基于稀疏编码的视频序列超分辨率重构

Video Super-Resolution Reconfiguration Based On Sparse Coding

把学习型算法用于稀疏编码的重建算法中来实现视频序列图像的超分辨率重构。该算法无需显式求取运动向量,能够克服传统方法对精确运动估计的要求,通过稀疏编码便能够自动利用邻近帧中最相关的那些样本块进行重构;另外,算法通过设置最大运动窗口,利用帧间运动的连续性特点,在相邻帧已经重建的基础上,提取其运动窗口内的高、低分辨率图像块来构建样本库,从而实现减小所需样本库的尺寸的目的。

The paper adopts learning algorithm into the reconfiguration algorithm for spare coding to realize a super resolution （SR） reconfiguration of the video sequence images. This method doesn＇t need explicit motion estimation, can overcome the demand for explicit motion estimation in traditional methods. By using the sparse coding technique, the method can automatically make use of the most related patch samples in the adjacent frames when reconstructing. Moreover, the method sets the maximum motion window for each location, utilizes the continuity feature of the inter-frame motions, and extracts the high resolution （HR） patches and low resolution （LR） versions in the adjacent frame within this window on the basis of adjacent frames having reconfigured to form the corresponding database, thereby reducing the necessary size of the database.