基于结构字典学习的图像运动模糊复原

Single image motion blur restoration problem has been widely concern in the fields of image processing , computer vision and other fields. Currently, most of existing research in this respect are based on fixed dictionary or gradient model , but the study of corresponding theory for researching on dictionary learning based image restoration is still in the infancy stage. How to design a new dictionary learning method to achieve image deconvolution is the core issue to be addressed in this project . First, the project intends to predict the strong edge of the clear image, and the point spread function can be calculated in the Fourier domain fast and accurately; then intends to propose a structure dictionary learning methods, respectively, according to the different structural types of images to generate many types of dictionaries; finally, we plan to complete the structure dictionary learning in terms of image restoration theoretical research and practical applications. This proposal aims to design a structure dictionary learning which will be more conducive to image representation and understanding relative to traditional dictionary learning methods, thereby improving the quality of the image deconvolution . The significance of this research project is that it will increase our understanding of image structure analysis and dictionary learning theory, while providing new ideas to explore a wider range of motion blur in restoration techniques. This research will give us a outlook for the application of the structure dictionary learning methods in the fields of image processing and computer vision.

由于科学探测及军事科技的需要，单幅图像运动模糊复原成为一个迫切解决的实际问题。目前国内外关于这方面的研究成果基本上都是基于固定字典或者梯度模型的，基于字典学习的图像复原研究尚处于起步阶段。本项目将要解决的核心问题是设计新的字典学习方法来实现图像反卷积。首先，拟通过预测清晰图像中强边缘，快速、准确的计算出点扩散函数；其次，提出一种结构字典学习方法，利用图像中的不同结构类型分别生成多类字典；然后，利用学习得到的字典推动实际应用。预期设计出的方法更有利于图像表示和理解，提高图像反卷积质量。本项目的研究意义在于，为图像运动模糊复原这一个需要迫切解决的实际问题，提供一个理论支撑且实际可行的方案。

由于科学探测及军事科技的需要，单幅图像运动模糊复原成为一个迫切解决的实际问题。目前国内外关于这方面的研究成果基本上都是基于固定字典或者梯度模型的，基于字典学习的图像复原研究尚处于起步阶段。本项目将要解决的核心问题是设计新的字典学习方法来实现图像反卷积。首先，预测清晰图像中强边缘，快速、准确的计算出点扩散函数；其次，提出一种结构字典学习方法，利用图像中的不同结构类型分别生成多类字典；然后，利用学习得到的字典推动实际应用。本项目设计出的方法有利于图像表示和理解，提高图像反卷积质量。在项目执行期间，以第一作者发表期刊论文4篇和国际会议论文1篇，其中英文SCI两篇（一篇2区，一篇4区），中文EI期刊2篇，英文EI会议一篇，申请专利两项。本项目的科学意义在于，为图像运动模糊复原这一个需要迫切解决的实际问题，提供一个理论支撑且实际可行的方案。