计算机应用 ›› 2013, Vol. 33 ›› Issue (02): 472-475.DOI: 10.3724/SP.J.1087.2013.00472

• 多媒体处理技术 • 上一篇    下一篇

新的基于稀疏表示单张彩色超分辨率算法

杨玲,刘怡光,黄蓉刚,黄增喜   

  1. 四川大学 计算机学院,成都 610065
  • 收稿日期:2012-08-06 修回日期:2012-09-21 出版日期:2013-02-01 发布日期:2013-02-25
  • 通讯作者: 杨玲
  • 作者简介:杨玲(1987-),女,重庆人,硕士研究生,主要研究方向:计算机视觉、图像超分辨率;
    刘怡光(1972-),男,四川内江人,教授,博士,主要研究方向:计算机视觉、图像处理、神经网络、机器智能;
    黄蓉刚 (1979-),男,四川富顺人,博士研究生,主要研究方向:机器视觉;
    黄增喜(1985-),男,广西百色人,博士研究生,主要研究方向:图像处理、生物特征识别、稀疏表示。
  • 基金资助:
    国家自然科学基金资助项目;四川省国际科技合作与交流研究计划项目;四川省应用基础项目

New approach for super-resolution from a single color image based on sparse coding

YANG Ling,LIU Yiguang,HUANG Ronggang,HUANG Zengxi   

  1. College of Computer Science, Sichuan University, Chengdu Sichuan 610065, China
  • Received:2012-08-06 Revised:2012-09-21 Online:2013-02-01 Published:2013-02-25
  • Contact: YANG Ling

摘要: 传统的基于学习的超分辨率算法普遍采用样本库来训练字典对,训练时间长且对样本库依赖较大。针对传统算法的不足,提出一种新的单张彩色图像超分辨率算法。该方法基于稀疏编码超分辨率模型,利用图像自相似性和冗余特性,并结合图像金字塔结构,采用低分辨率图像本身来训练高、低分辨率图像块的字典对。同时,针对彩色图像,该算法采用一种基于稀疏表示的彩色图像存储技术,将彩色图像的三通道值组合成一个向量进行图像稀疏处理,以更好地维持原始图像细节信息。实验结果表明,与传统的超分辨率算法相比,该算法不但有更好的视觉效果和更高的峰值信噪比(PSNR),而且计算速度快。

关键词: 基于学习的超分辨率, 稀疏编码, 字典对, 图像金字塔, 彩色图像存储

Abstract: Traditional learning-based super-resolution algorithms generally adopt training images for learning dictionary pairs, they are time-consuming, and the results strongly depend on the training images. To address these problems, a new super-resolution approach from a single color image was proposed based on sparse coding model. According to image self-similarity and redundancy features, this algorithm utilized low-resolution image itself for training dictionary pairs, combined with image pyramid structure. Meanwhile, in view of color images, the sparse representation based color image storage technology was used, which concatenated the values of three channels to a single vector and directly represented them sparsely. The experimental results illustrate that the proposed method not only can generate high-resolution images with better visual effects and higher Peak Signal-to-Noise Ratio (PSNR) but also has less computation time.

Key words: learning-based super-resolution, sparse coding, dictionary pair, image pyramid, color image storage

中图分类号: