ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2020, Vol. 57 ›› Issue (2): 413-423.doi: 10.7544/issn1000-1239.2020.20190333

• 人工智能 • 上一篇    下一篇


王 旭, 陈 强, 孙权森   

  1. (南京理工大学计算机科学与工程学院 南京 210094) (
  • 出版日期: 2020-02-01
  • 基金资助: 

Multichannel Spectral-Spatial Total Variation Model for Diffractive Spectral Image Restoration

Wang Xu, Chen Qiang, and Sun Quansen   

  1. (School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094)
  • Online: 2020-02-01
  • Supported by: 
    This work was supported by the National Natural Science Foundation of China (61673220).

摘要: 在衍射成像光谱仪成像过程中,准焦谱段成像会受到其他离焦谱段的干扰而产生模糊.现有的重构算法只利用了图像空间信息,并且对于此类不适定反问题的复原效果不佳.因此,提出了一种基于多通道空间光谱全变差的正则化方法来重构衍射光谱图像.首先根据衍射光谱成像原理构建退化光谱图像的观测模型,然后在最大后验概率框架下结合空间和光谱先验信息建立复原模型.该方法充分利用衍射光谱图像的局部空间平滑性和局部光谱平滑性,并使用交替方向乘子法对模型进行有效的优化.大量实验表明,与其他的衍射光谱图像重构方法相比,此复原模型在平均峰值信噪比、平均结构相似度、平均光谱角距离和视觉质量方面都具有一定的优越性.此外,对于多通道模糊重叠且受噪声干扰的病态问题,该模型能够在保证求解速度的情况下抑制噪声,保留边缘信息,减缓锯齿状光谱失真的情况.

关键词: 衍射成像光谱仪, 全变差, 局部空间光谱光滑性, 交替方向乘子法, 图像复原

Abstract: During the imaging process of the diffractive optic imaging spectrometer, infocus images are often blurred by other defocused images. The existing recovery algorithms only utilize the spatial information and show limitations on these ill-posed inverse problems. In this paper, a regularization method based on the multichannel spectral-spatial total variation prior is proposed to reconstruct diffractive spectral images. First, an observation model of the degraded spectral images is constructed carefully, relying on the principle of diffractive spectral imaging. Then, a reconstruction model is established by combining the spatial and spectral prior information under the maximum posteriori framework. The proposed model makes full use of the local spatial smoothness and local spectral smoothness of the diffractive spectral images. Meanwhile, the ADMM (alternating direction method of multipliers) is employed to efficiently optimize the model. A large number of experiments demonstrate that this new restoration model has better performance in terms of average peak signal-to-noise ratio, average structural similarity, average spectral angular distance and visual quality compared with other restoration methods. In addition, for the ill-conditioned problem with cross-channel blurs and noise interference, this model can suppress noise, preserve edge information, and reduce jagged spectral distortion while ensuring the solution speed.

Key words: diffractive optic imaging spectrometer, total variation, local spatial-spectral smoothness, alternating direction method of multipliers (ADMM), image restoration