基于控制核回归的多帧图像超分辨率重建

Multi-Frame Super-resolution Reconstruction Based on Steering Kernel Regression

针对多帧图像超分辨率重建问题,本文提出了一种基于控制核回归的重建方法.该方法先对低分辨率图像序列进行亚像素配准,再利用控制核回归进行非均匀插值得到高分辨率图像的初始估计,然后通过迭代控制核回归进行进一步的鲁棒估计,最后经过全变差正则化图像复原得到高分辨率重建图像.本文针对模拟图像序列及真实视频图像序列进行了对比实验,结果表明本文方法不但可以较好地保持图像中的细节信息,而且有较好的去噪能力,能够有效地提高图像重建质量.

Image super-resolution （SR） reconstruction can produce high-resolution （HR） images from single or multi-frame low-resolution （LR） images. This paper proposes a novel approach of multi-frame SR based on steering kernel regression （SKR）. First, the aligned LR image sequence is fused to generate an initial estimation of HR image with SKR. Then, a procedure of iterative SKR is performed to obtain robust estimation of HR image. Finally the reconstructed HR image is achieved after deblurring with TV-regularization. Experiments on synthetic LR images and real-world video sequence are carried out. The experimental results show that the proposed multi-frame SR method can reconstruct better images both quantitatively and perceptually and reserve effectively the detailed in- formation in the images as well.