【作者】 张玉伟；

【导师】 周以齐；

【作者基本信息】 山东大学 ， 机电产品创新设计与虚拟制造， 2014， 博士

【摘要】 运用先进的计算机技术辅助传统艺术创作,一直是计算机图形学、艺术设计学、计算机辅助设计和先进制造技术等领域的研究前沿和最富挑战性的课题之一,研究热点包括对反映视觉心理规律的艺术作品数字建模和风格化处理。本文以具有独特艺术表现形式的浅浮雕为对象,对3D模型到2.5D浅浮雕映射过程中涉及的视点选择、几何建模、编辑修改等关键技术和相关支撑算法进行了深入研究。目前主流的数字浅浮雕建模技术以2D图像建模为主,通过提取有用模型信息完成单幅图像到浮雕曲面的重构。由于单幅图像包含的深度信息有限,这种方法只有在增加用户操作或先验知识的情况下,才能得到相对精美的浮雕效果。相比而言,基于3D模型映射的浅浮雕建模技术在模型视点选择、深度信息获取、数字几何处理等方面具备明显优势,它以3D模型为输入介质,在观察视点定位后,将模型映射在背景平面或曲面上生成浮雕作品。虽然该技术目前还处于起步阶段,但随着国内外研究的不断进行,基于3D模型映射的浅浮雕建模方法必将在浮雕艺术创作领域得到广泛应用。由3D模型生成浅浮雕作品需要解决以下关键问题：(1)在输入模型已知的情况下,如何定位模型观察位置,选择最能反映模型形状特征的投影方向。(2)当模型压缩量很大时,浅浮雕视觉显著特征会变得模糊甚至丢失,如何将原始模型的细节信息“迁移”到浮雕表面,保证浅浮雕细节特征的清晰度。(3)浅浮雕映射需要具备实时性,如何在保证建模质量的前提下提高系统运算效率,实现“所见即所得”的可视化操作效果。(4)建模后的浅浮雕作品在局部形状需要调整时,如何为浅浮雕编辑提供实用、有效的解决方案。围绕上述关键技术问题,国内外研究人员已经提出了一系列的建模策略,虽然能解决部分存在问题,但距离实际应用还有一定差距。其中大部分方法在获取模型高度场信息后,采用图像增强和非线性压缩进行浅浮雕建模,这种规则化的像素操作模式丢失了原始模型的几何拓扑信息,缺少对模型不同部位的多分辨率控制,不利于浅浮雕作品的艺术再加工和后期编辑处理。与之不同,本文在3D空间域内完成数字浅浮雕建模和编辑工作,全部算法以非规则化的网格三角面片为基本处理对象,这样保持了建模前后模型拓扑结构的一致性,有利于各种网格几何处理算法在浅浮雕设计领域的应用和扩展,为具有复杂场景设计要求的浮雕艺术创作、融合和再利用奠定良好基础。具体来讲,本文在以下方面做了深入研究：(1)以信息论为理论基础,将网格面片显著度融入到视点属性的计算过程中,提出了基于模型显著度驱动的视点质量评价方法,为浅浮雕投影方向的选择构建了视点定量分析工具。本文定义了两种显著度计算方法来捕捉模型的视觉重要特征,第一种方法与视点无关,根据平均曲率梯度幅值来区分网格面片的重要度；第二种与视点相关,采用3D线画技术来标定当前视点下的面片显著度值,然后通过采样平均确定网格的显著度分布。为了提高计算效率,本文忽略了一般的视点质量评价属性,将最重要的视点信息量、视点稳定性、视点可见性等质量评价指标集合在一起,建立了视点综合质量评价标准,系统根据计算结果自动推荐最佳模型投影方案。(2)提出了一种空间域实时浅浮雕数字建模方法,解决了浅浮雕建模容易丢失细节并且生成速度较慢的难题。首先采用空间域双边滤波算法提取三维网格的多层细节信息,然后进行视点相关的坐标映射构建浮雕高度场,在以网格Z值最小点定位基准压缩平面后,将模型主体和细节特征分尺度线性压缩并进行自适应叠加,生成满足设计要求的浮雕曲面。该算法基于空间域操作,几何运算简单、高效,在细节特征保持的同时,高效重用了预处理阶段提取的几何细节信息,获得了交互级的建模和细节编辑效率,最终的浅浮雕生成结果令人满意。(3)以三维离散网格框架下的微分形变技术为理论基础,提出了一种基于微分操作的浅浮雕自动生成及编辑方法。该方法将用户的浮雕设计意图映射为微分域内网格梯度场的操纵,通过求解泊松稀疏线性方程实现浅浮雕建模和局部编辑。根据浅浮雕建模需要,本文给出了一系列具体的应用实例,包括平面浅浮雕映射、细节特征保持、高度场连续性优化、曲面浅浮雕映射等建模算法,以及轮廓边高度编辑、勾画式外形编辑、骨架驱动式形状编辑、浮雕网格融合等编辑算法。这些算法搭建了一个通用浅浮雕几何处理框架,为微分域浅浮雕建模技术的应用和扩展提供了实用的解决方案。(4)提出了一种由3D模型生成数字沉雕的建模方法,将线条雕刻和沉雕高度场有效结合在一起,生成具备良好外观和可加工性的数字沉雕作品。算法对提取的3D轮廓特征线进行裁剪、串联、平滑等基本操作,使轮廓线具备光滑、连续的矢量特征,更好与模型主体曲面进行融合。与传统的浅浮雕不同,本文采用郎伯光照结合深度信息的方法构建沉雕高度场,建立了三个独立的描述图层：外部轮廓线图层用于勾勒物体的整体形状,内部装饰线图层用于表达物体的局部特征细节,光滑的高度场图层在承载雕刻线条的同时,用于传递场景的立体印象。通过模拟刀具雕刻过程,三个图层最终合成在一起,生成最后的数字沉雕作品。更多还原

