带多种属性的三维模型的简化与压缩技术研究

发布时间：2018-03-12 23:00

本文选题：三维模型　切入点：多属性　出处：《山东大学》2017年博士论文　论文类型：学位论文

【摘要】：三维模型的描述形式有点云、三角形网格、四边形网格等,组成模型的元素也有很多属性,如基于面片(surfel)表示的点云模型有位置、法向、颜色和半径等属性,网格模型有顶点位置、颜色或纹理坐标等属性。其中,纹理图片、法向贴图等也是表现三维网格模型特征的重要数据。这些属性和数据直接影响着三维模型的表面特征。在三维模型的应用领域中,如三维电影、游戏、医疗、科学可视化等,真实感的要求越来越高。但是,有些系统受到存储空间、带宽和计算能力的限制,所以需要对高精度的三维模型进行压缩或者简化。模型数据的压缩和简化是模型处理的重要内容,一直在被广泛而深入的研究。模型压缩是一种有效的减少存储和传输的数据的技术。通常,模型的不同属性数据具有不同的空间分布,所以针对不同属性数据的特点制定不同的编码和解码算法,有利于增大压缩率、提高压缩效率。由于在编码时要对浮点数进行量化,所以即使是无损压缩,解码后的数据与压缩前数据可能不完全一致。因此,在压缩过程中,通常需要在模型质量和软硬件环境的限制之间做权衡。除了对三维模型进行压缩外,也可以通过简化技术来减少三维模型存储、传输和处理时的数据量。与模型压缩技术不同的是,模型简化不需要对模型的数据进行编码,而是在尽量保持模型原始外观的情况下去除一些不重要的信息,这也导致模型的简化算法一般都是有损算法。在通常情况下,简化是预处理的过程,需要根据不同的应用环境选择相应质量的简化模型。一种常用的策略是,为同一个模型预处理生成一系列不同质量的细节层次(LODs),并根据不同的应用环境选择相应的质量的LOD。虽然一些压缩算法可以在渐进编码和解码过程中产生一系列不同细节水平的LODs,但是这些算法的焦点通常是如何高效地编码,而不是如何提高简化模型的质量。虽然三维模型的压缩和简化技术得到了比较广泛的研究并且发表了大量的相关算法,但仍然不能全面并有效地解决目前模型应用中存在的问题。第一,已发表的模型压缩算法,普遍对低比特下模型压缩关注的比较少,而且其中一些算法还受模型拓扑结构的限制。第二,目前大多数的点云模型简化的技术能够较好地利用模型的几何信息(如位置和法向信息),却很少对纹理信息(如面片的颜色属性)进行充分的分析利用。并且,大多数的点云模型简化算法是基于局部启发式的简化,而不是系统地全局优化。类似地,早期的三维网格模型简化算法也致力在减少的顶点数目情况下保持模型几何特征。后来提出的许多简化算法同时最小化几何误差和纹理偏差,也能够较好的保留带纹理的模型的外观。但是,这些算法并没有充分地利用纹理图像内容。第三,尽管高精度纹理图像可能比模型的网格数据占用更多的存储、带宽或计算资源,纹理图像简化却很少得到研究人员的关注。本文全面研究了目前三维模型压缩和简化技术的相关工作,综合考虑了其中存在的问题,提出了:一种基于GLA(标准劳埃德算法)的点云模型压缩技术。该技术扩展了 GLA算法,对聚类中的点数进行控制,并贪婪地优化编码顺序。该算法在低比特压缩时具有明显的优势,并且算法具有通用性,适用于压缩任意拓扑结构的点云模型。一种基于邻域聚类合并的点云模型简化技术。该技术通过迭代地挑选并合并具有最小成本的局部邻域,直到达到期望的数据简化率。算法基于人类对视觉重要度的感知,利用几何和纹理信息的局部变化,计算局部邻域合并的成本;基于高斯差分的纹理锐化技术用于增强纹理细节。此外,还把算法扩展到了大规模模型简化。该技术能够较好地保留简化模型表面的几何和纹理特征。一种协作简化三角形网格及其纹理图像的技术。纹理图像用于引导网格简化,简化的网格用于辅助纹理图像简化。在简化模型的几何数据时,会充分考虑简化部分的纹理特征,贪婪地选择对模型形状和外观影响最小的区域进行简化;在简化纹理图片时,会考虑局部颜色的复杂性,并充分利用简化模型的几何信息。该技术在减少模型网格和纹理图片数据的同时,能够较好地保留简化模型表面的几何和纹理特征。另外,该技术的框架非常通用,各种基于迭代的网格简化的算法经过适当的调整,都可以应用于该框架。
[Abstract]:The 3D model described in the form of a point cloud, triangular mesh, quadrilateral mesh, model elements also have many properties, such as surface based on point cloud model (surfel) representation of the position, direction, color and radius attributes, mesh vertex position, color or texture coordinates and other properties. Among them, texture mapping method, the data is important to performance of 3D mesh model feature. These attributes and data directly affects the surface characteristics of the 3D model. In the field of 3D models, such as 3D movies, games, medical, scientific visualization, and realistic higher. However, some systems are stored space, bandwidth and computing capacity constraints, so the need for 3D model of high precision compression or compression and simplification. The simplified model of data processing is an important content of the model, has been widely and deeply. The model is an effective compression to reduce the storage and transmission of data technology. Usually, different attributes of data model with different spatial distribution, so according to the characteristics of the different attributes of the data set of encoding and decoding algorithms, can increase the compression rate, improve compression efficiency. Due to quantization in encoding on the float, so even a lossless compression, the decoded data and compressed data before may not be entirely consistent. Therefore, in the compression process, usually need to do the tradeoff between model quality and the limits of the software and hardware. In addition to the compression of 3D model can be reduced by simplifying the 3D model storage technology. When the amount of data transmission and processing. Unlike model compression technology, the simplified model does not need to model the data encoding, but try to keep the original appearance of the situation on the model In addition to some important information, which leads to a model simplification algorithm is generally lossy algorithm. In most cases, simplification is the process of pretreatment, according to different application environment selection and simplified model of quality. A