基于编码复杂度的混合结构稀疏人脸识别方法  

Face Recognition Method of Mixed Structured Sparsity Based on Coding Complexity

作　　者：蔡体健[1,2] 樊晓平[2,3] 谢昕[2] 徐君[2] 

机构地区：[1]中南大学信息科学与工程学院,长沙410012 [2]华东交通大学信息工程学院,南昌330013 [3]湖南财政经济学院网络化系统研究所,长沙410205

出　　处：《模式识别与人工智能》2015年第7期595-602,共8页Pattern Recognition and Artificial Intelligence

基　　金：国家科技支撑项目(No.2012BAH08B01);教育部人文社会科学研究青年基金项目(No.14YJCZH172);江西省自然科学基金项目(No.20142BAB207007);江西省科技厅工业支撑重点项目(No.20151BBE50055);华东交通大学科研基金项目(No.09111004)资助

摘　　要：利用编码复杂度表示数据的结构稀疏度,通过降低编码复杂度实现结构稀疏.在稀疏表示分类模型的基础上,通过聚类排序的方法构造结构化字典,形成混合结构稀疏模型.此模型结合类间样本的定长组结构与类内样本的动态可重叠组结构,以及误差的标准稀疏结构.为实现混合结构稀疏重构,提出改进的混合结构贪婪算法.实验表明对数据字典进行聚类排序可有效改进人脸的识别性能,在相同条件下,混合结构的性能优于其他结构,文中算法也优于其他算法.Coding complexity is utilized to represent the structural sparsity, and structural sparsity is achieved by means of reducing coding complexity. Based on the model of sparse representation classification, a structural dictionary is formed from clustering and sorting, sparsity model with mixed structure is constructed. This model combines fixed-length group structure between classes, and dynamic group structure within classes, as well as standard spare structure corresponding to error part. To reconstitute this mixed structural sparsity, an improved mixed structural greedy algorithm is proposed. Experimental results show that the clustering and sorting of the data dictionary can effectively improve the performance of face recognition. Under the same conditions, the performance of mixed structure is better than otherstructures, and the proposed algorithm outperforms other algorithms.

关 键 词：人脸识别 压缩感知 结构稀疏 编码复杂度 结构贪婪算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]