3D Model Retrieval Method Based On Content And Sketch

With wide application of three-dimensional(3D)models in China's manufacturing,processing technology and digital modeling technology,it becomes an important research direction to classify and retrieve 3D models.3D model has inherent complexity.The method of reducing the dimension of 3D model is adopted in this paper.Two-dimensional(2D)image is used as the condition of searching 3D model,in order to decrease retrieval costs further.2D image retrieval algorithm in this article consists of 2D sketch retrieval algorithm as well as 2D view retrieval one.Ordinary users may not have 2D view and 3D model to be retrieved directly.Rough filtering retrieval is performed with rough-hand sketch in the first part of this article.The selected 2D views are applied to detail classification and retrieval in the second part.This paper utilizes double search to achieve the goal of retrieving 3D models fast and accurately.The set of 3D model's 2D views is obtained according to the fixed projection method in freehand sketch retrieval part.From 2D view set,24 two-dimensional views with the longest contours are selected as the feature view set of 3D model.Then,the hand-drawn sketch is denoised.A weighted set of global view features and 2D shape distribution histogram is constructed as the integrated feature descriptor.The integrated feature descriptor is used to evaluate the similarity between hand-drawn sketches and 2D views in order to retrieve 3D models.Experimental results show that the performance of the algorithm in this paper is better than those of other 3D model retrieval methods based on hand-drawn sketches.It has higher recognition of regional features and contour ones.An ideal 3D model is selected from the retrieved views as the condition of view retrieval in next step.Firstly,a sphere is used to envelop the selected 3D model completely in 2D view retrieval part.The purpose is to obtain a set of 2D projection views for 3D model along a fixed perspective.Secondly,Apriori algorithm is adopted to select typical views as the input of convolutional neural network.Based on Alex Net model,parameters are adjusted and used for 3D model classification and retrieval.Thirdly,convolutional neural network is optimized by combining forward propagation and back one in order to improve its generalization performance.Experiments show that this method can improve accuracy and efficiency of 3D model classification and retrieval.