改进的序列图像超分辨率重建方法

Improved super resolution reconstruction method for video sequence

为解决局部运动的视频序列在超分辨率重建过程中,由于采用传统的图像间全图一致变换模型可能导致的运动估计误差增大,影响重建效果的问题,提出了基于三角网不规则分块运动估计思想和基于DTN-POCS的重建算法。从边缘点中提取特征点,利用配准获得的同名点集,在主/从图像对间构建同名不规则三角网,并以此分割图像,默认每个三角块中像素运动一致。在此基础上,把所有低分辨率帧分块投影到高分辨率坐标,以凸集投影(POCS)迭代优化。试验结果表明:重建的高分辨率图对图像中的局部运动鲁棒性更强,能有效改善重建精度。

To decrease the negative effect caused by local shift motions, this paper proposed an approach based on the triangulated irregular block motion estimation and DTN-POCS super resolution reconstruction. First of all, select feature points from edges, obtain matching points set between image pairs from the motion estimation process; ＇after that, for each image pair, creates the Delaunay triangulation net for slave image based on the matching feature points, and also creates its cprresponding triangular net for the master image; and then, suppose the homologous triangulations between images have the relationship of affine transformation, project all the low resolution images by sub-block onto the high resolution coordinate, generate the initial image by interpolation methods; in the end, iteratively optimize the super resolution image based on the improved method. In the experiment, the method with image uniform affine transformation parameters has less accuracy than that of the improved one. The simulation results testify that the proposed method can improve the reconstruction performance.