基于分块压缩感知的遥感图像多尺度融合

Multi-scale fusion for remote-sensing images by block-based compressed sensing

由于分块压缩采样缺乏全局特性,导致基于分块压缩感知的常规融合图像质量差、且存在分块效应。首先提出图像在小波变换域的非均匀分块压缩感知(WNBCS),然后对压缩采样结果用局部特征规则融合,再用迭代阈值投影算法重构融合图像和消除分块效应。最终提出基于分块压缩感知的遥感图像多尺度融合方法,并给出算法详细实现流程。仿真结果表明WNBCS改善了图像重构质量和速度。实际资料测试结果表明,局部特征压缩融合比最大值和加权融合结果具有更好的视觉效果和定量分析结果。所提融合方法考虑图像全局特性、简化融合决策过程,便于大数据量遥感图像的压缩融合。

Since compressive sampling of block-based compressed sensing （BCS） lacks global features, con- ventional fusion by BCS results in lower quality with blocking artifacts. Firstly the non-uniform BCS sampling of input images in the wavelet-transform domain is presented, and then these compressive samplings are fused by the rule of local-features. Finally, the fused image is reconstructed by iterative thresholding projection （ITP） algorithm with the consideration of blocking artifacts. A fusion method with wavelet-based non-uniform BCS sampling （WNBCS） is proposed for remote-sensing images, while the detailed implementation flows are given. Numerical experiment shows that ITP reconstruction with WNBCS can produce better recovery with low computational cost. Field test indicates that, image fused by the proposed method achieves better subjective visualization and quantitative analysis than those with rules of maximum-absolute-value and linear weighting. The proposed method simplifies the process of fusion decision with consideration of global features, which is advantage for the fusion of big images of remote sensing.