【作者】 薛峰；

【导师】 张佑生；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2006， 博士

【摘要】 基于样图的纹理合成技术(Texture Synthesis from Samples,TSFS)是近年来发展起来的一项新的纹理生成技术,它既克服了传统纹理映射技术可能带来的纹理接缝和纹理扭曲的缺陷,也避免了过程纹理合成(Procedural Texture Synthesis,PTS)参数选择的繁琐过程,因而成为计算机图形学、计算机视觉领域的研究热点之一。 本文首先介绍基于样图的纹理合成技术的国内外研究现状,对TSFS的基本模型、原理和经典算法进行了详细地介绍和讨论。在此基础上,本文对TSFS技术中若干关键问题展开了深入的研究,主要研究内容如下: 1.对基于样图的纹理合成技术的研究历史进行较全面的回顾,对其中的经典算法进行了分类,并总结它们的优缺点。 2.利用纹理图像及其子图像的直方图的相似性,提出一种Wei&Levoy算法中L邻域最佳尺寸的自动选取算法,避免因为L邻域尺寸选取不当而引起的合成时间的增加和合成质量的降低。 3.提出一种基于灰度辅助纹理和自组织特征映射的纹理合成算法:(1)使用纹理图像的灰度图像作为辅助纹理加速纹理合成过程;(2)提出一种改进的SOM神经网络的向量构造、学习、和测试方法,对纹理邻域集合进行分类,并使用分类结果进行纹理合成。 4.对二维实时纹理合成算法展开了深入地研究,主要内容包括:(1)介绍基于最大梯度和灰度相关匹配的Image Quilting加速算法,基本达到实时性要求;(2)提出一种新的纹理贴块——s-Tiles生成算法,并使用s-Tiles实时合成高质量的纹理。此外,还提出一种基于s-Tiles的应用——“纹理句子”的实时生成,达到了很好的效果。 5.研究三角网格曲面的快速纹理合成。首先由输入样本纹理使用随机顺序纹理合成算法生成一个新的用于曲面纹理合成的样本纹理,然后提出一种基于更多还原

【Abstract】 Texture Synthesis from Sample(TSFS) is one of the new techniques for texture generation in recent years. TSFS can eliminate some defects of Texture Mapping technique, such as texture seams and texture distortions, moreover, it can avoid the boring processes of parameter optimization of Procedural Texture Synthesis(PTS). TSFS has quickly become one of the research hotspots in computer graphics and computer vision recently.In this dissertation, we focus on the study on TSFS technique. In the first part of this dissertation, we review the research results of TSFS and introduce its basic model, principle and typical algorithms. The main part of this dissertation concerns some key problems on TSFS, including:1. We review the research history of TSFS and classify the typical TSFS algorithms into different categories and conclude their advantages and disadvantages.2. Based on the histogram similarity between texture image and its sub-images, we present a new method for determining the optimal size of L neighborhood in Wei&Levoy texture synthesis algorithm.3. We provide a new texture algorithm based on gray image of input texture and Self-Organized Feature Maps(SOM). Firstly, we use the gray image of input texture as an aiding guidance to accelerate the process of texture synthesis. Then, a new method for constructing, training and testing SOM vectors was proposed, which is used to classify the vectors constructing from L neighborhood of output pixels. We use the classified results to speed up the output texture generation.4. Real-time texture synthesis is well researched in chapter 4: (1) we present a new method of image mosaic based on maximum and intensity correlation, and we use this improved image mosaic method to accelerate Image Quilting algorithm. (2) Anew texture tiles—s-Tiles generating method is proposed, which can be tiled into arbitrary size of output in real-time. We also present a novel application of s-Tiles—"Sentence Tiling".5. A new technique of texture synthesis on arbitrary surfaces is proposed. We firstly generate a new tillable texture from input using Wei’s Random Order Texture Synthesis algorithm, which will be used as sample for consequent texture synthesis on surfaces, then, we synthesis "texture extension" method and template matching method to calculating texture coordinates for each triangle according the number of its synthesized neighbor. Our texture synthesis method on arbitrary surfaces can produce better results in little time than previous methods.6. In the sixth chapter, we provide a novel application—geological folds simulation using orientation-controlled textures synthesis algorithm. The experimental results show that our simulation algorithm is automatic, feasible and easy to be extended to any other shaped folds.更多还原