青铜器纹饰特征提取算法研究与实现

发布时间：2018-06-25 19:09

本文选题：青铜器修复 + 浮雕提取　；参考：《合肥工业大学》2017年硕士论文

【摘要】：青铜器是我国的文化瑰宝。然而,随着我国文物事业的快速发展,积累了大量残破青铜器,目前青铜器的修复还主要采用传统技术。传统青铜器修复不仅难以保证修复质量,而且时间花费长。因此,将现代数字化技术与传统青铜器修复技术相结合,发展青铜器修复技术,是解决日益凸现的青铜器修复难题的必由之路。青铜器的表面纹饰和铭文都可以看成是浅浮雕特征,在修复过程中要保证碎片之间纹饰、铭文的连贯性。因此,研究浮雕纹饰的提取对青铜器的数字化修复是至关重要。本文着重研究基于三角网格模型的浮雕纹饰提取算法,为实现青铜器数字化修复奠定基础。主要研究内容及取得的成果如下:(1)实现了一种基于三维曲面循环引导的浮雕提取算法。该算法将浮雕上的纹饰看成是附着在基曲面上的起伏或内嵌部分,通过准确地估算基曲面的法矢量,进而重构出基曲面的空间位置,然后将纹饰相对于基曲面的偏移量计算出来,最终使用一种聚类算法将浮雕过滤出来。实验结果表明:在基曲面占优的条件下,该算法能生成高质量的基曲面,并对浮雕特征具有较好的提取效果。(2)实现了一种基于高度估计的浮雕提取算法。该算法认为浮雕由两部分组成:基曲面和定义在该基曲面上的高度函数。由于浮雕所位于的基曲面是未知的,这使得基曲面上的高度函数分析非常困难。在没有准确估计基曲面情况下,该算法能够较为准确估计出高度函数,然后采用高斯混合模型分析该高度函数来自动确定分割阈值,从而提取出浮雕纹饰特征。实验结果表明该算法对于高浮雕的提取效果较好。(3)提出并实现了一种基于特征增强的浮雕尖锐特征提取算法。首先,为了减少网格均匀度对特征提取的不利影响,提出了一种加权投影距离;其次,为了增强尖锐特征,提出一种逆双边滤波算法,并利用该算法获得反锐化掩膜法,增强投影距离;最后,采用Otsu算法自动确定分割阈值,依据该阈值把网格顶点分为特征点集和非特征点集。实验结果表明,该算法能自动、高效的提取浮雕的尖锐特征,对于纹饰复杂青铜器有较好提取效果。
[Abstract]:Bronze ware is the cultural treasure of our country. However, with the rapid development of cultural relics in China, a large number of broken bronzes have been accumulated. At present, the restoration of bronzes is mainly based on traditional techniques. Traditional bronze restoration is not only difficult to guarantee the quality of restoration, but also takes a long time. Therefore, combining the modern digital technology with the traditional bronze restoration technology and developing the bronze restoration technology is the only way to solve the increasingly prominent bronze restoration problem. The surface ornaments and inscriptions of bronze ware can be regarded as shallow relief features. In the process of restoration, it is necessary to ensure the continuity between fragments and inscriptions. Therefore, it is very important to study the extraction of embossed ornaments for the digital restoration of bronze ware. This paper focuses on the extraction algorithm of embossed ornaments based on triangular mesh model, which lays a foundation for the realization of bronze ware digital restoration. The main research contents and achievements are as follows: (1) A relief extraction algorithm based on 3D curved surface cycle guidance is implemented. In this algorithm, the embossed ornaments are regarded as the undulating or embedded parts attached to the base surface, and then the space position of the base surface is reconstructed by accurately estimating the normal vector of the base surface. Then, the offset of the ornaments relative to the base surface is calculated, and a clustering algorithm is used to filter out the relief. The experimental results show that the algorithm can generate high quality base surfaces under the condition that the base surface is dominant, and it has a good extraction effect for relief features. (2) an algorithm for extracting relief based on height estimation is implemented. The algorithm considers that the relief consists of two parts: the base surface and the height function defined on the base surface. Because the base surface on which the relief is located is unknown, it is very difficult to analyze the height function on the base surface. In the absence of accurate estimation of the base surface, the algorithm can estimate the height function more accurately, and then analyze the height function by using the Gao Si mixed model to determine the segmentation threshold automatically, and extract the embossed decorative features. The experimental results show that the algorithm is effective for the extraction of high relief. (3) A feature enhancement based sharp feature extraction algorithm is proposed and implemented. Firstly, in order to reduce the adverse effect of mesh uniformity on feature extraction, a weighted projection distance is proposed. Secondly, in order to enhance the sharp feature, an inverse bilateral filtering algorithm is proposed, and the anti-sharpening mask method is obtained by using this algorithm. Finally, Otsu algorithm is used to automatically determine the segmentation threshold, according to which the mesh vertices are divided into feature point set and non-feature point set. The experimental results show that the algorithm can automatically and efficiently extract the sharp features of relief carving, and has a good effect on the extraction of complex bronzes with decorative patterns.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;K876.41

【参考文献】