Automatic Recommendation Algorithm Of 3D Home Ornaments

Currently,the choice and layout of indoor ornaments are usually done manually.It is tedious and time-consuming to choose proper candidates from large number of the ornament database.Besides,the result of the selection cannot be guaranteed.Therefore,it is meaningful to develop an automated method of ornaments recommendation and placement instead of manual methods.In this paper,we propose an automated recommendation and arrangement method of single part ornament and multi-part ornaments.Our method is essentially a data-driven method.However,there is no any open dataset of ornaments and 3D home scene.To address this problem,we firstly construct an ornament dataset from the 3D home scenes which are design by aesthetic experts.For the single ornament recommendation task,we design an auxiliary task to do ornaments type recommendation.We adapt the mixed recommendation idea,and make fusion features from embedding feature and original feature as the input of single ornaments recommendation algorithm.To recommend the type of the ornaments,we use matrix decomposition and convert recommendation task into a binary-classification problem.With the original features of the subspace and ornaments,we add the embedding feature and input the fusion features into ornaments recommendation model to generate the final recommendation list.For the multi-part ornaments group recommendation task,we use the Apriori algorithm to find association the associate rules among the ornaments in the dataset,and then get the groups with the same type,the same function,the same style,or the combinations.The proposed method in this paper is based on the content recommendation algorithm so that it avoids cold start problem to a certain extent.Moreover,it is easy to adjust recommendation list and it also conveniently increases the item-based collaborative filtering recommendation module to enrich the recommendation results.To evaluate the effectiveness and reasonableness of the recommendation results,we develop a visualization system of the ornaments recommendation.Experimental results show that our method can recommend comparable results with expert-level user,which can replace or ease the manual selections of the ornaments for 3D home layout.