Multi-Fourier Spectral Descriptor to Integrate Knowledge-Based Processing for Three-dimensional Shape Database

在本研究中，我们专注于将基于知识的工程集成到三维形状数据库中，该数据库假定是用我们的专利技术MFSD（多傅立叶谱描述子）特征进行编码。在过去的三年中，我们发明了几种新技术；第一种是“线性扩散投影（LDP）”，它将特征空间降至近1/10维，指标紧凑，在高维空间中保持非线性结构。二是“基于突起形状的分割”。这项技术是在2011年2月的一个研讨会上提出的。这允许我们“按部分”搜索相似的形状，这使得枚举部分相似的形状成为可能，即使形状作为一个整体可能与输入查询不相似。

In this research, we have focused on integrating knowledge-based engineering on three-dimensional shape database, which are assumed to be encoded in our patented technology called MFSD (Multi-Fourier Spectral Descriptor) features. During the past three years, we have invented several new techniques ; the first one is "Linear Diffusion Projection (LDP)", which reduces the feature space into almost 1/10 dimensions with compact indices, keeping non-linear structure in higher dimensional space. The second one is "Segmentation based on protruded shape". This technique was presented in a workshop in February, 2011. This allows us to search similar shapes "by parts", which makes it possible to enumerate partially similar shapes, even though the shape as a whole may not be similar to the input query.