On The Compressive Transmission Of Image Within The Compressive Sensing Framework

Multimedia and big data service will consume more and more data stream in future network. It is expected that the mobile data stream in2018will be increased twelve times compared to the total amount consumed in2012. Because of the explosive growth in the amount of data, it is imperative to reduce the content to be delivered without significant loss in the fidelity by fully exploiting the inherent correlation and redundancy of those multimedia traffic. And it turns out that the compressive transmission technique provides an effective candidate for the multimedia dissemination, and the related research efforts is of great theoretical and application importance. As a new but promissing technique, compressive sensing (CS) can realize almost accurate restoration from the measurements with sampling rate below the Nyquist sampling rate. Meanwhile, the CS itself has already comprised of both the measurement and the compression operation, the reduced measurement dimension makes it a viable compressive transmission technique. In this thesis, we focus on the image compressive transmission scheme to fullfill the high quality image transmission requirements in the future wireless and mobile network with tolerable implementation cost.Firstly, the low density framework based Bayesian compressive sensing is investigated, wherein the measurement and restoration can be equivalently transformed into the encoding and decoding steps of the LDPC code, and the a priori distribution of the sparse coefficients can be utilized to improve the restoration quality as well by following the Bayesian estimate mechanism. By introducing the Gaussian Scale Model (GSM) assumed in the traditional image processing to characterize the wavelet coefficient distribution, the thesis analyzes and compares different reconstruction algorithms by assuming Jeffreys and Laplace as the a-priori distribution, respectively. The analysis in this thesis shows that, the SuPrEM reconstruction algorithm can achieve better restoration performance by employing the GSM a-priori distribution model in the low density based compressive sensiong framework. Moreover, it is shown that the GSM model seems more suitable than the conventional Gaussian model in the low density based compressive restoration. Secondly, the Contourlet transform based compressive transmission of the image signal is studied in this thesis. By fully exploiting the orthogonality of high frequency sub-band components, the high frequency directional subband reassembling based compressive sensing algorithm is proposed, and it is shown that better signal reconstruction quality can be achieved than the wavelet based image transmission scheme. Finally, through altering the association method in the spatial directional tree generation, the wavelet-Contourlet transform based Set Partitioning in Hierachical Trees (SPIHT) image encoding scheme is investigated. It is validated through simulation that the wavelet-Contourlet transform based SPIHT image encoding scheme can achieve better image reconstruction performance than the conventional SPIHT scheme, especially when the code rate is low.