Shape-based comparison of 3D models and its applications

点击翻译按钮获取中文摘要

Our research has produced the following to contributions:・ Retrieval method based on semantic classes: Shape as well as semantics play equally important roles in comparing 3D models by their "shape". Tb incorporate semantics associated with 3D mcdri9 into their "shape-based" comparison, we have developed a learning-based method. The method learns, offline, a set of multiple semantic labels attached to models in a 3D model database 'lb efficiently learn labels from a relatively small set of labeled mark's we introduced a semi-supervised approach to learning, which employs both labeled and unlabeled 3D models for the learning. Our empirical evaluation showed that the proposed approach does improve the retrieval performance of several 3D model retrieval benchmarks significantly. The method is believed to be the first to successfully learn semantic classes for 3D model retrieval. We believe that, fit the SHREC 2006 benchmark, the said method is most likely the best performing method as of … More March 2008.・ Local, multi-male, image features for retrieval: Shape comparison methods can be classified by their ketones into two globalleatme based methods and Iccalleatme based methods. The proposed method employs the latest local fratare. The method uses a set of local, multi-scale, image-based features extracted from a 3D model in compare the chap, of the model The method has an advantage over global-feature based methods for articulated models and models having global deformations Experimental evaluation of the method using a databa.se of articulated 3D model is showed that the method performs comparably to the best known method. Furthermore, our method is more versatile than the competing method in that our method can handle diverse 3D shape representations.Noteworthy achievement ; We entered the CAD model track of the Shape Retrieval (latest (SHREC) 2007, a 3D model retrieval contest, and won the track, proving the superiority of our algorithm. The winning algorithm is what we proposed in 2006 that employs unsupervised distance metric learning. We are planning to enter the SHREC 2008 contest, as our method based on semantic classes (above) Ear outperforms our own method based on distance metric learning, which won the SHREC 2007 CAD track. Less