基于分块压缩感知的遥感图像融合

Compressive fusion for remote-sensing images by block compressed sensing

对大数据量遥感图像融合,常规融合方法需考虑图像所有像素点,而全局压缩采样融合重构计算成本高、存储需求大。首先利用分块压缩感知(BCS)对输入图像进行压缩采样,再对压缩测量采用线性加权策略融合,最后采用迭代阈值投影(ITP)重构算法重构融合图像,并消除分块效应。提出了一种基于BCS的遥感图像融合方法,并给出其详细实现流程。仿真结果表明了ITP算法计算成本低、重构精度高。实际资料测试表明BCS融合方法与常规小波加权融合结果相比,除了平均梯度有所差别外,在平均值、标准差和信息熵等定量分析和视觉特征上基本相同。该算法用较少采样点实现有效压缩融合,存储需求小、重构成本低,融合决策过程简单,有利于大数据量遥感图像的融合。

For remote-sensing images fusion with huge amounts of pixel, the conventional fusion methods need all pixels du- ring fusion decision, and common compressive fusion with global sampling requires a large memory space and a high computa- tion cost for reconstruction. This paper made compressive measurements of input images by the block-based compressed sensing （BCS） and fused them with the rule of linear weighting, and then reconstructed the fused image through the iterative threshol- ding projection （ITP） with consideration of block artifacts. It proposed a compressive fusion strategy based on BCS sampling of remote sensing images with large amounts of data, while presented detailed implementation steps of the fusion algorithm. The emulation results show that the ITP achieves low computation cost and better recovery. The field tests show that, comparing with the result from wavelet fusion with a linear weighting rule, the image fused by the proposed method produces the similar visual features and nearly equal quantitative analysis in mean-value, standard variance, and information entropy, except a smaller value of average gradient. With very small number of sampling, the proposed fusion method achieves good fusion with small memory requirement, and fast reconstruction and simple fusion decision, which is very beneficial to the fusion of big re- mote-sensing images.