Study On Dictionary Structure Methods Based On Compressive Video Sensing

With the development of wireless network, more and more mobile video terminalsappear in multimedia communication domain. But the traditional video compressionstandards could not be used in the above-mentioned application because of the encoderwhich always has high complexity and weak error resilience. Compressive sensing, anew decoding theory, brings hope to the video codec technology, and recentlycompressive video sensing has become a new research focus.As one of the key technologies in the compressive sensing, the dictionary structurehas great influences on the quality of signal reconstruction. Based on the introduction ofcompressive sensing theory, this paper firstly do an in depth analysis of the KSVDdictionary training method, and gives an initial dictionary structure method based on therelevant samples. this method select the sample signals which is the most relevant to thesignals to reconstruct as parts of the initial dictionary of the KSVD dictionary trainingmethod, and the selection of samples is upon a minimum measurements MSE standardswhich based on JL lemma, with the help of relevant samples it effectively improves thequality of the final training dictionary. Then a relevant sample weighted KSVDdictionary structure method is given, which improves the sparse representation ability ofthe dictionary by increasing the proportion of sparse representation error of the relevantsample. Finally, The above-mentioned dictionary structure method is applied to thedistributed compressive video sensing system.Experimental results show that, compared with the original KSVD dictionarytraining method, when the sample rate of key frames is0.5, the relevant sampleweighted KSVD dictionary structure method can effectively improve the objectiveperformance PSNR(Peak Signal to Noise Ratio) of the non-key frame by0.1to1.2dB.When the above-mentioned dictionary structure method applied in the distributedcompressive video sensing system, the reconstruction video quality is enhanced, eachframe needs to structure only one dictionary, and it is simple and easy to be realized.