Multi-view Perception Optimization For 3D Scenes

With the rapid development of computer technology,the resource sharing of 3D modeling database has been achieved.For the convenience of browsing and consulting 3D models,databases usually require one or more preview pictures to represent a 3D model.A set of high perceivable preview pictures can help users to quickly identify the corresponding 3D model,conversely,a set of low perceivable preview pictures will grievously affect users' identification.In particular,at the last stage of interior design,interior designers have a responsibility to provide a set of overall renderings to users for perceiving the overall style.Therefore,it is a significant problem that how to adaptively select multiple viewpoints to optimally represent and perceive the input 3D scene model.In the field of three-dimensional perception,most of researchers focus on the view-point selection problem of single model,conversely,a little for complex 3D scene model just pay close attention to best single viewpoint selection of global or local perception.In the light of the problem that the existing works haven't researched multi-view per-ception for 3D scenes,this paper firstly researchs this problem.And in the light of the viewpoint metric algorithms of the related works aren't suitable for the multi-view perception of 3D scenes,this paper defines a perception-aware metric function which ensures that the corresponding best viewpoint covers enough significant geometric in-formation and conforms to the perception habits of people.Given an indoor scene model,firstly,a model-based metric function and a view-based metric function,based on significant geometric information and aesthetic infor-mation respectively,are defined in this paper.Then,a perception-aware metric func-tion,which ensures that the corresponding viewpoints cover enough significant geo-metric information and conform to the perception habits of people,can be obtained by a weighted combination of the two metrics.Secondly,on the basis of the the perception-aware metric function and the number of viewpoints specified by users,a multi-view objective function to ensure collaborative perception is modeled.Finally,this paper uses the classical simulated annealing algorithm to solve the multi-view optimization model for avoiding local optimal solution.In order to verify the validity and rationality of the algorithm,the quantitative and qualitative analysis of experimental results are presented in this paper.The quantita-tive comparisons with metric function of related work show the effectiveness of our perception-aware metric function.In addition,the qualitative comparisons with excel-lent interior designers' works prove the rationality of our algorithm.Extensive experi-ments demonstrate that our method can adaptively obtain a set of all-sided and artistic viewpoints which cover enough significant geometric information and conform to the perception habits of people for a variety of 3D scenes.