Nowadays 3D models play an important role in many applications: viz. games, cultural heritage, medical imaging etc. Due to the fast growth in the number of available 3D models, understanding, searching and retrieving such models have become interesting fields within computer vision.ududIn order to search and retrieve 3D models, we present two different approaches: one is based on solving the Poisson Equation over 2D silhouettes of the models. This methoduduses 60 different silhouettes, which are automatically extracted from different viewangles. Solving the Poisson equation for each silhouette assigns a number to each pixel as its signature. Accumulating these signatures generates a final histogram-based descriptor for each silhouette, which we call a SilPH (Silhouette Poisson Histogram).ududFor the second approach, we propose two new robust shape descriptors based on the distribution of charge density on the surface of a 3D model. The Finite Element Method is used to calculate the charge density on each triangular face of each model as a local feature. Then we utilize the Bag-of-Features and concentric sphere frameworks to perform global matching using these local features.ududIn addition to examining the retrieval accuracy of the descriptors in comparison to the state-of-the-art approaches, the retrieval speeds as well as robustness to noise and deformation on different datasets are investigated.ududOn the other hand, to understand new complex models, we have also utilized distribution of electrical charge for proposing a system to decompose models into meaningful parts. Our robust, efficient and fully-automatic segmentation approach is able to specify the segments attached to the main part of a model as well as locating the boundary parts of the segments.ududThe segmentation ability of the proposed system is examined on the standard datasets and its timing and accuracy are compared with the existing state-of-the-art approaches.
展开▼
机译：如今，3D模型在许多应用程序中扮演着重要角色：即。游戏，文化遗产，医学影像等。由于可用3D模型数量的快速增长，理解，搜索和检索此类模型已成为计算机视觉中的有趣领域。 ud ud为了搜索和检索3D模型，我们提出了两种不同的方法：一种是基于在模型的2D轮廓上求解泊松方程。此方法使用60个不同的轮廓，这些轮廓是从不同的视角自动提取的。求解每个轮廓的泊松方程可为每个像素分配一个数字作为其签名。累积这些签名会为每个轮廓生成最终的基于直方图的描述符，我们将其称为SilPH（Silhouette Poisson直方图）。 3D模型。有限元法用于计算每个模型的每个三角形面上的电荷密度作为局部特征。然后，我们利用特征包和同心球框架使用这些局部特征执行全局匹配。 ud ud除了与最新技术相比检查描述符的检索准确性外，研究了不同数据集的速度以及对噪声和变形的鲁棒性。 ud ud另一方面，为了了解新的复杂模型，我们还利用电荷的分布提出了一种将模型分解成有意义的部分的系统。我们强大，高效且全自动的分割方法能够指定附加到模型主要部分的线段，并找到线段的边界部分。 ud ud在标准下检查了提议系统的分割能力将数据集及其时间和准确性与现有的最新方法进行比较。
展开▼
