基于高斯加权和因子分析字典学习的人脸姿态估计  

Face poses estimation based on Gaussian weighted and factors analysis dictionary learning

作　　者：廖海斌 邓树文 王电化 范平 陈友斌 LIAO Haibin;DENG Shuwen;WANG Dianhua;FAN Ping;CHEN Youbin(School of Computer Science&Technology,Hubei University of Science&Technology,Xianning 437100,China;School of Automation,Huazhong University of Science&Technology,Wuhan 430074,China)

机构地区：[1]湖北科技学院计算机科学与技术学院,湖北咸宁437100 [2]华中科技大学自动化学院,武汉430074

出　　处：《扬州大学学报（自然科学版）》2018年第4期47-51,56,共6页Journal of Yangzhou University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61701174);湖北省自然科学基金资助项目(2017CFB300);湖北省教育厅科学技术研究资助项目(Q20172805);湖北科技学院工科硕士点建设专项科研资助项目(2018-19GZ050)

摘　　要：针对传统的非约束环境下人脸姿态估计方法无法在统一框架下很好地处理各种姿态相关和姿态无关因子等问题,设计了基于字典学习和稀疏表示的鲁棒性人脸姿态估计框架,提出一种新的基于鼻尖点高斯加权的人脸预处理方法.此外,为了提高字典的鉴别性,提出一种基于姿态相关和姿态不相关因子分析的鉴别字典学习算法.通过在公开的XJTU、Multi-PIE、CAS-PEAL-R1和AFLW人脸库实验,结果表明:该方法在具有光照、噪声和遮挡变化的人脸库上识别率均约达95%,基本可满足实际应用的要求.Accurate estimation of facial pose in an uncontrolled environment presents a great challenge.However,extant conventional approaches lack the capability to deal with multiple pose-related and-unrelated factors in a uniform way.This paper proposes a robust pose estimation framework based on dictionary-learning and sparse representation.With the guide of this framework,a novel face image pre-processing algorithm based on Gaussian weighted and tip of nose is designed to enhance pose-related factors.Further,a new insight on discriminative dictionary learning is also provided.Specifically,the author formulates the discrimination term based on pose-related and-unrelated factors analysis.Several experiments are performed on XJTU,CMU MultiPIE CAS-PEAL-R1 and AFLW databases.Recognition results show that the proposed method can achieve recognition rate about 95%under illumination,noises and occlusion variations.

关 键 词：人脸姿态估计 字典学习 稀疏表示 因子分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]