利用边缘信息的多尺度分块压缩感知自适应采样方法

在小波域多尺度压缩感知框架下，被完整保留的低频系数存在着许多可利用的图像信息。本文在分析了不同尺度之间、以及同一尺度之内的系数块存在能量差异的基础上，提出了利用边缘信息的多尺度分块压缩感知自适应采样方法(EAS)。该方法首先利用低频系数提取出边缘信息，然后将边缘信息分块，加权计算每个块的边缘信息度，根据边缘信息度判断每个系数块的能量大小，将其转换成每个子块的自适应采样率，从而实现多尺度分块压缩感知的自适应采样。采用医学图像，含有复杂纹理的自然图像和含有严重噪声的SAR图像三类测试数据，验证了EAS方法的性能。数值实验结果表明，EAS方法对不同的压缩感知算法均有很大的提升，能够显著提高图像的重构质量和视觉效果。 

Edge-based Adaptive Sampling Method for Multiscale Block Compressed Sensing

Under the framework of multiscale compressed sensing in wavelet domain, the low frequency part of wavelet decomposition coefficients that contained much useful image information were completely preserved. Based on the analysis of different coefficient sub-blocks in the inter-scale and intra-scale had varied energy content, we proposed an edge-based adaptive sampling method for multiscale block compressed sensing. Firstly, we extracted edge information from low frequency coefficients, and then got the edge information value by the way of using weighted calculation on the each edge information block. Lastly, we transformed the edge information value into adaptive sample rate of each coefficient sub-blocks. The method achieved the more efficient multiscale block adaptive sampling for the compressed sensing of images. The performance of our method was verified on the three types test images: the sparse medical images, the natural images with complex texture and the SAR images with many speckle noise. The results show that it can significantly improve several compressed sensing algorithms on both the reconstructed image quality and the visual effect.