基于张量截断核范数的高光谱图像超分辨率重构

Super-Resolution Reconstruction of Hyperspectral Images Based on Tensor Truncated Nuclear Norm

针对高光谱图像在获取过程中存在多种不同程度退化的问题,提出一种基于张量截断核范数和空谱全变差正则化模型,实现了高光谱图像的超分辨重构。首先分析高光谱图像的两种先验信息:空谱低秩先验和空谱稀疏先验;利用空谱低秩先验建立基于张量截断核范数的低秩约束模型,实现对秩函数的准确逼近;利用空谱稀疏先验建立空谱全变差正则化模型,有效地保持图像的边缘信息;最后结合两种模型的优势,建立基于张量截断核范数和空谱全变差正则化的高光谱图像重构模型。实验结果表明新模型提高了视觉质量,与目前最新的超分辨率重构模型相比,本文方法的平均峰值信噪比提高了0.8dB。新模型充分利用高光谱图像的空间和光谱稀疏低秩先验,针对模糊化和下采样后的高光谱图像,能够有效实现高光谱数据的超分辨率重构。

Based on the tensor truncated nuclear norm and the spatial-spectral total variation regularization,a new model is proposed to realize super-resolution reconstruction of hyperspectral images to solve the problem that most hyperspectral images suffer from degradation in the acquisition process.First,two types of priori information in the hyperspectral images,i.e.,the low rank-based priori information and sparse priori in the spatial and spectral domain,are explored.Next,using the low rank-based priori information in the spatial and spectral domain,a lowrank constraint model based on the tensor truncated nuclear norm is proposed to achieve a more accurate approximation of the rank function.Subsequently,using sparse priori information in the spatial and spatial domain,a spatial and spectral total variation regularization model is proposed to retain the sharp edges and more detailed information of the original image.Finally,the low-rank constraint model based on the tensor truncated nuclear norm and spatial and spectral total variation models are integrated.This new restoration model possesses the advantages of both the aforementioned models.The peak signal-to-noise ratio of 0.8 dB is obtained,and structural similarity indices are adopted to provide quantitative assessments of experimental results.The experimental results demonstrate that the proposed model achieves better visual quality than those of several existing related methods.The proposed model can effectively achieve the super-resolution reconstruction of hyperspectral images after being blurred and downsampled.