基于形状检索的场景图像三维建模  

3D Modeling of scene images based on shape retrieval

作　　者：樊亚春[1] 谭小慧[1] 周明全[1] 陆兆老[1] 

机构地区：[1]北京师范大学信息科学与技术学院,北京100875

出　　处：《高技术通讯》2013年第8期781-788,共8页Chinese High Technology Letters

基　　金：国家863计划(2008AA01Z301);国家自然科学基金(61001168;61170203)资助项目

摘　　要：提出一种利用二维场景图像快速构建三维虚拟场景的三维建模方法,该方法有以下特点:利用改进的图割算法对场景图像进行分割,提取对象轮廓形状;基于尺度不变特征点进行轮廓形状局部特征描述,用量化后的视觉词汇向量描述三维模型轮廓及对象轮廓;计算不同形状轮廓之间的距离,从三维数据库中检索出与对象相似的三维模型;计算对象模型和背景场景模型的坐标变换,合成虚拟场景。实验证明,该方法检索效果好,用户操作简单,能实现复杂虚拟场景的快速建模。A new method for quick building of a 3D virtual scene by using 2D scene images is proposed.It has the charcteristis below：an improved graph cutting algorithm is used to segment objects and extract contour information in the scene image ;a new local shape description is used to describe the contour of 3D model projections and 2D image by using quantized visual vocabulary vectors which are based on scale-invariant feature transform; 3D models are retrieved from 3D databases by computing the distances between shape contours of different 3D models;coordinate position is transferred to combine object models to build the visual scene.The experimental results show that the performance of the proposed approach is quite well and it can effectively build complicated virtual scenes.

关 键 词：三维建模 图割 形状检索 视觉词汇 多尺度特征描述 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]