M-估计耦合双边滤波的正则化超分辨率重建

Regularized super-resolution reconstruction based on M-estimation and bilateral filtering

在正则化超分辨率重建框架下,基于M-估计理论和双边滤波思想,建立了一种鲁棒的超分辨率重建统一能量泛函。该能量泛函融合了M-估计的鲁棒性处理机制和双边滤波的双重异性加权机制,提高了算法的鲁棒性和边缘保持特性。鉴于采用最小二乘估计的CLS算法和采用最小一乘估计的Farsiu重建算法在边缘保持特性方面存在的不足,在算法实现时选用了Huber稳健M-估计。不论是视觉效果还是峰值信噪比(PSNR),实验结果都表明该算法的有效性。

In regularized super-resolution reconstruction framework, a unified and robust energy function for super-resolution reconstruction was constructed, which incorporated both the robustness of M-estimation and the double-weighting idea of bilateral filtering, and hence behaving much better in robustness and edge-preserving. Because of drawback of the constrained least square algorithm using least square estimator and Farsiu＇s algorithm using least absolute deviation estimator in the edge-preserving, the robust Huber estimator was used in the unified energy function. The experimental results demonstrate the effectiveness of proposed algorithm, both in the visual effect and the Peak Signalto Noise Ratio （PSNR） value.