电子学报 ›› 2016, Vol. 44 ›› Issue (5): 1189-1195.DOI: 10.3969/j.issn.0372-2112.2016.05.025

• 学术论文 • 上一篇    下一篇

基于耦合特征空间下改进字典学习的图像超分辨率重建

詹曙1, 方琪1, 杨福猛2, 常乐乐1, 闫婷1   

  1. 1. 合肥工业大学计算机与信息学院, 安徽合肥 230009;
    2. 三江学院电子信息工程学院, 江苏南京 210012
  • 收稿日期:2014-12-17 修回日期:2015-09-14 出版日期:2016-05-25
    • 作者简介:
    • 詹曙 男,1968年生于安徽合肥.副教授、硕士生导师,合肥工业大学计算机与信息学院.研究方向为三维人脸识别和医学图像处理.E-mail:shu_zhan@hfut.edu.cn;方琪 男,1991年生于安徽灵璧.硕士研究生,合肥工业大学计算机与信息学院.研究方向为数字图像处理.E-mail:fq9110@mail.hfut.edu.cn
    • 基金资助:
    • 国家自然科学基金 (No.61371156); 安徽科技攻关计划 (No.1401B042019)

Image Super-Resolution Reconstruction via Improved Dictionary Learning Based on Coupled Feature Space

ZHAN Shu1, FANG Qi1, YANG Fu-meng2, CHANG Le-le1, YAN Ting1   

  1. 1. School of Computer & Information, Hefei University of Technology, Hefei, Anhui 230009, China;
    2. School of Electronic Information Engineering, Sanjiang University, Nanjing, Jiangsu 210012, China
  • Received:2014-12-17 Revised:2015-09-14 Online:2016-05-25 Published:2016-05-25
    • Supported by:
    • National Natural Science Foundation of China (No.61371156); Science and Technology Research and Development Project of Anhui Province (No.1401B042019)

摘要:

针对目前基于字典学习的图像超分辨率重建效果欠佳或字典训练时间过长的问题,本文提出了一种耦合特征空间下改进字典学习的图像超分辨率重建算法.该算法首先利用高斯混合模型聚类算法对训练图像块进行聚类,然后使用更改字典更新方式的改进KSVD字典学习算法来快速获得高、低分辨率特征空间下字典对和映射矩阵.重建时根据测试样本与各个类别的似然概率自适应地选择最匹配的字典对和映射矩阵进行高分辨率重建.最后利用图像非局部相似性,将其与迭代反向投影算法相结合对重建后的图像进行后处理获得最佳重建效果.实验结果表明了本文方法的有效性.

关键词: 超分辨率, 字典学习, KSVD, 稀疏表示, 混合高斯模型

Abstract:

Image super-resolution reconstruction via Improved Dictionary Learning based on Coupled Feature Space is studied in the paper, in order to solve the following problems:1 the dictionary training process is time-consuming, 2 the results are not satisfactory in the existing algorithms.In the proposed algorithm, at first, the Gaussian mixture model clustering algorithm is employed to cluster the training image blocks, secondly, quickly obtain high and low resolution feature space of dictionary and mapping matrix by using dictionary updating based on improved KSVD dictionary learning algorithm, and then, the Super-Resolution image is reconstructed according to the likelihood probability of test samples, in which each category adaptively selected the most matching dictionary and mapping matrix for high-resolution reconstruction, finally, the non-local similarity and iterative back-projection are exploited to furtherly improve the quality of the reconstruction image.The experimental results demonstrate the validity of the proposed algorithm.

Key words: super-resolution, dictionary-learning, ksvd, sparse representation, gaussian mixture model

中图分类号: