Research On Attitude-based Model Retrieval

With the rapid development of both the hardware and software,Scanning technology and 3D modeling technology has already become mature.A large number of 3D models were constructed,which as a new information carrier has been got more and more attention and application.The usage lies not only in areas like computer aided design,bioscience,medicine,chemistry,industrial and military,but also can be seen in our daily life,such as 3D games and movies.With a growing number of demand of using 3D models and the massive appearance of 3D models on the Internet,users may confront a same problem that finding suitable models is a difficult job in a short period.Thus,with model as input,searching out the required information from the internet has becoming more and more valuable in commercial application and scientific research.There have developed many methods for model retrieval recently.Such methods to model retrieval currently include two major classes: one is rigid method,which is focused on how to extract the digital vector form the given model's inherent shape information,the digital vector can effectively express the model,and then use it to complete the similarity measure between models,which can used in classification and identification.But for the rigid feature descriptors,different forms of the same model may lead to inconsistent feature descriptor.However,we still expect for different deformation models have a similar feature descriptors.So it appeared the second kind of method for nonrigid retrieval.However,this kind of methods ignore the attitude information of models,so the retrieval results can't sorted by the attitude information.The purpose of this paper is to find an effective way of the combination of features in rigid and non-rigid method,to improve the discriminative of features,so as to improve the accuracy of image retrieval,and join the attitude information in the search result.We proposed an model presentation framework which based on rigid and non-rigid transformation in this paper.This framework uses HKS feature to ensure deformation invariant,and then use the shape distribution feature in the attitude information.In this way,the feature is more discriminative than single local feature,and with higher repeatability.By experimental verification,compared with traditional methods,the proposed method not only improves the precision of model retrieval effectively,but also considers the attitude information of models.