common strategy is to generate a series of different levels of detail for the same quality a model of pretreatment (LODs), and according to the different application environment to select the appropriate quality LOD. although some compression algorithms can produce a series of different level of detail LODs in progressive encoding and decoding process, but the focus of these algorithms is usually efficient encoding, rather than how to improve the quality of the simplified model. Although the three-dimensional model of compression and simplification technology has been widely studied and published a large number of algorithms, but still can not fully and effectively solve the existing problems in the application of the model. First, the published model compression algorithm, generally focus on low bit rate compression model is relatively small, and some of the algorithms is also affected by the model topology. Second, most of the point cloud simplification technology can better utilize the geometry information (such as position and normal information), but rarely on the texture information (such as patch color attributes) are fully analyzed by. And simplified algorithm of point cloud model is the most simplified local heuristic based on global optimization and not systematic. Similarly, simplified algorithm of 3D mesh models early also committed to preserving geometric features in the vertex number scenario reduction. Many a simplified algorithm while minimizing the geometric error and the texture deviation was put forward, and can retain the texture with the appearance of models. However, these algorithms do not fully exploit texture The content of image. Third, despite the high precision texture image may occupy more storage than the grid data model, bandwidth or computing resources, simplify the texture image has received little attention from researchers. This paper studies the relevant current 3D model compression and simplification technology, considering the existing problems, put forward: Based on the GLA (standard Lloyd algorithm) point cloud model compression technology. This technology expands the GLA algorithm to control the number of points in the cluster, and greedy optimization encoding sequence. This algorithm has obvious advantages in low bit rate compression, and universal algorithm for point cloud model compression of arbitrary topology structure. A simplified technique combined with neighborhood clustering of point cloud model based on this technique. By iteratively selecting and merging with local minimum cost, until the desired data. The rate of human vision. The algorithm of perception based on local variation using geometric and texture information, combined with the cost of computing the local neighborhood; Gauss texture sharpening techniques for enhancing texture detail based on points. In addition, the algorithm is extended to the large scale model is simplified. The technology can preserve the simplified geometry and texture the characteristics of the model surface. A simplified triangular mesh texture image and collaborative technology. The texture image is used to guide the mesh simplification, mesh simplification for texture image is simplified. The simplified model of the geometric data, will fully consider the simplified texture part, greedy selection for model shape and appearance of the minimum impact area is simplified in the simplified; texture image, will consider the complexity of local color, and make full use of geometric information model. The technology in the grid and reduce the model Texture and image data can also retain the geometric and textural features of the simplified model surface. In addition, the framework of the technology is very general. All kinds of iterative mesh simplification algorithms can be applied to the framework after proper adjustment.

【学位授予单位】：山东大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】