基于变分结构纹理分解的超分辨率图像复原方法研究

Via variational structure-texture decomposition,the low resolution image decomposes into meaningful image components. The super-resolution restoration for single image transforms into structure component reconstruction and texture component reconstruction. Furthermore, the interpolation technique and minimization energy function is used to handle these two sub-problems. Eventually, a novel algorithm for super-resolution restoration for single image is obtained. The main contents of this project include:(1)Variational structure-texture decomposition model, which mainly researches on using total variation and oscillationg function to describe structure and texture components seperately, and contructs the variational structure-texture decomposition model for low resolution image.(2)Variational structure-texture decomposition algorithm, which mainly focuses on changing the G space with proper other norm spaces, and transforms the variational structure-texture decomposition model into solvable partial difference equation.(3)Structure component and texture component reconstruction, which mainly studies on adopting the interpolation technique and minimization energy function method to implement the super-resolution reconstruction for structure and texture components. The research in this project is going to further expand the application of variational structure-texture decomposition theory in image processing realm, and promote the technology and approach innovation for super-resolution image restoration.

本课题通过变分结构纹理分解实现低分辨率图像“有意义”的图像分解，将单幅图像超分辨率复原问题转化为结构分量重建、纹理分量重建两个子问题，并利用插值技术及最小化能量泛函分别实现结构及纹理分量重建，得到用于单幅图像超分辨率复原的新算法。本课题的主要研究内容包括：（1）变分结构纹理分解模型,主要研究利用全变差和振荡函数分别描述图像的结构和纹理分量，建立低分辨率图像的变分结构纹理分解模型。（2）变分结构纹理分解算法，主要研究利用合适的范数空间替换G空间，将变分结构纹理分解模型转化成可解的偏微分方程。（3）结构分量及纹理分量重建，主要研究采用插值技术和最小化能量泛函方法实现结构分量及纹理分量的超分辨率重建。本课题的研究将进一步拓展变分结构纹理分解理论在图像处理领域中的应用研究，推动超分辨率图像复原技术和方法的革新。

图像结构纹理分解能够把图像分解为若干“有意义”的信息分量，用于图像的进一步分析和处理。基于对成像过程中的图像降质成因的研究，本项目引入变分结构纹理分解理论，构造变分结构纹理分解模型，研究具体的数值计算方法及其在图像超分辨率复原方面的应用。主要研究内容包括：（1）变分结构纹理分解模型,利用全变差和振荡函数分别描述图像的结构和纹理分量，建立图像的变分结构纹理分解模型；（2）变分结构纹理分解算法，利用合适的范数空间替换G空间，将变分结构纹理分解模型转化成可解的偏微分方程；（3）结构分量及纹理分量重建，采用插值技术和最小化能量泛函方法实现结构分量及纹理分量的超分辨率重建。相关成果申请发明专利8项；发表EI收录期刊论文3篇，CSCD期刊论文5篇，1篇论文投稿中。本课题的研究成果进一步拓展变分结构纹理分解理论在图像处理领域中的应用，丰富超分辨率图像复原的技术和方法。