【Abstract】 Over the past few years, computer aided art design has been one of the most challenging topics in computer graphics, art design, computer aided design and advanced mechanical manufacturing. The research hotspots include visual psychology analysis, model simulation and art stylization on different kinds of artistic products. In this paper, we take bas-relief as the research object, and investigate the key techniques of digital bas-relief generation, such as viewpoint selection, geometric modeling and relief editing.So far, most existing methods for bas-relief generation take advantage of2D image processing that tend to convert the useful information of a well-designed image to the height field of the relief. Since the depth information is often limited in a single image, the quality of the generated bas-relief depends largely on the experience of the designer. In recent research on automatic bas-relief generation, a great deal of attention has been devoted to view-dependent3D object compression, which begins with the input of original3D scenes. Compared with traditional2D image algorithms,3D object-based methods have competitive advantages in viewpoint selection, depth information extraction, and geometric processing. Although the technology is now in its developing stage, it will play a more important role in future bas-relief creation with the development of the research.To generate the desired bas-relief from a3D object, several problems need to be solved. First, for a given3D object, how to select a most representative viewpoint for subsequent geometric modeling? Second, when a large compression is conducted on the input object, the original feature details will be lost in the bas-relief, so how to preserve the fine details and ensure the generation quality? Third, the computation efficiency is critically important to bas-relief generation, so how to present a real-time algorithm and ensure the availability of the production? Finally, when local shape of the bas-relief needs to be modified, how to provide an efficient solution for further relief editing?Currently, there have been abundant researches to deal with the above problems. Although some satisfactory results have been obtained, there is still some difference from the practical application. Among them, most existing methods operate in image space that take the rendering of3D object as the input, and represent the depth information in z-direction displacements based on regular pixels sampling. Thus, the topology of the generated relief becomes completely different from that of the input mesh. Furthermore, the pixel-based representation of the output makes shape editing of the relief much more difficult, largely limiting the extensibility of the bas-relief. Our approach differs from existing methods in that it operates directly on triangular mesh, maintaining the topology of the mesh unchanged during geometric processing. On this basis, more achievements from3D mesh processing techniques can be explored for further bas-relief producing, merging and reusing. In this paper, the main research contents are as follows:A new saliency-driven approach for best viewpoint selection is proposed based on information theory. To discriminate some features are more salient than the others, we define mesh saliency in two ways:a view-independent method using third-order surface derivatives, and a viewpoint sampling approach based on the line drawing technique. After that, the mesh saliency is incorporated in the computation of the most influential descriptors:viewpoint information, viewpoint stability and viewpoint visibility. Finally, we combine the three descriptors together to evaluate the goodness of a viewpoint and to select the most representative views for the bas-relief generation.Aiming at preserving fine details and achieving high computational efficiency, a real-time bas-relief generation algorithm is proposed based on geometric compression. The feature details are first extracted from the original objects using a spatial bilateral filtering technique. Then, a view-dependent coordinate mapping method is applied to build the height domain for the current view. After fitting the compression datum plane, we use an adaptive compression function to scale and combine the Z values of the base mesh and the fine details. The proposed approach operates in object space that offers flexible control over the extracted details, making it possible for real-time bas-relief generation and details adjustment.A novel approach for bas-relief generation and editing is proposed based on gradient-based mesh deformation. By deforming the input mesh implicitly through gradient manipulation and Poisson reconstruction, our approach works well both for plane surface bas-relief and curved surface bas-relief generation. We put forward a series of gradient-based algorithms such as height field deformation, surface continuity enhancement, fine details preserving, curved surface flattening and so on to ensure high-quality of the generation. Besides that, we present several editing tools, including boundary height editing, sketching-based boundary shape editing and skeleton-driven shape editing that allow the user to interactively modify the bas-relief to a desired shape. On this basis, a basic framework for bas-relief generation and editing is developed.The problem of direct generation of line-based sunken relief from a3D mesh is studied. We present a suitable way for the combination of feature lines and relief height field. After extracting view-dependent feature lines from a3D mesh, we concentrate on the post-processing of the lines including clipping, chaining, and smoothing. Based on this, the extracted lines are organized in good forms and appearance, appropriate for line engraving. We construct the base height field of the sunken relief using Lambertian shading. The input3D mesh is finally transformed into three basic layers:a contour line layer to strengthen the sense of outlines, a decorative line layer to represent the relief details, and a smooth base layer to convey2.5D illusion of the depth. Using a continuous pitting corrosion method, we combine the three layers together and generate the sunken relief with desired line depth and stylization.更多还原