Multitask classification and reconstruction using extended Turbo approximate message passing. Issue 2 (February 2017)

• Record Type: Journal Article
• Authors: Wang, Ying-Gui; Yang, Le; Tang, Ze-Ying; Gao, Yong
• Is Part Of: Signal, image and video processing. Volume 11:Issue 2(2017)
• Page Start: 219
• Page End: 226
• Publication Date: 2017-02
• Subjects: Multitask CS -- Task classification -- Signal reconstruction -- Turbo generalized approximate message passing (Turbo-GAMP) -- State evolution (SE); Signal processing -- Digital techniques -- Periodicals; Image processing -- Digital techniques -- Periodicals; Digital video -- Periodicals; 621.3822
• DOI: 10.1007/s11760-016-0922-5
• Languages: English
• ISSNs: 1863-1703

Abstract

Abstract Approximate message passing (AMP) is a known compressive sensing (CS) algorithm, owing to it being computationally efficient, and having high performance and a deterministic state evolution (SE) trajectory. Turbo generalized AMP (Turbo-GAMP) was proposed based on AMP, and it was extended to multitask CS with multiple measurement vectors (MMVs). The resulting Turbo-GAMP-MMV can reconstruct multiple structured-sparse signals when they are well correlated. This paper considers the case where the CS tasks belong to various groups and signals from different groups may have weak correlation. We explore the SE property to enhance Turbo-GAMP-MMV for weakly correlated signals. The developed methods first conduct task classification via dividing CS tasks into groups and then reconstruct the original signals from each group jointly. Experiments using synthetic signals and grey images show that the new algorithms outperform several state-of-art benchmark techniques.