Study On Characteristics Of Image Reconstruction Based On Bayesian Compressive Sensing

Traditional Nyquist sampling theorem (Nyquist) requires that the sampling rate must be no less than double of the maximum frequency of signal, it may require a higher sampling rate in practical engineering. In some signal and image processing, however, it is difficult and expensive to realize very high sampling rate in hardware. In addition, the processing efficiency is not high, and limited by the time-consuming acquisition and redundant storage information under the condition of great amount of data.Compressed sensing theory indicated that the signal can be sampled much less than the Nyquist criterion under the conditions that the preprocessed signal is compressible or sparse, observing it by the measurement matrix and reconstructing the signal by solving the optimization problem. It has been a hotspot and difficult research on how to discard the insignificant information and reconstruction processing directly on the portion of the feature of interest.Within the sparse Bayesian framework, a series of signal reconstruction were achieved by the maximum a posteriori estimation of sparse signal based on Relevance Vector Machine (RVM). The image was reconstructed in the processing of all measurement vectors at the same time through Bayesian compressed sensing algorithm, which improves the efficiency of the algorithm and accurately reconstructs the original signal or image. What’s more, it enhances some detail characteristics of the processed image.Main work included:(1) Combining Bayesian theory frame with the compressed sensing technology, the original image and the sparse processing of gray-scale image were reconstructed through the compressed sensing algorithm based on Bayesian hypothesis test theory.(2) The changes of peak signal to noise ratio (PSNR) value of the same image without reconstruction and with reconstruction were analyzed in different measurement values.(3) The changes of peak signal to noise ratio (PSNR) value of the image without reconstruction and with reconstruction were analyzed after imposing sparseness of the different gray-scale upon the same image in same measurement values. In addition, the influence factors of reconstruction performance were analyzed and the experimental conclusions were obtained.In this paper, according to the difference of image reconstruction performance in the sparse processing of the gray-scale images and in the condition of different measured values, such as the ultrasound, CT and optical images, the following conclusions were obtained by experimental methods.(1) The significant difference of image visual effects resulted on the different types of image processing based on the resolution of image without reconstruction and with reconstruction. The proposed algorithm can be outstandingly applied in image reconstruction. The changes of the ratio between gray-scale and sparsity of image and the measured value will have effects on the performance of the image reconstruction.(2) The image reconstruction performance did not always increase as the ratio between gray-scale and sparsity of image increase through the sparse processing of different gray-scale in the same measurement value. The local optimum area of the image reconstruction performance resulted from different sparsity according to the sparse processing of the gray-scale original image.(3) The observation image detail components and the peak signal to noise ratio (PSNR) value of the reconstruction image increase as the measurement value increases. Therefore, the Bayesian compressive sensing algorithm has a brilliant future in the practical application of image feature extraction.