基于多曲面拟合和单帧学习信息的图像超分辨率方法

项目名称： 基于多曲面拟合和单帧学习信息的图像超分辨率方法

项目编号： No.61271393

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 廖庆敏

作者单位： 清华大学

项目金额： 88万元

中文摘要： 视频监控得到了日益广泛的应用，但现有的视频监控系统存在目标图像分辨率低、清晰度不够等问题，导致了目标辨识的困难。为此，本项目提出基于多曲面拟合与单帧学习信息的图像超分辨率方法，拟在现有的视频监控图像质量基础之上显著提高图像分辨率，以便提高目标的辨认程度。围绕超分辨率所涉及的关键技术，本项目主要研究内容包括：针对投影变换的局部目标亚像素配准新方法、基于多曲面拟合新思路的插值重建、基于不相关子字典的学习型超分辨率新方法、以及单帧学习和多帧重建相结合的新方案等四部分内容。预期研究成果将提高图像超分辨率的理论研究水平及其在视频监控、遥感遥测、医学成像及高清显示等领域的应用能力。

中文关键词： 多帧图像超分辨；单帧图像超分辨；图像重建；多曲面拟合；学习

英文摘要： Although video surveillance system has been widely applied in many fields, the existsing surveillance images and videos are of low resolution and poor visibility, which results in the difficulty of object recognition. Therefore, we propose a super-resolution method based on multiple surfaces fitting and single frame learning, with the aim of enhancing image resolution and promoting the discriminability for objects on the basis of current surveillance images. The main contents contain several key techniques,including new approaches of sub-pixel registeration for objects with independent perspective motions, interpolation-based reconstruction using multiple surfaces fitting, learning-based method via incoherent sub-dictionaries learning and the mergence of the single frame learning with multiple frames method. The results obtained from this project are expected to improve both theorical level and applicational ability of our country on the areas such as video surveillance,remote sensing & telemetry, medical imaging and high-definition display.

英文关键词： multi-frame super-resolution；single image super-resolution；image reconstruction；multiple surfaces fitting；learning