Research On Block-based Compressed Sampling Method

With the rapid development of information technology,the demands for information are increasing.How to store,process and transmit large amount of data is an important subject to be studied.Compressed sensing breaks through the bottleneck of the Shannon sampling theorem,which provides a new method for information acquisition,and therefore,the compressed sensing theory has attracted great attention in many application fields.Compressive sensing has made some progress,but due to the complexity of the modern signal,and the compressive sensing reconstruction algorithm,the original compressed sensing algorithm has a lot of disadvantages in the practical application.Because the signal is large,the measurement matrix becomes large,resulting in a large amount of memory consumption.In the reconstruction process,due to the large amount of measurement signal,the reconstruction algorithm has a large amount of computation,which makes the running time be too long.This paper focuses on the application of block compressed sensing method in different fields.In the image processing field,we propose a novel sampling method based on clustering knowledge,which optimize the sampling process and improve reconstruction quality.In the field of sensor networks,the idea of block compressed sensing is jointed with the random walk mechanism.We provide random walk with a good start,which optimizes the sampling process and result in a better reconstruction quality.The main work is as follows:(1)Summarizing compressed sensing theory and pointing out the shortcomings of the traditional compressed sensing.Introducing some block compressed sensing method and presenting some representative block compressed sensing method.Also advantages as well as the universal existence of some issues of block compressed sensing is summarized.(2)Proposing saliency-based clustering block compressive sampling for image signals.According to the saliency of the image signal,the K-means clustering algorithm is used to gather rows and columns in sub-block to a block that they are similar to.Sampling rate is assigned based on the saliency of sub-blocks that high sampling rate is assigned to saliency regions while background regions get low.The proposed method optimizes the allocation of sampling rate to a signal and experimental results demonstrate its better performance in PSNR and MSSIM and the visual effect.(3)Proposing a novel scheme of random walk in wireless sensor network.The scheme first obtains the feature,sparsity or entropy,of data in network.Then it divides network to serial blocks and calculate their feature weight.The number of starting point in per block is allocated based on its weight.This scheme optimizes the start of random walk,which results in better reconstruction quality and results in a more stable reconstruction.