【作者】 刘安平；

【导师】 梁金星；

【作者基本信息】 武汉纺织大学 ， 软件工程（专业学位）， 2023， 硕士

【摘要】 我国拥有大量的彩色文物艺术品,它们承载着我国灿烂文明和历史文化,具有非常重要的历史价值,而颜色信息是文物自身历史文化属性表达与传递的重要载体。由于受到外界因素的影响,文物发生严重的病害问题,致使其可显现的价值信息发生改变或丢失。本文针对计算机辅助的数字化修复技术中存在的古代壁画图像分割及颜色高保真修复问题,提出一种基于超像素的图像优化分割方法和基于光谱替换的颜色高保真数字化虚拟修复方法,取得了较为理想的结果。1)针对实现古代壁画图像精细化分割问题。古代壁画的特点是颜色种类丰富,并且不同颜色之间彼此融合,再加上壁画颜色出现褪变、脱落、磨损等严重病害,使得传统分割算法在像素级水平下单一使用来分割目标图像会出现过分割现象,而在超像素水平下会出现欠分割的问题,要实现精细化分割目标基本采取传统分割方法与人机交互结合的分割模式。基于深度学习的图像分割方法需要大量的训练样本数据集,而制作古代壁画图像训练样本难。针对上述存在的问题,提出了基于超像素的古代壁画图像优化分割方法,实现了古代壁画图像的精细化分割目标。2)针对颜色高保真数字化修复问题。传统的古代壁画图像颜色数字化修复技术受制于色度学所固有的“同色异谱”问题,因此,以色度学为基础的颜色数字化修复技术,都是有条件的颜色处理,在艺术品颜色高保真数字化采集和复制再现等应用中,有很大的局限性。针对古代壁画图像颜色高保真修复问题,采用光谱反射率替换实现颜色的高保真采集和再现,并在基于光谱的褪变色模型基础上实现古代壁画图像颜色的褪变色模拟,实现了古代壁画图像颜色高保真数字化修复。研究以敦煌模拟壁画和实际壁画为对象,对方法的有效性和优越性进行了检验。并依据提出的分割算法和光谱反射率褪变色模型即光谱褪变色模型,对仿真壁画含铅颜料的褪变色过程进行了模拟,实现了基于光谱替换的古代壁画图像颜色高保真虚拟演变。实验结果表明,本文提出的算法能够取得边界清晰的精细化分割结果,并在优化分割方法的基础上进行了颜色高保真数字化仿真实验,整体上为褪变色古代壁画图像的颜色高保真数字化修复提供了方法支撑。更多还原

【Abstract】 China has a large number of coloured cultural relics and works of art,which carry the splendid civilisation and history and culture of our country and have significant historical value,while the colour information is an essential carrier for the expression and transmission of the historical and cultural attributes of the relics themselves.Due to the influence of external factors,cultural relics have suffered serious disease problems,resulting in the change or loss of information about their apparent value.In this paper,we propose a super-pixel-based image segmentation method and a spectral replacement-based colour high-fidelity digital virtual restoration method to address the problems of image segmentation and high-fidelity colour restoration of ancient wall paintings in computer-aided digital restoration technology and achieve more satisfactory results.1)To address the problem of achieving fine-grained segmentation of ancient mural images.Ancient frescoes are characterised by a rich variety of colours and the fusion of different colours with each other,coupled with fading,peeling,wear and tear and other serious damage to the colours of the frescoes,making the traditional segmentation algorithm at the pixel level a single use to segment the target image will result in over-segmentation,while at the superpixel level will result in under-segmentation problems,to achieve the goal of refinement of segmentation basically take the traditional segmentation methods and humancomputer interaction In order to achieve finer segmentation of the target,traditional segmentation methods and human-computer interaction are basically adopted.Deep learningbased image segmentation methods require a large number of training samples,and it is difficult to produce training samples for ancient wall painting images.To address the above existing problems,a super-pixel-based optimised segmentation method for ancient mural images is proposed to achieve the goal of refined segmentation of ancient mural images.2)For high-fidelity digital restoration of colour.Traditional colour restoration techniques for ancient fresco images are limited by the inherent "homochromatic" problem of chromaticity,therefore,colour restoration techniques based on chromaticity are conditional on colour processing and have significant limitations in applications such as high-fidelity colour digital capture and reproduction.In order to address the problem of high-fidelity colour restoration of ancient wall painting images,we use spectral reflectance replacement to achieve high-fidelity colour acquisition and reproduction and realise the faded colour simulation of ancient wall painting images based on a spectral-based faded colour model,realising high-fidelity colour digital restoration of ancient wall painting images.The validity and superiority of the methods are examined using Dunhuang simulated murals and actual murals.Based on the proposed segmentation algorithm and the spectral reflectance fading model,i.e.the spectral fading model,the fading process of lead-containing pigments in simulated murals is simulated,and a high-fidelity virtual evolution of the colour of ancient mural images based on spectral replacement is achieved.The experimental results show that the algorithm proposed in this paper is able to obtain refined segmentation results with clear boundaries,and the colour high-fidelity digital simulation experiments are carried out based on the optimised segmentation method,which provides overall methodological support for the colour high-fidelity digital restoration of faded ancient wall painting images.更多还原