激光与光电子学进展, 2018, 55 (7): 071014, 网络出版: 2018-07-20
空间光谱联合稀疏表示的高光谱图像超分辨率方法 下载: 1068次
Hyperspectral Image Super-Resolution Method Based on Spatial Spectral Joint Sparse Representation
图像处理 超分辨率重建 高光谱图像 联合稀疏表示 同步正交匹配追踪 信号非负性 空间结构相似性 image processing super-resolution reconstruction hyperspectral image joint sparse representation simultaneous orthogonal matching pursuit signal non-negative spatial structure similarity
摘要
针对获取的高光谱图像空间分辨率较低的问题,设计了一种空间光谱联合稀疏表示的超分辨率方法:提取图像中不同的反射光谱,通过压缩感知字典学习算法得到强稀疏性、弱相干性的光谱字典;利用高光谱图像信号的稀疏性、非负性以及空间结构相似性,通过同步正交匹配追踪算法,从相同场景的高空间分辨率的低光谱图像求解得到稀疏编码矩阵;联合光谱字典和稀疏编码矩阵得到目标图像。由于联合使用高光谱图像的空间与光谱信息,仿真实验数据和真实实验数据结果表明,相比于传统方法和矩阵分解方法本文方法,能够有效重建图像细节信息与纹理结构,有效提高波段平均峰值信噪比、波段平均结构相似度以及光谱角映射,并且更好地保持光谱信息。
Abstract
In order to solve the problem of low spatial resolution of hyperspectral image, a method based on spatial spectral joint sparse representation is designed. Firstly, we extract the different reflectance spectra of scenes and obtain a spectral dictionary with strong sparsity and weak coherence by exploiting compressed sensing dictionary learning algorithm. Then using sparsity, non-negativity and spatial structure similarity of hyperspectral image signals, we obtain the sparse coding matrix from the high-spatial resolution low-spectral image of the same scene by the simultaneous orthogonal matching pursuit algorithm. Finally, we combine the spectral dictionary with sparse coding matrix to get the target image. As a result of the combined spatial and spectral information, the simulated data and real data experimental results show that this method can effectively reconstruct image detail information and texture structure compared with bicubic interpolation method and matrix decomposition method, and effectively improve the value of average peak signal-to-noise ratio, average structural similarity, and spectral angel mapper, and maintain the spectral information better.
许蒙恩, 谢宝陵, 徐国明. 空间光谱联合稀疏表示的高光谱图像超分辨率方法[J]. 激光与光电子学进展, 2018, 55(7): 071014. Xu Meng′en, Xie Baoling, Xu Guoming. Hyperspectral Image Super-Resolution Method Based on Spatial Spectral Joint Sparse Representation[J]. Laser & Optoelectronics Progress, 2018, 55(7): 071014.