基于GPU加速的几何纹理合成算法研究

发布时间：2018-10-24 14:18

【摘要】：在计算机虚拟现实中通常采用纹理数据模拟模型和场景的表面细节,使得模型或者场景的真实感大幅提升。纹理根据其数据组织结构不同可分为二维纹理和几何纹理。其中二维纹理以具有自相似性的二维图像数据为主,二维纹理合成具有合成速度较快,且内存占用小等特点;同时,由于二维纹理不支持遮挡、阴影、轮廓等重要的效果,二维纹理无法满足虚拟现实对表现形式越来越精细化的要求;几何纹理具有更丰富、细致的表现能力,在虚拟现实技术中得到了大量应用,大大提高了模型和场景的真实感。现有的几何纹理合成方法普遍存在计算量大、存储占用高并且需要大量的人工交互等问题,因此如何快速高效高质量的合成几何纹理成为计算机图形学的研究热点。首先,对纹理合成的研究背景、意义、及国内外研究现状做了分析,并且对二维纹理合成和几何纹理合成的一些经典算法进行详细的介绍和分析总结其优缺点。然后,对马尔科夫模型、像素点邻域、纹理块邻域以及邻域匹配原则等纹理合成基本理论进行了详细的阐述,结合现有的经典方法对二维纹理合成和几何纹理的一般步骤和方法进行详细的描述,以此分析得到影响纹理合成速度的原因是在于基于邻域匹配的纹理合成方法中大量的邻域搜索成了纹理合成速度的瓶颈。其次,对纹理合成加速技术从软件和硬件层面进行了分别介绍,根据邻域搜索存在的瓶颈问题,本文深入研究了相关的软件加速技术和GPU多线程并行加速技术。最后,提出了一种基于GPU加速的几何纹理合成方法,以解决几何纹理合成过程中的高计算量、高存储占用和高耗时等问题。通过采用GPU多线程并发技术设计并行加速算法,将串行的几何理纹理合成过程并行化,加速几何纹理合成。实验结果表明,本文算法不仅存储占用更小,而且能够在保证合成质量的同时,极大的降低几何纹理合成耗时。
[Abstract]:In computer virtual reality, texture data is usually used to simulate the surface details of the model and scene, which greatly improves the reality of the model or scene. Texture can be divided into two dimensional texture and geometric texture according to its data organization structure. Two-dimensional texture is mainly two-dimensional image data with self-similarity. Two-dimensional texture synthesis has the characteristics of faster synthesis speed and less memory footprint. At the same time, two-dimensional texture does not support occlusion, shadow, contour and other important effects. Two-dimensional texture can not meet the requirements of virtual reality for more and more refined representation; geometric texture has a richer and more detailed performance ability, has been a large number of applications in virtual reality technology, greatly improve the reality of the model and scene. The existing geometric texture synthesis methods have many problems, such as large amount of computation, high storage cost and large amount of manual interaction. Therefore, how to quickly and efficiently synthesize geometric texture with high quality has become a hot topic in computer graphics. First of all, the background, significance, and current situation of texture synthesis at home and abroad are analyzed, and some classical algorithms of two-dimensional texture synthesis and geometric texture synthesis are introduced in detail and their advantages and disadvantages are summarized. Then, the basic theories of texture synthesis, such as Markov model, pixel neighborhood, texture block neighborhood and neighborhood matching principle, are described in detail. Combined with the existing classical methods, the general steps and methods of two-dimensional texture synthesis and geometric texture are described in detail. The reason why the speed of texture synthesis is affected by this analysis is that a large number of neighborhood searches in the texture synthesis method based on neighborhood matching become the bottleneck of the speed of texture synthesis. Secondly, the acceleration technology of texture synthesis is introduced from the software and hardware levels. According to the bottleneck problem of neighborhood search, the related software acceleration technology and GPU multi-thread parallel acceleration technology are deeply studied in this paper. Finally, a new method of geometric texture synthesis based on GPU acceleration is proposed to solve the problems of high computational complexity, high storage footprint and high time consuming in the process of geometric texture synthesis. The parallel acceleration algorithm is designed by using GPU multi-thread concurrent technology to parallelize the serial geometric texture synthesis process and accelerate the geometric texture synthesis. Experimental results show that the proposed algorithm can not only save less memory, but also greatly reduce the time of geometric texture synthesis while ensuring the quality of synthesis.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP391.9

【参考文献】