Structure Information - Guided Compression Sensing Method And Algorithm Evaluation Software

Based on the sparsity of the signal, compressed sensing has been a popular topic in various scientific fields, because of obtaining signal information by using compressive sampling instead of the traditional method that both sampling and compression, which way breaking the traditional Nyquist sampling theorem, so that the sampling rate of the signal, the sampling time and storage costs are significantly reduced. Compressed sensing theory mainly involves three core issues:sparse representation of the signal, designing the measurement matrix and reconstruction algorithm, which have a direct bearing on the quality of the reconstructed signal.In this paper, as the starting point by the three core issues, based on researching the algorithm model and development status of compressed sensing, we analyze the compressed sensing reconstruction algorithms that have been proposed, and combine the inherent structural information of the image, so that improving the performance of reconstructed image. Meanwhile, we also studied the relative objective evaluation algorithm for the quality of the reconstructed image. The main work and achievements of this paper include:Firstly, we researched the method of designing random measurement matrix, and according to the traditional star sampling methods in the Fourier domain, presented methods of compression sampling which can be variable density, based on the image significant information in the Fourier domain, combining with the saliency map of image. This method can sample adaptively variable density for an image’s distinct regions in the Fourier domain. The experimental results show that under the same reconstruction algorithm, compared with the traditional star-sampling methods, the proposed variable density adaptive sampling method can adaptively adjust the sampling density according to the significant content of the image, with reducing sampling redundant in the non-significant area, and increasing the sampling density in the significant region, thereby improving the visual quality and objective evaluation index of the reconstructed image.Secondly, we proposed a method based on the image’s edge structural information which guided the compressive sensing reconstruction, because the Total Variation (TV) regularization priori in the some reconstructed algorithms, cannot effectively restore the image edge and texture information, and could easily lead to over-smoothing effect for the reconstructed image. In the process of the iterative solution, we assign different weights in the front of total variation regularization depending on the morphology of the front edge, after detecting the edges of the image by using of sophisticated edge detection method, so as to maintain the image edge information in the reconstruction process. The experimental results show that the way that we combine the edge information in the process of reconstructed image, can improve the effect of reconstructed image in the part of edge and texture, so that the structural similarity and PSNR of reconstructed image are higher than the image’s that reconstructed by other algorithm.Thirdly, according to the approach that guided the compressed sensing process by the edge information of image, we propose another compressive sensing method which based on the level set normal of priori image structural information. The algorithm is divided into two alternating steps:the first step is to estimate the normal vectors of the image level set curve; the other step is to reconstruct the compressive image by the estimated normal vectors. Through simulation experiments, we confirmed the rationality and validity of the model, but also it is beneficial to improve the effect of the reconstructing the part of directional structure in the image that using the normal information of image edge.Finally, in the MATLAB environment, this paper design a software about compressed sensing evaluation by using compression sampling rate, objective quality indicators, time of the reconstruction and so on. And the software included SSIM, PSNR, UQI (Universal Quality Index), VIF (Visual Information Fidelity) and other image quality assessment algorithm and integrate the classic compressed sensing algorithm and the new compressive sampling and compression algorithm that designed in this paper. The evaluation experiments confirm further that the approach of the structure of information guiding can help to improve the performance of compressed sensing algorithm.