电子学报 ›› 2015, Vol. 43 ›› Issue (7): 1275-1280.DOI: 10.3969/j.issn.0372-2112.2015.07.004

• 学术论文 • 上一篇    下一篇

基于二阶广义全变差的多帧图像超分辨率重建

任福全1, 邱天爽1, 韩军2, 金声3   

  1. 1. 大连理工大学电子信息与电气工程学部, 辽宁大连 116024;
    2. 大连现代高技术集团有限公司, 辽宁大连 116025;
    3. 澳大利亚纽卡索大学信息技术学院, 纽卡索, 2300
  • 收稿日期:2013-11-04 修回日期:2014-12-17 出版日期:2015-07-25
    • 通讯作者:
    • 邱天爽
    • 作者简介:
    • 任福全 男,1984年生于辽宁营口.现为大连理工大学博士研究生.研究方向为图像与视频处理,压缩感知. E-mail:renfu_quan@hotmail.com
    • 基金资助:
    • 国家自然科学基金 (No.81241059,No.61172108); 国际合作专项资助 (No.2012DFA10700)

Multiframe Image Super Resolution Based on Second Order Total Generalized Variation

REN Fu-quan1, QIU Tian-shuang1, HAN Jun2, JIN Sheng3   

  1. 1. Faculty of Electrical and Electronic Engine, Dalian University of Technology, Dalian, Liaoning 116024, China;
    2. Dalian Modern High Techology Group Ltd, Dalian, Liaoning 116025, China;
    3. School of Information Techrology, University of Newcastle, Newcastle 2300, Australia
  • Received:2013-11-04 Revised:2014-12-17 Online:2015-07-25 Published:2015-07-25
    • Supported by:
    • National Natural Science Foundation of China (No.81241059, No.61172108); International Cooperation Project (No.2012DFA10700)

摘要:

图像超分辨率重建是图像处理领域的重要问题.本文将二阶广义全变差用于基于正则化的多帧图像超分辨率重建问题,构建了基于二阶广义全变差正则项的图像超分辨率模型.为了更好地保持重建图像的边缘和细节,采用图像空域自适应正则化参数,并针对该重建模型的非光滑性,给出了基于半二次正则化和交替方向法的求解算法.实验结果表明该模型和数值算法能够较好地提高图像的分辨率,同时可以较好地保持图像的细节信息.

关键词: 超分辨率重建, 二阶广义全变差, 自适应正则化参数, 半二次正则化, 交替方向法

Abstract:

Super resolution is a very important issue in the field of image processing.In this paper,Second order total general variation (TGV) is used to solve a reconstruction based multi-frame images super resolution problem.A second order total general variation based super resolution model is built in the paper.In order to preserve the edges and detail information,a spatial adaptive regularization parameter of the image is applied.For solving the non-smooth problem,a half quadric and alternating direction based numerical algorithm is proposed.Experiments verified that the proposed model and numerical algorithm can effectively increase the resolution of an image,and they can preserve the texture and detail information.

Key words: super resolution, second order total generalized variation, adaptive regularization parameter, half-quadratic regularization, alternating direction method

中图分类号: