Research Of Image Reconstruction Algorithms Based On Block Compressed Sensing

Compressed Sensing(CS) theory breaks the limitation of traditional Nyquist sampling theory, as it makes it possible that signal sampling and compression at the same time and the sampling rate is much more less than Nyquist sampling rate. Block Compressed Sensing(BCS) further notes that image can be divided to a number of image blocks with the same size and each image block can be measured and reconstructed independently, through which speeding up the speed of sampling and reconstruction. The thesis studies image sparse and reconstruction methods based on BCS and mainly does the research in the following aspects:(1) The thesis achieves the reconstruction of one-dimensional signal and two-dimensional image using Orthogonal Matching Pursuit(OMP) algorithm based on CS theory. The thesis also puts forward a image block reconstruction method based on discrete cosine transform(DCT) to address major issues of direct CS reconstruction in image reconstruction, such as long time and large storage space of matrix in the reconstruction process. The method reduces the image reconstruction time and improves the reconstruction quality compared with the direct CS method.(2) The thesis puts forward a method to classify and handle image block based on the texture characteristics, which can select the observed value in a self-adaption way depending on the sparsity. And this method acquires the reconstruction image with high quality, reducing the observed value at the same time. Then the thesis handles the fake boundaries which are derived from the adjacent of texture blocks and smooth blocks while processing the reconstruction deploying the mean filter method. Finally, an improved total variation method was proposed to adjust the reconstruction image and the method can improve the PSNR and reduce the blocking effect.(3) The image block reconstruction of wavelet transform can reserve the coefficient of low frequency in the image and only measure the high frequency. The thesis realize the image block reconstruction based on discrete wavelet transform(DWT) by using OMP algorithm. Based on the previous, an method was proposed to process 3 high frequency subbands with different observed value. Experimental result revels that this method saves the image reconstruction time.