The conventional signal acquisition systems must follow the Shannon-Nyquist sampling theorem: to avoid losing information when capturing a signal,the sampling rate must be at least two times larger than the signal bandwidth.With people's growing demand for information (especially for images andvideos), however, the sampling with the so-called Nyquist rate results in a datadeluge, which increases the burdens of signal processing, transmission andstorage and becomes a bottleneck for the digital systems. As an entirely newsignal processing theory, compressed sensing/compressive sampling (CS) hasgained a widespread interest in the signal processing community, which cansample and compress signals simultaneously at a sub-Nyquist rate, whileoffering a highly precise reconstruction from few measurements through anoptimization method. In this thesis, the research focuses on the applications ofCS theory in the image coding and distributed video coding (DVC), and themain works are as follows:1. A progressive image coding scheme based on an adaptive block-basedcompressed sensing (ABCS) is proposed, in which both image acquisition andreconstruction are carried out in two layers. At the base layer, an original imageis sampled and restored by the block-based compressed sensing (BCS) methodwith a low and fixed measurement rate. Second, all blocks in the enhancementlayer are re-sampled with different rates according to a block classification. Thefinal reconstruction of a block at the enhancement layer is performed in multiplestages where each stage only knows a part of sampled coefficients, and theoverall quality of the recovered image in each stage is improved successively.2. A directional block-based compressed sensing (DBCS) scheme isproposed for image coding. In this scheme, the directional information withinimage blocks is exploited by the DBCS method. The sampling of an image isdriven by a BCS method along with a scan mode following the dominatingdirection of each image block. At the decoder, each recovered image block isrearranged by the corresponding inverse-scan mode so as to obtain the original- order image. Due to the exploitation of the directional information within imageblocks, the DBCS method outperforms BCS method.3. An adaptive sparse basis (ASB) based distributed compressive videosensing (DCVS) scheme is proposed, which integrates the CS and DVC theory.The proposed scheme incorporates a low-complexity encoder and shifts mostcomputation burdens to the decoder-side. The video sequences are divided intoKey-frames and CS-frames. The Key-frames can be coded by the conventionalvideo compression standards such as MPEG/H.26x with intra-mode and eachKey-frame serves as a reference frame for its neighboring CS-frames. On theother hand, the CS-frames are coded using a CS approach. The sampling of aCS-frame is driven by BCS method independently; however, at the decoder,each block in a frame is recovered jointly using the state-of-the-art sparse basisASB, which generated by a few temporal neighboring blocks in previouslyreconstructed preceding and/or following Key-frames. Experimental resultsshow that the proposed scheme outperforms DCVS schemes with fixed sparsebasis. |