Image Compressed Sensing Reconstruction by Collaborative Use of Statistical and Structural Priors

In this paper, we propose a novel compressed sensing (CS) algorithm by collaborative use of statistical and structural priors of natural images. The statistical priors include two aspects which are the statistical dependencies of wavelet coefficients in transform domain and non-local self- similarity among pixels in spatial domain. And the structural prior refers to the structural dependencies of wavelet coefficients in transform domain. Our algorithm which employs both multi- domain as well as multi-class prior information is realized under the framework of iterative hard thresholding (IHT). The reconstruction process is divided into two stages. In the first stage, the local statistical prior model is used to correct the signal estimation to obtain the preliminary estimation. In the second stage, first the non- local self-similarity model, and then the global structural prior model are employed to further refine the preliminary estimation. The results show that our algorithm outperforms the state of art. Our algorithm can be utilized in efficient communication in multimedia internet of vehicles (IoV). We demonstrate the effectiveness of our algorithm for multimedia IoV devices by showing its capacity in reducing the amount of multimedia data need to be transmitted while improving the recovery quality.