摘要
针对人脸识别中的姿态变化问题,提出了子区域关联映射的方法识别多姿态的人脸图像。人脸被分割为若干子区域,姿态变化对图像的影响被分解为关联子区域的形状映射与纹理映射。提出了2维耦合成分分析的方法构造关联子区域的映射关系。2维耦合成分分析采用2维矩阵方式直接表达人脸图像,在此基础上获取不同观测空间上的低维耦合空间,根据局部几何关系不变性的原理学习耦合空间上投影特征矩阵之间的非线性映射。在应用贝叶斯框架评估子区域可分性的基础上,综合全体子区域的信息给出最终的判别结果。比较实验结果表明,关联子区域映射方法能有效补偿姿态变化带来的影响;对应的多姿态识别方法判别率高,对姿态变化敏感度低。
As for the pose variation in face recognition, a correlative sub-region mapping method was proposed to recognize the multi-pose face images. The face was divided into several sub-regions, and the influence to image caused by the pose variation was decomposed into shape mapping and texture mapping between correlative sub-regions. A new technique coined two-dimensional coupled component analysis（2D CCA） was developed to construct these mapping functions. Based on 2D image matrices rather than 1D vectors, 2D CCA obtained the low-dimensional coupled spaces embedded in two observation spaces, and learned the nonlinear mapping relation between the projected feature matrices according to the principle of local geometry preserving. The discriminative power of each sub-region was estimated using Bayesian framework, and thereafter the final recognition result was obtained by combining the measures in all sub-regions. The results of comparing experiment show that, correlative sub-region mapping method compensates the pose variation effectively, and the corresponding muhipose recognition method is of high recognition rate and low sensitivity to pose variation.
出处
《中国图象图形学报》
CSCD
北大核心
2007年第7期1254-1260,共7页
Journal of Image and Graphics
关键词
多姿态人脸识别
关联子区域映射
2维耦合成分分析
姿态补偿
multi-pose face recognition, correlative sub-region mapping, two-dimensional coupled component analysis, pose compensation
