Research On Compressed Sensing-based Inter-frame Coding For Digital Videos

With the rapid development of information techniques, the demand for information of people is increasing dramatically. The traditional temporal sampling theoremindicates, signals can be recovered without distortion on condition that the signals are sampled at a rate at least twice the highest frequency present in the signals. As the amount of information increases, signals will have wider bandwidth, which raises higher demand for sampling devices and makes it more difficult to process signals. In recent years, Tao, Candès, Donoho, etc have proposed the Compressed Sensing Theory (CS), which manages to compress signals based on the sparsity of signals by means of incoherent measurement with proper measurement matrices. Signals can be reconstructed with high probability from only a few observations with optimization algorithm.Compared with MPEG-2 predictive coding techniques, a compressed sensing-based inter-frame coding scheme for digital videos is proposed in this thesis. Combined with the traditional video coding scheme, motion evaluation and motion compensation are used to achieve inter-frame differences. Inter-frame signals are sampled and codedwith CS theory block by block based on the sparsity, and entropy coding are utilized for further compression.A series of video coding experiments are performed on standard test sequences. The influence on the coding performance is analyzed in terms of the selection of measurement matrices, reconstruction algorithm and parameters such as sparsification threshold, sampling ratio, size of blocks and so on. Meanwhile, the coding performance is compared with that of MPEG-2 Test Model 5 coder on condition of similar video quality.