Joint Analysis Of 2D Image And 3D Shape Features With Simple Interaction

In recent years,due to the rapid development of 3D modeling technology,the scale of available 3D model database on the Internet arises an explosive growth.More and more 3D models can be easily downloaded through the Internet.This directly led to the development of 3D shape retrieval technology,which is,given the specific search information,the system needs to return a similar 3D model according to the user's requirement.In this paper,two related methods about 3D shape retrieval are proposed under the topic of 2D image and 3D model features analysis.The first one is a 3D shape retrieval method which is based on local features extraction and matching on 2D sketch image.The method takes a 3D model as a query input,and for each 3D model,it generates many 2D sketch images which contain information about the structure of 3D model.Next,the method obtains a feature histogram for each 3D model by extracting local features of 2D sketch images and finally searches out the most similar 3D model from the database through matching feature histograms.The second one is another 3D shape retrieval method which is based on deep convolution network.This method takes hand-drawing sketches as query input and build a joint embedding space of 2D sketches and 3D models by features extraction and matching based on the deep convolution network.The distance between similar 3D models and corresponding 2D sketches in the joint embedding space will be nearer than those dissimilar ones.Therefore,the spatial relationship between 2D sketches and 3D models is built up in order to retrieve out the most similar 3D models with the input 2D sketch.The two approaches we propose tightly combine the relationship between 2D images features and 3D models features.The numerical experiments clearly demonstrate that our proposed algorithms perform well on some public datasets.