基于半耦合稀疏表达的极低分辨率人脸识别  

Very low resolution face recognition via semi-couple sparse representation

作　　者：杨威[1,2] 卢涛[1,2] 汪浩 

机构地区：[1]武汉工程大学计算机科学与工程学院,武汉430205 [2]智能机器人湖北省重点实验室(武汉工程大学),武汉430205

出　　处：《计算机工程与应用》2017年第22期169-175,共7页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61502354);国家留学基金委资助;湖北省自然科学基金(No.2012FFA099;No.2012FFA134;No.2013CF125;No.2014CFA130;No.2015CFB451);湖北省湖北省青年科技晨光计划;测绘地理信息公益性行业科研专项(No.201412014);武汉工程大学研究基金项目(No.K201403)

摘　　要：现有基于学习的人脸超分辨率算法假设高低分辨率特征具有流形一致性(耦合字典学习),然而低分辨率图像的降质过程使得高低分辨率特征产生了"一对多"的映射关系偏差,减少了极低分辨率图像特征的判决信息,降低了超分辨率重建图像的识别率。针对这一问题,引入了半耦合稀疏字典学习模型,松弛高低分辨率流形一致性假设,同时学习稀疏表达字典和稀疏表达系数之间的映射函数,提升高低分辨率判决特征的一致性,在此基础上,引入协同分类模型,实现半耦合特征的高效分类。实验表明:相比于传统稀疏表达分类算法,算法不仅提高了识别率,并且还大幅度降低了时间开销,验证了半耦合稀疏学习字典在人脸识别中的有效性。Existing Super Resolution（SR）algorithms are based on the assumption that High Resolution（HR）and Low Resolution（LR）features share the same intrinsic geometry structure（coupled dictionary learning）. However, the degradation process of LR image makes the features between HR and LR data produce a one to many mapping, which reduces the discriminative information of LR image and decreases the recognition rate of the super-resolution reconstruction image.To solve this problem, this paper introduces the semi-couple sparse dictionary learning model and loosens up the manifold consistency assumption, while learning a sparse dictionary and a mapping function between sparse coefficients, which improves the consistency of the feature between HR and LR data. On this basis, the introduction of Collaborative Representation Classification（CRC）model achieves efficient classification for semi-couple characteristics. Experimental results show that compared to traditional sparse expression classification algorithm, algorithm in this paper not only improves the recognition rate, and also greatly reduces the time overhead, which verifies the effectiveness of the semi-couple sparse dictionary learning in face recognition.

关 键 词：稀疏表达 半耦合 协同表达分类 极低分辨率 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]