Study On Low Complexity Recovery Algorithms Of Image Compressed Sensing Based On Statistics Prior

As a popular data traffic, image transmission has been progressively applied in various fields. However, a great challenge is posed in realizing high-efficiency transmission for high-definition image which results in large data volume. In order to improve the image reconstruction efficiency and reconstruction quality, an innovative image transmission and compression method should be designed.Benefiting from the sparse characteristics of transmission signal in different domain, Compressed Sensing(CS) theory provides an innovative solution for sampling data with extremely reduced sampling cost and transmission cost. As a result, CS can be utilized for image transmission, which has the sparse feature induced by the high correlation and redundancy of the contiguous natural image pixels. Moreover, the amplitude of the image in specific domain also presents specific energy distribution characteristics, which can be used as a statistical a priori to improve the performance of the image compression.In this paper, a method with two strategies for compressed image sensing based on the statistical prior information is proposed aiming at enhancing the transmission efficiency of image compression. To be specific, we first summarize the statistical information of the image in the wavelet domain used for constructing a statistical model and put forward the concept of “energy level”. Second, based on the statistical model, a low complexity compressed sensing method leveraging image statistical information is proposed. This method contains two reduction strategies: the one-time direct recovery(OTD) and the two-times iterative recovery(TTI). The recovery process is composed of two steps, where the row-wise(or column-wise) intermediates and column-wise(or row-wise) final results are reconstructed sequentially. In each step of the recovery, the reconstruction is constrained to conform to the statistical prior by introducing a weight matrix, and the complexity is as simple as the linear matrix multiplications. While for the TTI strategy, the weight matrix for the second step is iteratively refined to achieve more accurate recovery results. Finally, extensive simulation results exhibit that, compared to the traditional method, the proposed method boosts the performance of compressed sensing recovery. In particular, the recovery method with OTD strategy can achieve much faster recovery speed and better recovery quality. Meanwhile, the best recovery quality can be guaranteed with TTI recovery strategy at the expense of slight degradation in recovery speed, yet still faster than traditional methods.