Abstract: Recently, representation-based classifications, such as Sparse Representation Classification （SRC）, Collaborative Representation Classification （CRC）, etc. have attracted extensive attention. These methods use a single picture to recognize the testing subject but ignore the relationship among a collection of pictures which lead to inadequate details. To take advantage of the correlation of multiple images, this paper presents a collective representation classification in face recognition. The new method uses a matrix to represent all the testing images then the most compact representation error is evaluated for classification. Multi-representation helps to take account of the similarity and difference hidden in the testing image-set. Especially when the image set includes more side-face images than frontal ones, Joint-collective Representation Classification （JRC） outperforms the state-of-the-art method which is also validated by the practical experiments in two public databases.

Key words: joint-collective representation, face recognition, sparse optimization

摘要： 近年来，基于表示的人脸图像识别方法吸引了众多学者的关注，如稀疏表示分类方法（Sparse Representation based Classification，SRC）、协作表示方法（Collaborative Representation based Classification，CRC）等。这些方法均利用单张图像的表示信息进行识别，而忽略了集体图像之间的关联性，容易存在信息不足的缺陷。为了能够充分利用多张人脸图像的相互关系，提出了一类集体表示分类方法。该方法将多张待识别图像映射为一个稀疏表示矩阵，并对每类测试图像集体重构，以最小残差为准则对每类人脸图像集分类。这种方法通过同时表示多张图像，关注到不同图像之间的相似与不同，获取到同一主体的更多信息，从而提高识别正确率。尤其在只有多张侧脸图像而无正脸图像的情况下，集体表示分类方法更能发挥优势，在两个公开人脸图像数据集上的实验结果也验证了该方法的有效性。

关键词: 集体表示, 人脸识别, 稀疏优化