基于全变差和稀疏表示的图像超分辨率重构

Image super-resolution based on total variation and sparse representation

为了解决监控图像中目标区域分辨率较低的问题,文中提出了一种图像超分辨率重构算法,算法分为两个阶段:字典训练和重构。在训练阶段,首先将低分辨率样本图像采用全变差的方法扩大到与其对应的高分辨率图像的大小,然后将图像分块并提取特征,最后基于高低分辨率图像块共享稀疏系数的思想,通过一定的算法,训练获得高低分辨率字典对。在重构阶段,首先按训练阶段相同的方法提取输入图像特征块,然后结合低分辨率字典获得稀疏系数,接着通过稀疏系数和高分辨率字典获得重构的图像块,最后将所有图像块融合为高分辨率图像。实验结果表明,算法在监控图像中取得了较为理想的超分辨率重构效果。

In order to solve the problem that the resolution of the target area in the monitored image is relatively low,an image super-resolution algorithm is proposed. There are two stages： dictionary learning and reconstruction. In the learning stage,each low resolution example image is firstly up-sampled to the same size with its corresponding high resolution image by total variation. Then the example images are divided into patches,and features of the patches are extracted. At last,a dictionary pair of high resolution and low resolution is obtained by some algorithm. In the reconstruction stage,firstly,the input image feature is extracted in the same way as in the learning stage. Then,combined with low resolution dictionary,sparse coefficients for each feature patch are solved. After that,the high resolution image patches are obtained by the coefficients and high resolution dictionary. At last,all the patches are merged into one high resolution image. The experimental results show that algorithm achieves good effect for monitoring images.