欢迎访问《图学学报》 分享到：
图学学报

图学学报

• 视觉与图像 • 上一篇    下一篇

基于倒易晶胞特征增强的图像超分辨算法

  

  1. 1. 南京理工大学计算机科学与工程学院，江苏 南京 210094；
    2. 南京信息工程大学信息与控制学院，江苏 南京 210044
  • 出版日期:2017-08-31 发布日期:2017-08-10
  • 基金资助:
    国家自然科学基金项目(61273251，61673220)

A Single Image Super Resolution Algorithm Based on Reciprocal Cell Model

  1. 1. School of Computer Science and Technology, Nanjing University of Science and Technology, Nanjing Jiangsu 210094, China;
    2. School of Information and Control Engineering, Nanjing University of Information Science and Technology, Nanjing Jiangsu 210044, China
  • Online:2017-08-31 Published:2017-08-10

PDF (PC)

156

摘要： 为提高单幅降质图像的分辨率，利用倒易晶胞模型改进了基于样例学习的超分辨
算法。首先，在Freeman 样例学习超分辨理论框架下，结合倒易晶胞滤波模型增强低分辨率图
像特征；然后，将特征增强的低分辨图像与高分辨率图像进行细节对应关系训练；最后，利用
训练好的对应关系实现低分辨图像的超分辨重建。该算法削弱了样例学习算法训练阶段“一对
多”的病态问题，有效减小了高、低分辨率图像特征空间内在“维度差”。实验结果表明，与双三
次插值、邻域嵌入、样例学习超分辨算法相比，该算法在超分辨重建图像主观视觉质量和峰值
信噪比(PSNR)客观评价指标中均优于比较算法。

关键词: 样例学习, 倒易晶胞, 维度差, 超分辨重建

Abstract: In order to improve single image resolution, a new algorithm based on the reciprocal cell
model is proposed. First, based on the framework of example-based learning super resolution theory
by Freeman, a pre-filtering step is introduced by reciprocal cell model. Then a corresponding
relationship is established, within the feature enhanced low resolution image and the original high
resolution image. At last, the super-resolution reconstruction is completed by using the trained
correspondence. The characteristics of the low resolution image are enhanced after the new
pre-filtering algorithm. The problems of “one-to-many” and “dimension difference” in training
database between the low and high resolution image feature space is effectively weakened.
Comparing with other algorithms such as bi-cubic interpolation, neighbor embedding and the
example-based learning algorithm, our experimental results show that the new approach has better
effect on subject image quality and PSNR index.

Key words: example-based learning, reciprocal cell, dimension difference, super resolution