图像集闭包建模的协同表示人脸识别算法

针对样本集距离分类算法忽略样本集内部变化的不足，利用图像多重描述提供的互补信息，提出图像集闭包建模的协同表示人脸识别算法。首先，扩展具有多重描述能力的图像集，图像的中等强度像素携带鉴别信息利用原始图像生成中等像素图像，镜像图像可增添图像细节信息利用原始图像产生镜像图像，将此两种源域图像与原始图像联合构成扩展的图像集。然后，以无参建模构建扩展的图像集为字典闭包，同类异源域的测试图像构成图像集且构建为测试闭包，借鉴协同表示思想利用字典学习迭代求解闭包系数。最后，采用残差判别函数进行模式分类。本文方法不仅构建具有多重描述能力的图像集，而且充分利用样本集内部关联性从而获得较好的分类结果。本文分别在ORL、GT（Georgia Tech Face Database）、CMU PIE人脸数据库上进行实验。 

Hull of Image Set based Collaborative Representation for Face Recognition

The set-to-set distance based methods ignores the relationship between gallery sets, while representing the query set images individually over the gallery sets ignores the correlation between query set images. In view of multiple representations of images contributing to providing complementary information, hull of image set based collaborative representation for face recognition is proposed. Firstly, the extended image set with multiple representations is structured. Due to the images with pixels with moderate intensities of the original images carry discriminatory information and the mirror images can somewhat overcome the misalignment problem of the face image in face recognition， the extended image set can be obtained by jointing the domain images of the original images and mirror images and pixels with moderate intensities images. Secondly, the extended dictionary is modeled as hull dictionary with non-parametric approaches for image set modeling and the query set from the same class of different domain image sets is modeled as a hull. The idea of collaborative representation and iterations are used to solve the coefficient of hull. Finally, the query set is classified by using of residual error SRC criterion. This method not only structures the image set with multiple representations contributing to the accuracy of image classification, but also makes full use of the relationship between image sets. Experimental results verify the proposed algorithm effectiveness respectively in ORL、GT（Georgia Tech Face Database）、CMU PIE facial database.