图像信号多空间特征建模与优化重建方法研究

项目名称： 图像信号多空间特征建模与优化重建方法研究

项目编号： No.61071170

项目类型： 面上项目

立项/批准年度： 2011

项目学科： 轻工业、手工业

项目作者： 刘哲

作者单位： 西北工业大学

项目金额： 10万元

中文摘要： 近几年兴起的压缩感知理论突破了奈奎斯特采样定理的限制，论证了稀疏信号可以通过其少量的观测值精确重构出来，为信号的采集和处理提供了新的思路。本项目针对压缩感知理论的两项核心内容：信号的稀疏表示建模以及稀疏信号的优化恢复方法进行研究，提出了基于级联字典的图像稀疏分解新方法，改善了具有丰富纹理信息图像信号的稀疏分解效果；提出了自适应字典集学习的新方法，克服了当前稀疏表示算法中固定字典无法自适应图像局部结构的缺点，给基于字典学习的图像恢复/压缩感知带来了新的思路；提出了基于变量p迭代加权的图像重构算法，以及基于罚函数的稀疏信号重构算法，改善了基于lp范数优化的图像重构质量。在进行压缩感知基础理论和算法研究的基础上，拓展其应用技术研究，提出了一种基于压缩感知理论框架的图像融合算法，为图像融合技术的发展提供了新的发展思路。已发表论文6篇，其中国际期刊IEEE Transaction on Image Processing 1篇、国内期刊2篇、国际会议3篇，被SCI检索1篇，EI检索4篇。

中文关键词： 压缩感知；稀疏分解；lp范数优化；级联字典。

英文摘要： The recently emerged Compressive Sensing(CS) provides an alternative to Nyquist sampling for acquisition of sparse or compressible signals. CS exploits the surprising fact that sparse signals can be exactly recovered from a far less compressive measurements than the Nyquist samples. This project addresses two fundamental questions in CS theory, the sparse representation of images and the optimal recovery algorithms. Firstly, a new image sparse representation method based on concatenate dictionaries is proposed to enhance the representation performance for images with rich texture information. A new sparse domain selection strategy is proposed to overcome the drawback that image local structures cannot be adaptively reserved in certain domain. These new methods propose some new ideas for the image sparse representation based on dictionaries. Secondly, a new iterative reweighted algortihm and a penalty function based algorithm are proposed to solve the lp norm optimization problem for image recovery. The proposed algorithms are numerically testified for improving the quality of the reconstructed images. Finally, besides the basic theoretical work on CS, a new image fusion method based on the CS framework is proposed from the application aspect. This work extends the application of CS in traditional signal and image processing. Supported by this project, our team has published 6 articles on journals and conferences, including 1 article published by the famouse IEEE Transaction on Image Processing, 2 articles published on domestic journals and 3 articles published on conference proceedings. Among these articles, 1 article is indexed by SCI database, 4 articles are indexed by EI.

英文关键词： compressive sensing；sparse representation； lp norm optimization；concatenated dictionary.