基于低秩字典学习的高光谱遥感图像去噪

Hyperspectral Image Denoising Based on Low Rank Dictionary Learning

针对高光谱遥感图像(Hyperspectral Image,HSI)去噪问题,提出了基于非局部低秩字典学习的图像去噪算法。该算法利用高光谱遥感图像各波段之间的强相关性,结合图像非局部自相似性和局部稀疏性提高去噪性能。首先,结合各波段图像的强相关性、非局部自相似性和局部稀疏性建立非局部低秩字典学习模型,然后,利用迭代法求解该模型得到冗余字典和稀疏表示系数,最后,利用冗余字典和稀疏表示系数复原图像。相比较现有先进的算法,由于充分利用了高光谱图像各波段的强相关性这一内在特征,使得该算法能够很好地保持高光谱遥感图像的细节信息,达到了预期效果。

For the multi-spectral remote sensing image denoising problem, the image denoising algorithm based on nonlocal low rank dictionary learning is proposed. The basic idea of this algorithm is to improve the image denoising effect by simultaneously using high correlation between each band image, nonlocal similarity and local sparity. Firstly, the nonlocal low rank dictionary learning model is constructed by the high correlation between each band, nonlocal similarity and local sparity. Then this model is solved by the iterative method to obtain redundant dictionary and sparse represent coefficients. Finally, the denoising image can be restored by the redundant dictionary and sparse represent coefficients. Compared with state-of-the-art methods, the PSNR of the proposed algorithm is higher for taking advantage of the high correlation between each band image, while image detail information can be preserved to achieve the expected results.