【摘要】 针对人脸识别中小样本数据集缺少分布信息难以获得鲁棒图像表示问题,提出深度子空间联合稀疏表示单样本人脸识别算法。首先,使用深度加权子空间构建抽象特征描述网络,获得单样本人脸深层抽象描述子。进而利用样本类间差异信息,引入邻域排斥度量学习实现低维度有鉴别力特征提取。最后基于协同表示分类器完成模式分类。在FERET、ORL、Multi_PIE等数据库上验证本文算法在单样本人脸识别问题上的有效性,鉴于深度子空间强大的特征描述能力,即使训练样本集很小,依然可以保证训练样本能够紧凑的表示有变化的测试样本。更多还原

【Abstract】 To solve the problem of not acquiring robust image representation with the small sample set lacking of distribution information,the single sample face recognition algorithm joined sparse presentation and deep subspace model is proposed. First,a deep feature extraction network with weighted subspace model is built to get the deep abstract representation of single sample face. Then,using intra-personal variations,neighborhood repulsed metric learning model could extract low dimension and more discriminating feature. Last,Collaborative Representation Classification( CRC) is used to classify. The effectiveness of proposed algorithm is varified on FERET,ORL,Multi_PIE and other databases,The results show that,because of the formidable feature extract ability,it is still possible to ensure that the training set could compact represents the test sample with variation,although training set is very small.更多还原