基于联合卷积分析与合成稀疏表示的遥感图像分量替换融合方法

针对遥感图像融合中传统分量替换方法光谱失真严重问题，提出了一种基于联合卷积分析与合成稀疏表示的改进分量替换融合方法。与传统分量替换方法不同，该方法旨在改进融合过程中空间细节信息提取和注入策略，以生成具有更高光谱与空间质量的遥感图像。首先利用联合卷积分析与合成稀疏表示算法分别对强度分量和直方图匹配后的全色图像进行分解，获取二者的基础层和细节层；其次分别采用平均融合策略与最大值选择策略对基础层和细节层进行融合，将融合后的基础层和细节层求和并与强度分量相减以获取空间细节信息；然后将得到的空间细节信息与传统分量替换方法获取的空间细节信息进行加权平均；最后将最优空间细节注入到上采样多光谱图像中以获得最终融合图像。实验结果表明，与另外7种融合方法相比，所提方法得到的融合图像具有较低的光谱失真度和较高的空间分辨率，其光谱失真指标、空间失真指标、无参考质量评价指标在真实数据上明显优于对比方法。 

Joint Convolutional Analysis and Synthesis Sparse Representation-Based Component Substitution Fusion Method for Remote Sensing Images

Aiming at the spectral distortion problem of traditional component substitution-based methods in remote sensing image fusion， an improved component substitution fusion method based on joint convolutional analysis and synthesis sparse representation is proposed. Different from the traditional component substitution methods， this method aims to improve the spatial detail extraction and detail injection strategy in the fusion process to generate remote sensing images with higher spectral and spatial qualities. Firstly， the joint convolutional analysis and synthesis sparse representation algorithm is used to decompose the intensity component and the histogram-matched panchromatic image to obtain the base layer and the detail layer， respectively. Secondly， the average fusion strategy and the choose-max fusion strategy are respectively applied to fuse the base layers and the detail layers， and the spatial detail image is obtained by subtracting the intensity component from the sum of the fused base layer and the fused detail layer. Then， the weighting average is employed to the obtained spatial detail image and the detail image obtained by the traditional component substitution method to acquire the optimal spatial detail image. Finally， the optimal spatial detail image is injected into the upsampled multispectral image to obtain the final fused image. Compared with the other seven fusion methods， the experimental results show that the fused images obtained by the proposed method have lower spectral distortion and higher spatial resolution， and the proposed method provides the best values in terms of the spectral distortion index， the spatial distortion， and the QNR index.