数字图像超分辨率重建技术综述  

Survey of Digital Image Super Resolution Reconstruction Technology

作　　者：肖宿[1] 韩国强[1] 沃焱[1] 

机构地区：[1]华南理工大学计算机科学与工程学院,广州510006

出　　处：《计算机科学》2009年第12期8-13,54,共7页Computer Science

基　　金：国家自然科学基金资助项目(60573019);广东省自然科学基金博士科研启动基金资助项目(07300561);广东省重点实验室开放基金项目(CCNL-200704);广东省科技计划项目(2007B010200050)资助

摘　　要：图像超分辨率重建的目的是通过一幅或多幅低分辨率降质图像来估计一幅视觉效果较好的高分辨率图像。它从传统的图像恢复与重建技术而来,利用图像之间的信息互补来获得比单幅图像更多的细节。超分辨率技术主要分为两大类:基于重建的超分辨率技术和基于学习的超分辨率技术。基于重建的超分辨率技术按照特定的退化模型,通过输入的图像来估计高分辨率图像。基于学习的超分辨率技术从训练样本中获取先验知识,对输入图像的信息进行补充,可以获得比基于重建的算法更好的效果。对超分辨技术的算法作了系统的介绍,并指出图像的配准、退化模型的建立、盲估计问题、学习模型的建立、学习算法等仍是图像超分辨率技术中存在的主要问题,也是进一步研究的方向。The purpose of image reconstruction is to estimate a high-resolution image with better vision effect from one or a sequence of images. Super resolution that comes from previous image restoration and reconstruction technology can take advantage of redundant information among images. Super resolution reconstruction could include two kinds of technology：reconstruction-based technology and learning-based technology. The reconstruction-based technology estimates a high-resolution image from input images according to specific degradation model. Learning-based technology supplies in-put images with prior knowledge from training examples to get a better result, The paper systematically introduced algorithms of super resolution technology. Finally, we pointed out that image registration, reconstruction of degradation model and learning model, blind estimation, learning algorithms were main problems and future direction in image super resolution technology.

关 键 词：超分辨率 退化模型 学习模型 MARKOV随机场 最大后验概率(MAP) 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]