基于稀疏建模的褪色/变色石窟壁画彩色图像超分辨率重建方法研究

针对褪色/变色石窟壁画彩色图像超分辨率重建这一科学问题，本项目拟基于稀疏表示理论，以麦积山石窟壁画为对象，重点研究彩色壁画图像多通道特征信息联合提取及超分辨率重建模型构建，力争在超分辨率重建字典设计、稀疏表示系数快速估计、超分辨率重建全局优化算法以及褪色/变色壁画图像色彩信息还原等方面实现创新。主要内容包括：利用可控迭代核回归模型设计彩色壁画图像各通道形态特征信息提取方法；基于局部约束四元群稀疏框架构造多分量过完备字典；构建四元卷积神经网络实现稀疏表示系数快速估计；采用双重异性加权保持策略设计色彩相关性约束正则项；根据多通道信息联合优化思想设计重建图像全局优化模型及数值算法；基于VGG-Net模型探索高分辨率重建彩色壁画图像色彩还原方法。本项目旨在促进石窟壁画图像的保护与展示研究，充实彩色图像稀疏表示理论，并为超分辨率重建技术在文物保护等领域的应用提供新思路。

Color super-resolution reconstruction of faded or discolored grotto murals is an urgent research work. The project intends to study color image super-resolution reconstruction of Maijishan grotto murals based on sparse representation theory. The project focuses on jointing extraction of multi-channel features information and designing of super-resolution reconstruction model. And the project aims at how to design effective sparse dictionaries, how to estimate sparse codes, how to design the global optimization algorithm of super-resolution reconstruction model and how to restore color information of faded or discolored murals. The project consists of several parts. Firstly, the controllable iteration kernel regression model is used to extract features of color mural images. Secondly, Quaternion-based multi-component locality-constrained group sparse dictionaries are designed. Thirdly, a convolutional neural network model is utilized for fast sparse coding. Then considering color channels correlation, several constraint regularization terms are designed. Moreover, the numerical algorithm of global optimization is proposed to remove color artifacts. Last but not least, a color restoration method based on the VGG-Net is proposed. Consequently, the purpose of this project is to promote the research and protection of grotto murals, enrich the color sparse representation theory, and provide an innovative idea for the heritage preservation research and application by super-resolution reconstruction technology.