基于冗余字典的高光谱图像超分辨率复原算法

为了提高高光谱图像的空间分辨率,将基于冗余字典的信号稀疏表示理论应用到高光谱图像的超分辨率复原领域,提出一种基于冗余字典的高光谱图像超分辨率复原算法.该算法通过训练一组高低分辨率相对应的冗余字典对,使得高低分辨率相对应的像元曲线在基于各自的冗余字典进行稀疏分解时,具有相同的稀疏表示系数.超分辨率复原过程中,将待复原的低分辨率高光谱图像基于低分辨率冗余字典进行稀疏分解,利用所得的稀疏表示系数和对应的高分辨率字典,重建高分辨率的图像.实验结果表明:与基于图像块字典的超分辨率复原算法及传统的双线性插值图像放大方法相比,重建图像的峰值信噪比(peak signal to noise radio,PSNR)得到了显著提高.该算法将高光谱图像沿光谱维方向进行整体稀疏分解,避免了传统算法逐波段进行超分辨率复原带来的波段间的光谱失真问题,显著降低了算法的运算量.

Algorithm of Spectral Super-resolution of Hyperspectral Imagery Based on Redundant Dictionary

To enhance the spatial resolution of hyperspectral image, a hyperspectral image super-resolution restoration algorithm based on redundant dictionary was presented in this paper. By training a group of high and low resolution redundant dictionary, the corresponding image element curve of high and low resolution was made to have the same sparse representation coefficients in sparse decomposition based on redundant dictionary in this algorithm. During the process of super-resolution restoration, the low resolution of hyperspectral image sparse decomposes based on low resolution redundant dictionary. The high resolution image was reconstructed by using the sparse representation coefficients and the high resolution dictionary. The experimental results show that, compared with the image patch based sparse super resolution algorithm and the traditional image bilinear interpolation method, the PSNR of image reconstruction is significantly enhanced. The algorithm sparse decomposes the hyperspectral image along the spectral dimension to avoid the traditional algorithm problem of spectral distortion caused by restoration. The computational complexity of the algorithm is significantly reduced.