基于峭度图像的超分辨率重建算法

针对在强噪声环境下，传统的超分辨率重建算法重建图像效果不佳的问题，提出一种基于峭度图像的超分辨率重建算法。定义峭度图像，从统计学角度分析得到峭度图像的2个重要性质，即具有高斯不变性，并且图像越模糊，峭度绝对值越小。在满足高分辨率图像与低分辨率图像之间反卷积的剩余误差有界的前提下，通过最大化峭度绝对值求解未知的高分辨率图像，采用Lagrange乘子法则求解此约束优化问题。分析高斯噪声和非高斯噪声环境下算法性能。仿真结果表明，该算法在主观视觉和客观评价上都明显优于传统算法。

Super-resolution Reconstruction Algorithm Based on Kurtosis Image

Performance of the traditional Super-resolution（SR） algorithm will degrade in intensive noise environment. In this paper, a novel kurtosis based SR method is proposed which is robust for SR reconstruction from Low-resolution（LR） images with low SNR. The definition of the kurtosis image is given and its two properties are analyzed： （l）The kurtosis image is Gaussian noise invariant; （2）The smoother the image is, the smaller the absolute value of a kurtosis image. The SR image is estimated by maximizing the local absolute kurtosis with the constraint that the reconstructed image is consistent with the observed data of the same image. The constraint optimization problem is solved by using the Lagrange multiplier method. The proposed algorithm is tested on images corrupted by a Gaussian noise and non-Gaussian noise. Experimental results demonstrate that the proposed method is superior both on subjective and objective evaluation.