结合稀疏编码模型的多帧图像超分辨率重建  

Multi- frame Image Super- resolution Reconstruction Combined with Sparse Coding Model

作　　者：卢健[1] 孙怡[1] 

机构地区：[1]大连理工大学信息与通信工程学院,辽宁大连116024

出　　处：《计算机工程》2015年第5期264-269,273,共7页Computer Engineering

摘　　要：传统序列超分辨率方法对低分辨率视频序列的要求较高,一旦序列中没有包含足够的信息,会造成重建高分辨率图像质量的下降。为此,提出一种结合稀疏编码模型的序列超分辨率算法。利用概率运动场从低分辨率序列中重建一幅高分辨率图像,根据自适应阈值确定重建有效和无效区域,使用稀疏编码模型对无效区域进行补全重建。实验结果表明,该算法可以采用序列自身的信息和稀疏字典中的信息来重建高分辨率图像,在序列信息有破缺时,与仅利用序列自身信息或仅利用单幅图像的算法相比,具有更好的鲁棒性和广泛的适用性。Classic multi-frame Super-resolution（ SR） techniques strongly rely on the supportability of Low-resolution （LR） frames. When the frames contain insufficient information,annoying artifacts often appear in the SR outcome. To solve this problem,a multi-frame SR combined with sparse coding technique is proposed in this paper. A high-resolution frame is reconstructed by the help of probabilistic motion estimation,and meanwhile effective/ineffective regions can also be determined by using an adaptive threshold segment. A sparse-coding-based completion technique is applied to recover the ineffective regions. Experimental results show that the proposed algorithm can essentially exploit the information from both LR frames and sparse coding dictionary. Compared with SR methods which depend only on image sequence itself or a single frame,the proposed algorithm has better robustness and extensive applicability.

关 键 词：超分辨率 稀疏编码 图像补全 非局部正则化 线性反问题 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]