改进的有参超分辨率图像盲恢复

Improved Parametric Blind Super-resolution Image Restoration

改进了有参超分辨率图像盲恢复中的边界与正则处理以及模糊参数估计方法,用交替迭代优化来估计模糊参数和高分辨率图像,用Neumann边界图像模型和修改边界像素来避免边界错误,用近似重排系统矩阵的带结构矩阵,来构造求解高分辨率图像的预处理共轭梯度算法。人工退化图像序列上的实验结果表明了改进方法的有效性。

To improve the boundary condition processing, the regularization processing, and the blurring parameter estimation of parametric blind super-resolution image restoration, an alternative iteration optimization is used to estimate the image blurring and high-resolution image. The neumann boundary condition and the boundary pixel modification of the lower-resolution image are adopted to avoid the boundary errors. In solving the high-resolution image, the matrixes close to the reordered system matrix and with special structures are used to make the preconditioner in the preconditioned conjugate gradient algorithm. The test results using the synthetic low-resolution sequences are presented to show the validation of the proposed improvements.