基于边信息的高光谱图像恢复模型

在高光谱图像(HSI)恢复中,如何在模型中有效嵌入先验信息和正确建模噪声一直是研究的两个重点.边信息作为一种基于域的先验知识已经在许多方向取得了成功,然而在高光谱去噪领域仍未受到关注.为了将这种领域知识与高光谱恢复模型自然耦合,提出的方法采用双线性映射的方式将边信息链接到表示观测数据潜在低秩结构的底层矩阵,并使用E-3DTV(enhanced 3-D total variation)正则编码了HSI局部平滑先验.此外该方法使用Lp范数进行噪声建模,进一步增强对腐败的鲁棒性.该方法在两个数据集、七种加噪方式下与五种竞争方法在三个数值指标上进行了比较,结果充分反映了提出方法对复杂噪声场景的有效性和鲁棒性.

Hyperspectral Image Restoration Model With Side Information

For hyperspectral image (HSI) restoration, how to effectively embed prior information in the model and correctly model the noise has always been the two focus of research. As a domain-dependent prior knowledge, side information has succeeded in many aspects, but it has not received much attention in the field of hyperspectral denoising. In order to naturally couple this domain knowledge with the hyperspectral restoration model, the method linked side information to the underlying matrix representing the potential low-rank structure of the observation data via a bilinear mapping, and used E-3DTV (enhanced 3-D total variation) to encode HSI local smoothness prior. In addition, this method used the Lp norm for noise modeling to further enhance the robustness against corruption. This method was compared with five competitive methods on three numerical indicators in two data sets and seven noise addition methods. The results fully reflect the effectiveness and universality of the proposed method for complex noise scene. The code was published in https://github.com/zsj9509/TVF.