结合灰色建模与多字典稀疏表示的图像超分辨率研究

项目名称： 结合灰色建模与多字典稀疏表示的图像超分辨率研究

项目编号： No.61273260

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 唐英干

作者单位： 燕山大学

项目金额： 78万元

中文摘要： 高分辨率图像能够提供更多、更丰富的细节和敏感的颜色转变，使得图像内容更加容易理解。本项目以灰色建模和稀疏表示理论为指导，研究多字典自适应稀疏表示的图像超分辨率的技术。首先，以图像分割和机器学习理论为基础，提取图像块包含空间位置在内的广义特征，通过图像块拓扑空间自组织聚类，发掘图像中所包含的不同结构模式。其次，针对图像不同的结构模式，以稀疏向量的非凸项为优化目标，研究图像块多字典自适应稀疏表示方法和理论。应用灰色理论，建立刻画图像块局部相关性的灰色模型，将这一模型与图像块多字典稀疏表示相结合，构建图像超分辨率重建的优化目标函数。最后，设计群智能优化算法与传统算法相结合的混合优化算法，有效地求解所构建的图像超分辨率优化目标函数。本项目是一个多领域的交叉研究，不仅能为图像超分辨率提高新的方法和思路，也能推动机器学习、信号稀疏表示和优化理论的进一步发展。

中文关键词： 图像预分割；聚类；字典学习；智能优化；深度学习

英文摘要： High resolution image can provide more image details and sensitive color change, making the image contents more easily understandable. This project studies image super resolution techniques based on grey modeling and spare representation theory. Firstly, on the basis of image segmenation and machine learning, by extrating the generalized features of image patches, we propose the image patches clustering methods to explore the different structure modes in the image. Then, for each different structure modes, the adaptive multi-dictionary sparse representation theory of image patches is studied using non-convex regularization term for sparse coefficient vector. Accoring to grey theory, the local relationship between image patches are modeled uisng grey modeling method. Integrating the grey local model and the multi-dictionary sparse representation of image patches, the optimization objective functions of image super resolution are built. Finally, the hybrid optimization methods combining swarm intelligence and conventional optimization are proposed to effectively optimize the built image super resolution objective function. The study of this project include multi-fields, it can not only present new idea and methods for image super resolution, but also promote the development of machine learing,signal sparse respre

英文关键词： image pre-segmentation；cluster；dictionary learning；intelligence optimization；deep learning