【作者】 宋磊；

【导师】 孟军；

【作者基本信息】 大连理工大学 ， 计算机应用技术， 2006， 硕士

【摘要】 随着三维模型数据越来越庞大和复杂，使得图形绘制系统和网络传输的压力越来越大。为了能更好的显示和传输模型数据，需要对模型进行简化，删除对模型显示贡献不大或者对模型所描述的对象影响较小的图形元素，以此来达到加速绘制和降低网络负载的目的。传统的模型简化算法所考虑的简化依据主要是几何特征(曲率、边长等)，通过计算图形元素几何特征值来确定图形元素的重要性进而确定简化的顺序。 本文首先根据Garland原始的二次误差简化算法提出了一个基于边收缩的快速的网格简化算法，该算法属于利用几何特征进行模型简化的算法，但它提高了模型简化的效率，适于实时应用。从人类视觉的生理角度考虑，即使几何特性相同的模型区域，由于光照、纹理、颜色等特征的不同，它们所产生的视觉刺激也是有所差别的。本课题在传统算法的基础上，引入了视觉感知因素，提出了一种基于视觉感知的模型简化算法，并且结合视觉感知因素和显示质量度量模型提出了一种细节层次控制方法。首先对一些虚拟场景中基本的视觉感知因素进行了较为详细的分析，包括线框分辨率、纹理、距离、光照、颜色和亮度，通过对这些视觉因素的分析，得到了如何从模型中提取和度量这些信息的方法。然后把它们作为标量场信息引入到了Garland的任意维度的二次误差模型简化度量公式中，得到了新的基于视觉感知因素的误差度量公式。为了能够根据人类视觉感知特性来确定模型简化过程的结束条件，本文根据视觉感知因素得到一个细节层次控制参数，通过把这个参数与显示质量度量模型的结合得到了基于视觉感知的细节层次控制的方法。通过新的模型简化算法和细节层次控制方法，使模型的简化结果更符合人类的视觉特征，达到更好的视觉效果。 本文利用公用的三维网格模型进行了有关实验，通过比较传统简化算法和基于视觉感知的简化算法的简化结果，发现引入视觉感知因素后，在模型的视觉特征区域可以保留较高的细节信息，在其他区域则可以简化掉更多冗余的图形元素。并且通过基于视觉的细节层次控制方法来约束模型简化过程，可以使得细节层次的选择更加符合人类的视觉习惯和适应虚拟场景的状态要求。最后本文通过一个虚拟场景构建工具说明了如何把基于视觉感知的模型简化算法应用到实际的虚拟场景中。更多还原

【Abstract】 The pressure of rendering systems and transmission by network is higher and higher because of the more and more complex model data. The models need to be simplified for better display and transmission. Model simplification deletes graphics elements whose effect for model display or the object characterized by model is exiguous as much as possible so as to accelerate the rendering process and depress the network load. Geometric characters of meshes are mainly surveyed in traditional algorithms. The sequence of simplified elements is determined by calculating the eigenvalue of them.A Fast mesh simplification algorithm based on edge-collapse is presented in this paper. Although it just adopts only geometric characters to simply models, it accelerates the simplification efficiency and is fit of real-time applications. At the point of human vision physiology, some model regions with the same geometric characters maybe have different vision stimulation. This subject imports visual perception factors into traditional algorithms and presents a simplification algorithm based on visual perception and combines the visual perception factors and the display quality metrics to form a strategy of controlling the LOD. Firstly, some basic virtual perception factors in virtual scene are analyzed, including wire-frame resolution, texture, distance, illumination, color and luminance. By these analyses, the way of distilling and measuring these factors is found out. Secondly, these perception factors are imported into Garland’s quadric metrics simplification in any dimensions as scalar information. Then a new error metrics is founded. In order to determine the end condition of the simplification procedure, a parameter to control LOD based on visual perception is computed out in this paper. By this parameter and display quality metrics, the strategy to control LOD based on visual perception is presented. The simplification algorithm and LOD control strategy can make the simplified meshes adapted to human visual perception.Some public 3-d models are adopted in the experiments. By analyzing the experiment results, it can be found that more details are kept in the visual perceptual areas, while more redundancy data are deleted in other areas. The choice of LOD is suitable for human vision and the demand of virtual scene state. At last, a constructor of virtual scene is introduced to explain how to adopt the new simplification algorithm in the actual virtual scene.更多还原