【作者】 王克刚；

【导师】 耿国华；

【作者基本信息】 西北大学 ， 计算机软件与理论， 2009， 硕士

【摘要】 文化遗产保护数字化已成为信息技术与考古学交叉的新兴研究方向。瓷器类物品是中国考古发现的重要元素之一,而瓷器易碎的特点导致遗存至今的古瓷器破损很多,考古过程所发掘的大量古瓷碎片往往混杂在一起,瓷器类物品的人工修复、分类管理等工作过程中面临很多困难。本文针对瓷片数字图像特征的分析与应用进行研究,利用色彩、纹理、纹饰形状等特征实现瓷片图像的模式分类,为古瓷碎片的类别自动划分提供辅助手段。本项研究得到了国家自然科学基金的支持,主要研究进展如下:(1)研究提出了一种简便易行的图像自适应平滑与增强算法,实现了图像区域内部平滑和边缘增强的同步处理,强化了瓷片图像的视觉感观特征。(2)改进了基于RGB空间上的色彩对聚类算法,使算法的时间复杂度由O(n~2)提升到O(nlogn);定义了基于HSI颜色空间的一种非均匀色彩量化方法,对瓷片图像色彩特征进行提取并应用于瓷片分类,取得了好的效果。(3)从结构化方法、统计性方法、频域变换方法三个方面实现了瓷片图像的基元纹理以及灰度共生矩阵、自相关函数、边界频率、二维直方图、Gabor变换等纹理特征的提取。提出了瓷片图像色彩—纹理特征的提取模型与具体方法,对色彩信息和纹理信息进行了有机融合,与其它方法相比较,瓷片分类的正确率得到了大幅提升。(4)研究提出了一种KFCM彩色图像分割方法。通过提取彩色图像的色彩—纹理基元特征,并引入核函数思想,实现了彩色图像的有效分割。利用该方法,得到了瓷片纹饰区域的准确划分,为纹饰形状特征的有效表示提供了良好基础。(5)应用支持向量机(SVM)分类方法,对瓷片图像的色彩、纹理、纹饰形状等多种特征的分类识别能力进行了测试、分析和比较。Matlab与VC结合,开发了基于数字图像特征的古瓷碎片分类原型系统,为古瓷碎片的自动分类提供了辅助平台。更多还原

【Abstract】 The research on cultural heritage protection with information technology, which belongs to intersectant research area of computer science and archeology, has become a hotspot in recent years. Chinese porcelain is one of important elements in archaeological discoveries, and porcelain has the characteristics of brittle, so many remains are discovered in the form of pottery fragments. During archaeological excavations, because a large number of porcelain shards usually mix up on the spot, there are a lot of difficulties in the artificial porcelain repair and classification management. In this paper, the extraction and application of the digital images characteristics has been researched penetratingly, and shards digital image mode is classified rightly according to the features such as color, texture and ornamentation shape in order to provide the assistant means contributing to shards automatic classification. This research is supported by national natural science foundation. The major advances are as following:(1) The paper proposes a simple adaptive smoothing and enhancement algorithm of digital image processing, which can realize the coinstantaneous processing of interior smoothing and edge enhancement of target area in image, strengthening the visual effect on the processed image.(2)This paper improved the algorithm of color pair wise clustering based on RGB so that the time complexity is promoted from O(n~2) to O(nlogn). Meantime, a new color un-equidistribution quantization method in HSI color space is defined to extract the color feature and apply to shards’ classification. As a result, the method can achieve good effect.(3)In this paper, the extraction technology has been research on three aspects: structuring method, statistical method and frequency-domain transformation method. Methods based on primitive texture, gray level co-occurrence matrix, autocorrelation function, edge frequency, two-dimensional histogram and Gabor transform are realized. New color-texture model and method for the feature extraction are proposed to integrate color information and texture information effectively. Comparing with the existing methods, the new method can promote accuracy of shards’ classification remarkably.(4) This paper puts forward a new segmentation algorithm for color image,which is called KFCM, that is based on image’s color-texturetexton feature, kernel function and fuzzy clustering method. The algorithm can realize the effective segmentation for color image. Through the algorithm, shard ornamentation shape can be divided accurately so that the shape feature of ancient shards may be presented effectively..(5) Using support vector machine classification technology, the recognition performance about ancient porcelain shard’s color, texture and ornamentation shape feature has been tested, analyzed and compared. Combing Matlab with Vc, an ancient shards classification prototype system based on the characteristics of digital image is developed as a assistant platform for automatic classification of ancient porcelain.更多还原