非刚性三维形状的内蕴相似匹配算法研究

编辑:指甲网互动百科 时间:2020-01-19 02:56:10
编辑 锁定
随着科技的不断进步,三维模型在现代计算机应用中扮演了越来越重要的角色。对于三维形状之间的相似性匹配与检索,是重利用这些数据资源的一项非常有意大量的非刚性三维模型为建立鲁棒的形状匹配算法带来了挑战。这类三维模型自身的形状存在较多的差异往往比类间的差异更加明显。为了有效衡量非刚性三维形状之间的内蕴相似性,形状匹配算法必须对等距形变足够鲁棒。
中文名
非刚性三维形状的内蕴相似匹配算法研究
外文名
Intrinsic non rigid 3D shape similarity matching algorithm
类    型
三维模型
领    域
现代计算机应用

非刚性三维形状的内蕴相似匹配算法研究中文摘要

编辑
本文研究了非刚...>> 详
大量的非刚性三维模型为建立鲁棒的形状匹配算法带来了挑战。这类三维模型自身的形状存在较多的可变性,它们通常是对原始刚性三维形状进行一定程度的等距非刚性变形而得到的。于是,其类内形状差异往往比类间的差异更加明显。为了
  本文研究了非刚性三维形状内蕴相似性匹配中的相关算法,其中,主要贡献包括:
  第一,提出一种高维内蕴形状的局部特征描述方法。为了减少形状标准化映射过程中的距离近似误差,选用一个高维欧氏空间来计算非刚性三维模型的内蕴标准形。然后利用子空间分解技术,从中构造内蕴自旋图像特征,来直接刻画高维形状的局部几何信息。内蕴自旋图像无关于非刚性模型的等距形变,同时有效继承了自旋图像的非参数性及局部描述能力。
  第二,提出一种非刚性模型高维内蕴形状的优化方法。为了提升标准形的描述能力,定义嵌入误差函数来度量模型标准化映射过程中产生的形状扭曲程度,并选取误差足够低、同时又最为紧致的临界欧氏空间来计算其高维内蕴形状。在异维标准形之间,推广内蕴自旋图像特征进行高维形状的直接匹配。得益于嵌入空间的优化,内蕴形状信息得以更为精准地刻画。
  第三,提出一种高效的形状标准化方法,自动地将非刚性三维模型映射为等值线标准形。通过计算测地距离等值线,并将其重对齐到引导轴上,保证了形状部件被局部映射为统一的非弯曲结构。同时,利用模型表面的测地距离约束来优化引导轴的朝向,保证整体形状结构的内蕴性质。等值线标准形显式消除非刚性模型的等距形变差异,不仅减少了形状扭曲,而且具有更高的计算效率。
  第四,提出一种基于自动逆关节形变的非刚性三维形状检索算法。借助于刚性形状分割与网格变形技术,自动调整模型的形状结构,消除关节形变信息。然后结合刚性匹配方法,来计算非刚性形状之间的内蕴相似性。逆关节形变算法保持原始非刚性三维模型的几何细节信息,显著改善了内蕴形状匹配的效果。此外,模型检索的计算复杂度也得以大大降低。
  关键词:非刚性三维形状,内蕴相似匹配,标准形,内蕴自旋图像,等测地线

非刚性三维形状的内蕴相似匹配算法研究外文摘要

编辑
With the rapid advance of technologies, three-dimensional models have been playing an increasingly important role in many modern computer applications. To better organize and reuse these valuable resources, it desperately requires the development of an efficient shape matching and retrieval system. It is also significant to promote the researches of other 3D model processing algorithms.
  However, many nature objects are non-rigid, which poses a great challenge in developing robust shape matching algorithms. Non-rigid 3D shapes are generated by isometrically deforming the original rigid ones. Therefore, their intra-class shape variations are often at the same magnitude or even larger than the inter-class variations. In order to evaluate the intrinsic similarity between non-rigid objects, the shape matching method is required to be insensitive to isometric deformations.
  This dissertation concerns on the algorithms regarding intrinsic similarity matching of non-rigid 3D shapes. The main contributions of this thesis include,
  First, an efficient local signature is proposed to represent the high-dimensional intrinsic shape. To preserve more embedding accuracy, a higher dimensional Euclidean space is preferred for calculating the canonical form of the non-rigid 3D model. Then, the subspace decomposition technique is employed to construct intrinsic spin images, which directly characterize the local geometric properties of high-dimensional shapes. The intrinsic spin image is not only invariant to isometric shape deformations, but also inherits the locality and non-parametric nature of the spin image descriptor.
  Second, a method for optimizing the canonical form is proposed for each individual non-rigid 3D model. To obtain the best representation power, a strategy is first proposed to select the most compact yet expressive Euclidean space. It is based on the embedding error, which measures the amount of shape distortion caused by the canonical mapping. Then, the intrinsic spin image is employed to conduct the high-dimensional shape matching across different canonical spaces. Benefit from the optimization of canonical spaces, intrinsic shape properties are characterized with more accuracy.
  Third, an efficient intrinsic embedding method is proposed to automatically map the non-rigid shape to its contour canonical form. Locally, each subpart is regularized with an unbent shape structure, by adaptively re-aligning the geodesic contours along a straight guidance axis. The isometry-invariance of its holistic structure is guaranteed by an global optimization under the geodesic constraints defined on the shape surface. The contour canonical form explicit eliminates the isometric deformations of the non-rigid object. It not only reduces the shape distortion, but also enhances the computational efficiency.
  Finally, a new non-rigid retrieval framework is proposed based on automatic shape anti-articulating. With the 3D segmentation and mesh editing technique, the shape structure is adjusted, in order to eliminate existing shape articulations. Afterwards, the rigid shape matching technique is introduced to measure the intrinsic similarity between the original non-rigid objects. The anti-articulating algorithm preserves more geometric details and guarantees the smoothness of the generated shape, which definitely improves the intrinsic matching performance. Besides, the time complexity for shape retrieval is greatly reduced.
  Keywords: non-rigid 3D shapes, intrinsic similarity matching, canonical form, intrinsic spin image, geodesic contours[1] 
参考资料
词条标签:
其他文化 书籍