基于光谱相似性的高光谱图像超分辨率算法

光谱相似性是指高光谱图像中的大量像元具有相似光谱的性质.提出了一种基于光谱相似性的高光谱遥感图像超分辨率算法,利用遥感图像中广泛存在的结构自相似性提升图像的空间分辨率,利用高光谱图像的低维子空间性通过主成分分析降低光谱维数提高运算效率,利用具有相似光谱的像元构建光谱约束项保证重建图像光谱的准确性.该算法在将单波段图像超分辨率方法推广到处理具有数百、乃至上千波段的高光谱图像过程中,既保证了重建图像光谱的准确性,又具有较高的运算效率.实验表明,与双三次插值和基于稀疏表示与光谱正则化约束的高光谱图像超分辨率算法相比,该算法具有更高的空间分辨率提升能力和更好的光谱保真能力.

Spectral Similarity-based Super Resolution for Hyperspectral Images

Spectral similarity refers to that there are many pixels with similar spectrum in a single hyperspectral image. In this paper, we propose a spectral similarity-based super resolution method for hyperspectral images. In our method, the extra information exploited from structural self-similarity which widely exists in remote sensing images is used to promote the spatial resolution. The principal component analysis is used to reduce the spectral dimension for increasing the computational efficiency according to the inherent low dimensionality of hyperspectral images, and the spectral similarity is used to construct spectral regularization for ensuring the accuracy of the spectrum in the reconstructed image. Our method can achieve accurately reconstructed results as well as high computational efficiency, when we extend the single-band super resolution method to the hyperspectral image with hundreds of bands. Experimental results demonstrate that our method can improve the spatial resolution more effectively and reconstruct the spectrum more accurately than the bicubic interpolation and the sparse representation and spectral regularization method (SRSRM).