Research And Application Of Dictionary Learning Algorithm In Compressive Sensing

Compressive sensing theory(CS) is a signal compression coding theory which breaks through the limit of Nyquist sampling theorem, and can reconstruct original signal accurately by using a random sampling approach with fewer sampling data. Sparse representation is the basis and prerequisite of CS, so how to find the right sparse dictionary to achieve optimal sparse representation of signals becomes a major research objective in the field of CS. Among all kinds of sparse dictionary, dictionary learning algorithm has the best performance. The sparse dictionary generated by dictionary learning algorithm is an adaptive one, which can get rid of the fixed structure and make the dictionary's atomic scale and features closer to the original signal itself. But it takes too long to generate the dictionary. So increasing the speed of the dictionary learning algorithm has a very big significance.In recent years, video codec in wireless multimedia sensor networks(WMSNs) gains more and more attention. Research in this field focuses on two issues:(1) how to reduce the complexity of the encoder;(2) how to improve the resiliency to channel errors. Both above issues can be solved by using CS in this field. And dictionary learning algorithm can improve the accuracy of reconstructed video. So using CS and dictionary learning algorithm in video codec has a great prospect.This dissertation includes three research contents:(1) we analyze various types of sparse dictionaries and using them in CS. We make a detailed comparison of these sparse dictionaries' characteristics through experiments, including their structure, sparse representation ability and quality of reconstructed image;(2) in order to solve the problem of taking too long to generate the dictionary, we propose an improved dictionary learning algorithm(IK-SVD) which makes big improvements both in the sparse representation stage and dictionary update stage of algorithm to reduce the time loss. Experiment results show that IK-SVD algorithm reduces the time of generating dictionary by 1/3 compared with traditional algorithm; and(3) for the limitations of traditional coding scheme in WMSNs above-mentioned, we propose a dictionary learning-based compressed video sensing codec model. The model uses CS to reduce the complexity of encoder effectively and improve resiliency to channel errors. And in the decoder, dictionary learning algorithm helps enhance images' sparse representation, thereby improve reconstructed video quality. This model switches the computational complexity from encoder to decoder and has high coding efficiency, so it can be applied to the resource-constrained embedded devices in WMSNs. Theory analysis and experiment results have verified the feasibility and efficiency of this model.