Style Pattern Mining For 3D Interior Scene Analysis And Furniture Recommendation

With the rapid development of 3D modeling and virtual reality technology,online interior scene design and 3D furniture shopping will gradually move towards the public vision,so the multi-dimensional scene analysis and recommendation technology will play an increasingly important role.Traditional analytical work based on semantic categories has been unable to meet the demand.Inspired by the recent works of three-dimensional shape style analysis,this paper applies the middle-level patch thought in the field of image analysis to the tree-dimensional field and presents a three-dimensional interior scene style analysis method and a three-dimensional furniture recommendation method.The corresponding prototype system is realized as well.The methods and system presented in this paper can help professional interior designers and ordinary users to complete style analysis tasks of 3D scene and choose three-dimensional furniture shape.The main work is reflected in the following three aspects:(1)Style Analysis of 3D Interior Scene Based on Style Pattern Mining.This paper presents the first three-dimensional interior scene style analysis problem,and gives a solution based on the style pattern mining.Inspired by the middle-level patch thought in the area of image analysis,randomly part-based sampling is conducted on three-dimensional shape firstly,then multi-view silhouette features are extracted for each part and a multi-view clustering method is used to find the style elements of each type of 3D shape.Each 3D scene can be represented as a transaction through style elements,in order to describe the overall style of the scene,a classic frequent pattern mining method is adopted to dig the match patterns of style elements between different shapes.The experimental results prove that the method proposed by this paper not only can analyze the style of 3D interior scene effectively,but also can intuitive display what and where the style is through back projection of the style elements,which lays the foundation for the follow-up furniture recommendation.(2)3D Furniture Recommendation Based on Style Rule Mining.Based on the previous style analysis word of 3D interior scene,this paper continues to mine the association rules of style elements,and proposes an interactive recommendation method for 3D furniture.This method draws on the idea of collaborative filtering in the recommendation system,regards each scene in the training set as a shopping basket record,the furniture elements of each furniture are regarded as items,and the association rule mining algorithm is used to mine the association rules between style elements of different furniture shapes.In the recommendation phase,the user is asked to interactive specify the furniture category which expected to be recommended and chose the results recommended by the system.The system finds the most appropriate style rules based on the overall style elements of the currently selected furniture shapes,and then recommends the appropriate furniture shapes.User survey results show that this method can not only recommend the interior scene with a consistent style,but also achieve a high degree of satisfaction after the user's interaction.(3)Design and Implementation of HomeStyler System.Previous two work are integrated into a prototype system,called HomeStyler.The main function and the related data structure of the system are introduced from the point of view of software design,and the main modules of the system are demonstrated through examples.The system can be used as a three-dimensional interior scene style analysis tool,help users complete the interior design,and can help researchers to implement further style analysis work.