Research On 3D Model Retrieval Technology Based On Deep Learning

With the rapid development of technologies,the forms of multimedia data information are more and more diversified.Due to the specific characteristics of actual situations,three-dimensional models are able to carry more information and more expressiveness than textual information and two-dimensional images.And three-dimensional models are applied more and more widely in three-dimensional movies,games and stereoscopic imaging,medical technology,geographic information system and other aspects.The technology of 3D model retrieval has also gradually developed under the progress of the times.With the development of these technologies,3D models are deployed in various fields more and more frequently.With the exponential growth of 3D models,to obtain the desired target model quickly and effectively from massive data becomes the most important and urgent issue.This paper firstly presents the significance and background of 3D model retrieval,then,this paper introduce the Deep Learning into the field of 3D model retrieval.Based on this,two kinds of 3D model retrieval algorithms are presented: 1)3D model retrieval algorithm based on circular convolution neural network: this method uses convolutional neural network(CNN)to extract low level translation invariant underlying features.Then sequence learning is carried out by recurrent neural network(RNN)to extract more robust features under different views.2)3D model retrieval algorithm based on residual network and long-short term memory: The circular convolution neural network is optimized by applying Residual Network(ResNet)structure and long short term memory(LSTM)model to avoid the disappearance of gradient.Finally,this paper verifies the actual retrieval results of the above two kinds of depth-based retrieval techniques by experiments and comparison with the current mainstream algorithms.And feasibility and effectiveness of the proposed algorithm are also verified by attached experimental results.