基于形态分量分析的图像超分辨重建机理与算法研究

项目名称： 基于形态分量分析的图像超分辨重建机理与算法研究

项目编号： No.60802039

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 肖亮

作者单位： 南京理工大学

项目金额： 20万元

中文摘要： 近年来，稀疏表示和变分方法在图像超分辨问题中各自发展了系列新方法，但是如何将两者的优势有机地结合起来，设计更加高效并保持图像几何结构和纹理的图像超分辨率模型和算法是目前国际上的热点问题。本项目对基于形态分量分析（MCA）的图像超分辨重建模型和优化算法作了系统和深入的研究。创新性贡献包括：系统研究了稀疏表示字典对图像形态成分的统计不相干性和稀疏性机理，建立了符合类内强稀疏而类间强不相干的几何结构和纹理分量稀疏表示的参数化字典;提出了更广义MCA 框架下的稀疏性度量、非局部结构正则化和噪声先验度量模型，并在凸分析和稳健统计学思想下，提出了MCA 框架下联合处理图像放大、去噪、去模糊、去马赛克效应的超分辨重建的系列推广模型，提出了多形态稀疏正则化图像超分辨及其他反问题处理的新模型与方法，通过子空间投影\迭代收缩方法\算子分裂方法等,设计了超分辨率重建的系列优化算法。并且在形态分量框架,研究和设计了新的压缩传感模型及压缩传感去模糊等新的处理算法。本项目对于推动超分辨重建、图像理解、稀疏表示、压缩传感等理论发展都有重要意义. 本项目图像超分辨率重建的成果在航空、层析重建和遥感等具有重要应用前景。

中文关键词： 形态分量分析;稀疏表示;变分方法;图像超分辨

英文摘要： In recent years, although sparse representation and variational approaches have been used to create a series of new methods to deal with image super-resolution problem respectively, how to provide a better mechanism to combine them to design more efficiency image super-resolution algorithms, which can preserve the geometrical structure as well as texture, has become the international popular issues. The project focuses on the research on the Morphological Component Analysis (MCA) based image super-resolution reconstruction algorithms. The main contributions have been made as follows:1)We made investigations on the sparsity and statistical coherence among different sparse representation dictionaries, and then established parameterized dictionaries for sparse representation of image geometrical structure and texture components,respectively. Both of them have strong sparsity in intra-class and weak coherence between different component classes.2)Second, the measures of sparsity, prior models of non-local regularization and noise were proposed in a generalized MCA (GMCA) framework. 3）Furthermore, motivated by the convex analysis and robust statistical theory, we presented several image super-resolution models, which can be used to deal with image magnification, de-nosing, de-blurring, and demos icing simultaneously under the proposed GMCA framework. And we proposed multimorphology sparsity regularized model and algorithms for image super-resolution and related inverse problems.3)Finally, several novel muti-step iterative algorithms based on subspace projection,iterative shrinkage and operator splitting approaches have been proposed. Our project will have great significant in the theories of image super-resolution, image understanding,sparse representation and compressive sensing. Our contributions in image super-resolution reconstruction has a wide application prospects in aviation, remote sensing, medicine, chromatography imaging, etc.

英文关键词： Morphological Component Analysis;Sparse Representation;Variational Approach;Image Super-resolution