Compressed Image Sensing by Jointly Leveraging Multi-Scale Heterogeneous Priors for the Internet of Multimedia Things

Abstract

As an important branch of the Internet of Things, the Internet of Multimedia Things (IoMT) has complex application scenarios, thus putting forward new challenges to data communication. With the application of compressed sensing (CS)-based image compression and transmission in the field of IoMT, the traditional CS image reconstruction algorithms have been extended to model-based algorithms, which exploit more prior information of multimedia objects other than sparsity/compressibility in the reconstruction process. However, these algorithms mainly exploit just one type of prior information, still leaving room for further improvement. In this paper, we propose a novel CS image reconstruction algorithm by jointly leveraging multi-scale (local and global) heterogeneous (statistical and structural) priors of natural images, named jointly leveraging statistical and structural priors for CS image reconstruction (JLSSP-CS), to enable high-quality CS image recovery for IoMT. Specifically, the proposed JLSSP-CS algorithm is realized under the iterative hard thresholding framework, and the reconstruction process is composed of two phases. In the first phase, the local statistical correction and the global statistical correction are considered sequentially. Then, the global statistical and structural priors are exploited in nested iterations to further refine the recovery result in the second phase. The extensive simulations have been conducted, and the results indicate that the proposed JLSSP-CS algorithm outperforms the current state of the art by realizing high-quality image reconstruction with a small number of measurements for IoMT end devices.