Research Of Video Stream Compression Based On Multi-dimensional Compressed Sensing

With the development of mobile communications and the arrival of big data floods,as well as the Internet and the Internet of things are running through every corner of life,a new information age is opening.Because the video information has the advantage of vivid,visual,specific,completed and the large amount of information,people's demand for all kinds of video is growing fast.The video information of the proportion is over half in big data.Ensure that large amounts of video information,therefore,highly efficient,high quality,standardized compression are more and more important.Images and videos have a lot of redundancy,which provide the basis for modern signal compression algorithms.In traditional image and video acquisition process,the well-known Shannon theorem is applied to uniformly sample a large number of data above the Nyquist sampling rate.In order to facilitate efficient storage and transmission of images and videos,compression algorithms are applied to remove the redundancy and convert the large amount of raw data into a relatively small bit stream.However,this process wastes valuable resources.Compressed sensing is an innovative concept in signal processing,and provides a new scheme to collect data at a rate that is below the Nyquist sampling rate.Compressed sensing directly acquires signals in a compressed form if they are sparse in certain transform domains,and combines data sampling and compression into a single step.Natural images are good candidates for compressed sensing applications,and compressed sensing for image acquisition and reconstruction has been well studied.Compressed image sensing techniques explore the spatial redundancy within an image and can be easily extended to video applications by considering each frame in the sequence independently.However,this simple extension is essentially a 2-D method and fails to address the temporal redundancy in videos and ignores the temporal correlation.There are spatial redundancy and temporal redundancy in the video signal.According to this feature,the laboratory proposed the multi-dimensional vector matrix theory.The core idea of this theory is to extend the traditional two-dimensional matrix to multi-dimensional matrix,and establish a multi-dimensional vector matrix model.This model can take full account of the temporal redundancy of the video on the basis of spatial redundancy in the video processing.This theory can be applied in the video compressed sensing technology effectively,and can remove the redundancy in video multiple dimensions as well as preserve the spatial position information of the video.In this thesis,multi-dimensional DCT(Discrete cosine transform)transform and multi-dimensional Walsh transform are proposed based on the theory of multi-dimensional vector matrix.Combining it with compressed sensing theory,this thesis completes the compressed sensing process of video signal.In this thesis,the standard gray scale image is processed according to the compression sensing theory.Four reconstruction algorithms are used to reconstruct the image,and the performances of the four algorithms are compared.At the same time,according to the theory of multi-dimensional vector matrix,the two-dimensional compressed sensing model is extended to multi-dimensional,and a multi-dimensional compressed sensing model is established.This model is used to process the video.Firstly,the video is divided into blocks and reassembled,and the orthogonal transform is performed to make the video signal sparse in the corresponding transform domain.The last step is to measure and reconstruct.The evaluation standard of this thesis is the peak signal-to-noise ratio(PSNR)at different sampling rates.The video block size is 8×8×8.Comparing the performance of multi-dimensional DCT and multi-dimensional Walsh transform in compressed sensing theory? Our simulation results show that compared to K-L transform and wavelet transform,the results of the proposed method are far better than the former two.