基于线描图智能化生成的计算机辅助壁画修复

【作者】 周宇；

【导师】 鲁东明；

【作者基本信息】 浙江大学 ， 计算机应用， 2003， 硕士

【摘要】 敦煌壁画是我国的珍贵历史文化遗产之一，敦煌壁画艺术是中国古典艺术文明中的杰出代表。然而，由于自然风化的破坏以及重大自然灾害的威胁，这个人类艺术的宝库变得非常脆弱。因此，保护敦煌壁画已经成为一个刻不容缓的任务。数字化壁画保护修复工作能够将壁画信息永久保存，并能够利用计算机对壁画进行虚拟修复、辅助进行壁画保护、辅助进行壁画临摹、对壁画病害过程进行虚拟演变、实现壁画真实感虚拟展示等丁作。数字化壁画保护修复为壁画的物理保护修复过程提供了充足的科学依据和测试环境，将保护工作的危险性降至最低。本文提出了壁画数字化保护与虚拟修复的一种新方法。根据敦煌艺术家对壁画修复过程的研究、颜料成分的研究、艺术家积累的经验知识以及保存得比较完好的一些壁画，综合利用数字图像处理、计算机图形学、人工智能等技术，实现壁画虚拟修复。计算机辅助壁画修复是在计算机上把一些脱落、变色、褪色的数字化敦煌壁画修复成符合敦煌艺术家认同的没有受到侵害时的图像。虽然传统的图像处理技术提出了很多经典的图像复原的方法，但是我们也发现这些方法用在壁画虚拟修复领域有一定的局限性。本文根据艺术家对敦煌壁画修复性临摹过程和计算机图形图像技术，提出了一套针对壁画的计算机辅助修复性临摹方法。首先我们根据图像分割和边沿提取算法对输入的壁画图像进行区域的提取和分割处理。然后我们根据敦煌壁画内容的层次化组织，建立了基于特征点的图像元素样本库。在建立的样本库的基础上，我们根据专家知识对壁画边沿缺失严重的部分进行替换。之后再交互的对替换后的图像边沿轮廓线进行精细的调整得到线描图。最后，我们采用具有特定风格的笔刷模型和纹理合成技术在线描图上进行着色修复，得到计算机辅助生成的壁画修复图。目前本文提出的基于线描图智能化生成的图像处理方法，已经应用于计算机辅助敦煌壁画虚拟修复领域，成功地开发了原形系统。研究论文在国内外各种期刊、会议中发表。对于目前研究中的一些不完善的地方，有待以后进一步的探索。更多还原

【Abstract】 Dunhuang murals are the key historical and cultural relics of Chin4 which are representativecollections of ancient Chinese art. Because Dunhuang Grotto locates the Gobi area havingatrOcious weather, muras are threatened by many diseases and tUm tO be very frail. So it is anurgent mission to protect these murals. Digital protecting and virtual restoring murals exploits newmethods and proposes effective tools. Digitalized methods are efficient and repeatable, and willnot do any harm to the relics. They break through the limitations of physical methods and presenta new area of the protection and restOtation of murals.This thesis presents a new method fOr digitalized protection and virtual restoration of murals.According to the research of muraI pigment, the experiential knowledge of artists andarcheologists and some well-protected murals, synthetically applying chromatics, graphics, imageprocessing and AI. murals can be virtually restored. Virtual restoration is to restore somecolorchanged, color-faded, or part-shedded murals to its original splendid appearance.Although traditional image processing covers many classical image recovery techniques, we findthese they are limited in mural image recovery because they did not consider the knowledge ofthe domain, the physical and chemical models of murals, or the restrictions of the domain.Therefore, based on the analyzing of the process of mural restoration by artists, we propose aseries of methods aim at computer aided mural restoration copying. We first aPply imagesegmentation and edge detection to the source mural image. Then we constructed an example basewhich can be retrieved by the control points of the examples. We rep1ace the heavily damaged partof the mural with the matched examples in our database. After the interactive revise process, weget the line drawing image of the source mural image. Finally, we can color the line drawing bydifferent style brushes and get the restoration target mural image.At present, this new image processing techniques based on the intelligent line drawing generationhave been aPplies to Dunhuang mural virtual restoration. The archetypal system has beendeveloped successfully. Several papers were published or admitted in sorts of periodicals andconferences at home and abroad. And we wiIl continue our research on some imperfect featuresstill exiSting in our work in the future.更多还原