首页 | 官方网站   微博 | 高级检索  
     

基于最大化自相似性先验的盲单帧图像超分辨率算法
引用本文:李键红,吕巨建,吴亚榕.基于最大化自相似性先验的盲单帧图像超分辨率算法[J].计算机科学,2018,45(2):147-151, 156.
作者姓名:李键红  吕巨建  吴亚榕
作者单位:广东外语外贸大学语言工程与计算实验室 广州510006,广东技术师范学院 广州510665,仲恺农业工程学院科学技术处 广州510225
基金项目:本文受广东省科技计划项目(2016A020210131),广州市科技计划项目(201609010032),语言工程与计算实验室项目(LEC2016ZBKT004)资助
摘    要:图像的自相似性质和图像质量之间存在着密切的关系,清晰的自然图像中几乎所有的图像片都在其自身或较低尺度内存在着重复。然而,在存在噪声或模糊等降质处理的图像中,这一性质明显减弱。针对这一现象,提出一种最大化自相似性先验的盲单帧图像超分辨率算法。该算法通过迭代计算求解超分辨率图像和降质过程的模糊核,使得到的超分辨图像中的任一图像片在输入的低分辨率图像中都以最大的概率存在。这一算法不仅能够准确地计算降质过程的模糊核,得到高质量的高分辨率图像,而且其先验知识随着输入图像的不同而自动进行调整,使得算法具有更强的鲁棒性。大量实验表明,该算法的PSNR,SSIM参数结果较主流算法都有着明显的优势。

关 键 词:单帧图像超分辨率  盲超分辨率  自相似  成像模型  概率密度函数
收稿时间:2017/4/11 0:00:00
修稿时间:2017/5/28 0:00:00

Blind Single Image Super-resolution Using Maximizing Self-similarity Prior
LI Jian-hong,LV Ju-jian and WU Ya-rong.Blind Single Image Super-resolution Using Maximizing Self-similarity Prior[J].Computer Science,2018,45(2):147-151, 156.
Authors:LI Jian-hong  LV Ju-jian and WU Ya-rong
Affiliation:Laboratory of Language Engineering and Computing,Guangdong University of Foreign Studies,Guangzhou 510006,China,Guangdong Polytechnic Normal University,Guangzhou 510665,China and Department of Science and Technology,Zhongkai University of Agriculture and Technology,Guangzhou 510225,China
Abstract:The relationship between the self-similarity property of image and image quality is close,and almost all the patches in the clear natural image have recurrence patches in itself or its lower scale.However,in the image which was processed by blur or noise,this appearance is not dramatically.Aiming at this phenomenon,this paper proposed a blind single image super-resolution algorithm using the prior of maximizing self-similarity.This algorithm estimates the high-resolution image and the blur kernel by iterative computation,thus making any patch in the final estimated high resolution image exists in the inputted low resolution image with maximizing probability.The proposed algorithm not only estimates the degradation kernel and the high-resolution image accurately,but also adapts the prior according to the inputted image to make the result more robust.Extensive experiments illustrate that our algorithm shows obvious advantages when comparing to other main stream algorithms in terms of PSNR and SSIM.
Keywords:Single image super-resolution  Blind super-resolution  Self-similarity  Imaging model  Probability density function
点击此处可从《计算机科学》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号