Signal Reconstruction Based On Block Compressed Sensing

Compressed sensing (CS) is a new area of signal processing for simultaneous signalsampling and compression. Since being put forward, CS has been concerned broadly byscholars both at home and abroad. According to the CS can’t sample in real time and have alarge amount of calculation, this text divide the image into small blocks, and make someimprovement on traditional blocking algorithm. So we can get higher precision for thereconstruction of the signal in a shorter time when we use the same amount measurements.The contributions of the thesis include:1. The CS method can compress and rebuild signal when the signal is sparse enough.Different blocks contain different texture, so some blocks contain much high frequency,and some blocks contain much low frequency. In this case different blocks have differentsparsity after wavelet transform. In this text we use block compressed sensing based onclassification. First, we divide the image blocks into flat blocks and non-flat blocksaccording to the amount of high frequency. And then non-flat blocks use moremeasurements and iterations, and flat ones use a little measurements and iterations.Finally, the false boundaries will be removed by an improved total-variation algorithm.The experiment proved that this method can have a good effect of image reconstruction,and the signal-to-noise ratio improved2.5dB～3dB.2. The image contains a low frequency sub-band and three high frequency sub-bands afterwavelet transform. Considering the low frequency sub-band has lower sparsity, we remainlow frequency sub-band coefficients, and processed the high frequency sub-bands only.But single method can’t make sure every block is sparse. According to this, the textproposes a block division method in wavelet domain. First, we divide the image afterwavelet transform into small blocks. According to the characteristic of different sub-bands,we use different methods to divide the wavelet sub-bands. And then the text use OMP torebuild the image. Experiment shows that our approach in the paper is effective.