Image Reconstruction Algorithm Based On Block Compressed Sensing

Compressed sensing (CS) is a new technique for sampling and compression. It breaks through the Nyquist sampling for acquisition and made quite a stir in academic and industry circles. On this basis, Lu Gan proposed block compressed sensing, which divided the original image into small block, and each block is processed independently. Block CS is faster in sampling and reconstruction and lower in memory using. In this paper, we mainly do some research in reconstruction. Under the frame of block CS, We modify sections of spare representation and reconstruction and propose our reconstruction algorithm. The main contents of the thesis can be summarized as follows:(1)Block CS based on the wavelet transform is proposed. In order to get rid of blocking artifacts, we use wavelet transform instead of Discrete Cosine Transform (DCT), which is the image is divided into small blocks in time domain, and then each block is processed by wavelet transform. The experimental results show that there are not any blocking artifacts in the reconstruction image, and visual qualities are better than block CS based on the DCT. With our method, the step of processing blocking artifacts can be elided. So the negative effect which is brought by processing block artifacts can be avoided, and the time is saved.(2) An improved matching pursuit algorithm (MP) algorithm based on the edge detection is proposed. When reconstruct the image, we add in the edge detection technique. By the method of edge detection, we can make sure the position of the edge point approximately. And the purpose of that is coefficients located in the edge are mainly reconstructed, and the other coefficients are replaced by generalized inverse. In this way, the edge and the outline in the reconstruction image are clear. In our algorithm, we use different reconstructing methods to deal with the block which contains the edge and the block which contains no edge. And the purpose of this is increasing the reconstructing speed as much as possible on the premise of guaranteeing visual qualities of the reconstructing image. Because of mainly reconstructing the coefficients of the edge, the edge position of reconstructing images is clear. The experimental results show that the algorithm proposed in this paper is batter than MP and basis pursuit (BP) algorithm in time and visual qualities.(3) In this paper, we do a plenty of experiments to test and verify advancement of our algorithm. In the reconstruction of nature images, our algorithm shows enormous advancements no matter in time and visual qualities. In the reconstruction of SAR images, on the premise of similar reconstruction qualities, compared with other algorithm, our algorithm show a huge advancement in time, and with the increasing of images, this advancement will become more and more clearly.