自适应约束下的双边全变差正则化超分辨率重建

在经典的双边全变差(BTV)超分辨率重建中,加权系数和正则化参数的恒定性导致重建结果边缘保持能力受限。为此，提出了一种自适应约束的BTV正则化先验模型。算法首先定义了图像的局部邻域残差均值以区分当前像素属于平坦区域还是边缘区域；然后针对加权系数的不变性导致边缘削弱的问题,利用边缘方向和垂直边缘方向扩散性的不同,设计自适应权重矩阵；最后根据代价函数的极值问题推导出迭代公式,从而进行图像的超分辨率重建,重建过程中采用自适应的方法确定正则化参数,以便求得代价函数的全局最优解,提高了算法的鲁棒性。实验结果表明：与双三次线性插值法和经典BTV算法相比,该算法取得了更好的视觉效果和更高的峰值信噪比,更多地保留了图像的边缘细节信息。

Regularized super-resolution reconstruction based on bilateral total variation model with adaptive constraints

In the classical super-resolution reconstruction algorithm based on the Bilateral Total Variation(BTV) model,the ability of edge preserving is restricted due to the constancy of the weighted coefficient and the regularization parameter. To solve the problem, an adaptive regularization BTV model is proposed. Firstly,the algorithm defines the local neighborhood residual mean of the image to distinguish wheather the pixel belongs to flat area or edge area.Then, the adaptive weighting matrix is designed in case that the constant weighted coefficient leads to the weaken ability of edge preserving,which takes advantage of the different diffusivity between the direction of the edge and the vertical of edge direction;Finally,iterative formula is deduced about the extremum problem of the cost function,so that the super-resolution reconstruction can be achieved. In the process the adaptive regularization parameter is adopted in order to get the global optimal value of the cost function.The algorithm improves its robustness. The experimental results show that the proposed algorithm acquires the better visual effect and the higher peak signal-to-noise ratio(SNR),in addition,it retains the more image details information,compared with the super- resolution method based on classical BTV model and the bicubic method.