• 学术论文 •

### 基于稀疏补分析模型的近似最优子空间追踪

1. 1. 东莞理工学院电子工程学院, 广东东莞 523808;
2. 东莞理工学院网络中心, 广东东莞 523808;
3. 东莞理工学院计算机学院, 广东东莞 523808;
4. 汕头大学电子工程系, 广东汕头 515063
• 收稿日期:2013-01-18 修回日期:2016-03-04 出版日期:2016-10-25 发布日期:2016-10-25
• 作者简介:张宗念,男,1963年6月生于河北深州,副教授.2000年获华南理工大学博士学位.主要研究方向为压缩感知理论与应用、图像处理.E-mail:zzn99@sohu.com;李金徽,男,1980年4月生于辽宁沈阳,工程师.2004年获重庆工商大学学士学位.主要研究方向为分布式计算机网络.E-mail:li@dgut.edu.cn;黄仁泰,男,1964年12月生于广东东莞,副教授.2006年获华中科技大学硕士学位.主要研究方向为分布式计算机网络.E-mail:huangrt@dgut.edu.cn;闫敬文,男,1964年7月生于吉林磐石,博士,教授,博士生导师.1997获中国科学院长春光机所博士学位.主要研究方向为图像处理和分析、遥感图像处理等.E-mail:jwyan@stu.edu.cn
• 基金资助:

国家自然科学基金（No.40971206）；广东省自然科学基金（No.2015A030313654）

### Approximately Optimal Subspace Pursuit Based on Cosparse Analysis Model

ZHANG Zong-nian1, LI Jin-hui2, HUANG Ren-tai3, YAN Jing-wen4

1. 1. Department of Electronic Engineering, Dongguan University of Technology, Dongguan, Guangdong 523808, China;
2. Network Center, Dongguan University of Technology, Dongguan, Guangdong 523808, China;
3. Department of Computer Science, Dongguan University of Technology, Dongguan, Guangdong 523808, China;
4. Department of Electronics Engineering, Shantou University, Shantou, Guangdong 515063, China
• Received:2013-01-18 Revised:2016-03-04 Online:2016-10-25 Published:2016-10-25

Abstract:

An approximately optimal subspace pursuit algorithm under cosparse analysis model was studied to reconstruct the original signal from the noisy measurement vectors.To overcome the drawbacks of the non steepest gradient during the pursuit process and the low successful reconstruction probability for sparse synthesis model,an approximately optimal subspace pursuit algorithm based on cosparse analysis model was presented and realized.The approximately optimal optimization object function for the algorithm was designed according to the structure of the different analysis dictionaries,the iterative pursuit process of the algorithm was revised,and the methods of selecting cosparsity was optimized.The simulation experiments show that the complete reconstruction probability of the new algorithm is evidently larger than that of the algorithm for ASP,AHTP,AIHT,AL1 and GAP when the cosparse operator is a random compact frame or a two dimension total variant matrix.The comprehensive average PSNR of the output signal for the new algorithm is larger than that of the algorithm of ASP,AHTP,and AIHT for 0.8dB,1.38dB and 3.13 dB respectively and is less than that of the algorithm of GAP and AL1 for 0.32 dB and 0.6dB when the input signal is with Gaussion noise.The complete reconstruction probability of the new algorithm was greatly improved by adopting the above measures,and the convergence condition for the new algorithm was simplified.