was proposed.Finally,the synthesised sketch was recognized with the one to be recognized by combining ICA/linear discriminant analysis(LDA) and PSM.Experiment results indicate that the synthesis sketch performs very well by new algorithm,the recognition rate of one rank is 94.7%,and the recognition rate of ten rank is 99.1%.Recognition with subspace and PSM,one rank rate can be increased by 5.3%,and the first three rank rate can be increased by 4.2% evenly.The algorithm has higher recognition rate and basically meets the requirements of automatic face sketch recognition system.