针对液晶可调滤波片高光谱成像系统记录动态场景的成像特点, 提出一种图-谱结合的压缩感知高光谱视频图像复原方法。首先, 通过前景目标检测获得运动前景目标的高光谱图像, 实现运动前景目标与背景区域分离, 并根据前景目标检测结果将背景区域划分为运动区域(被前景目标遮挡区域)与静止区域(未被前景目标遮挡区域)。然后, 基于高光谱图像空间维、光谱维相关性, 对静止区域进行字典学习获得稀疏先验信息, 结合压缩感知理论用于运动区域恢复, 得到完整的背景区域高光谱图像。最后, 将运动前景目标高光谱图像与背景区域高光谱图像相结合, 得到高光谱视频图像。实验结果表明: 本文提出的高光谱视频图像复原方法在峰值信噪比和视觉效果上都要优于现有算法, 峰值信噪比平均提高5 dB以上。

In this paper, a new graph-spectral hyperspectral video restoration method regarding the imaging characteristics of dynamic scenes recorded by liquid crystal tunable filter hyperspectral imaging system is proposed. Firstly, the hyperspectral image of the moving foreground target is obtained by the foreground target detection, and the moving foreground target is separated from the background region. Then the background region is divided into the motion region which is obscured by the foreground target and the still region which is not obscured by the foreground target according to the foreground target detection result. Based on the correlation of the spatial dimension and spectral dimension of the hyperspectral image, dictionary learning is performed on the still region to obtain sparse prior information. Combined with compressed sensing theory for motion region recovery, a complete background region hyperspectral image is obtained. Finally, the moving foreground target hyperspectral image is combined with the background region hyperspectral image to obtain a hyperspectral video image. The experimental results show that the proposed method of hyperspectral video image restoration outperforms the existing algorithm in terms of peak signal-to-noise ratio and visual effect, and the peak signal-to-noise ratio is increased by an average of more than 5 dB.