Research On Virtual Interior Items Retrieval System Based On Local Features

With the development of e-commerce household sales,and the rise of vertical shopping guide platform,it is urgent that how to enhance the user experience.At present,e-commerce sales are also facing the problem of information explosion.According to statistics,the types of goods on Taobao have been refined into billions.The way using keywords to search already can't satisfy the needs of users,we must provide consumers with goods retrieval and visualization.In particular,retrieval function is very important,directly affects to the user's experience.Considering that a large amount of data is not only described in the form of text,but also displayed as images.Therefore,Content-based image retrieval has great practical significance.Compared with the image,the 3D model carries more information such as shapes and texture.The current development trend for the commercial platform is to provide users with more intuitive display way of three-dimensional items.The main purpose of this paper is to find out how to retrieve the 3D model using the indoor image taken by user,and provide users with interactive retrieval results.This paper we mainly focus on three aspects: initially,how to find out the correspondence between the reality image and virtual 3D model,the main difficulty of this problem lies in the real image and the image captured form 3D virtual model have different characteristics.In this paper,we reformulate the correspondence problem as an object classification problem and solve it using exemplar LDA.The second main work is to study the correspondence between 3D models.In order to improve the accuracy of 3D model retrieval,this paper combines the geometric features of the 3D model and image features for 3D model retrieval.On the basis of this,we also compute the fuzzy correspondences between the models considering the local features of 3D models.Finally,we integrate the first two works above to achieve a model retrieval system based on the indoor image taken by user.The system is competent for the feature extraction of image and ROI-based 3D model retrieval.We demonstrate the actual availability of the previous two studies on several datasets.