基于多视图和显著性分割的古生物三维模型检索  

Paleontological 3D model retrieval based on multi-view and saliency segmentation

作　　者：周宇航 冯宏伟[1] 冯筠[1] 刘建妮[2] ZHOU Yuhang;FENG Hongwei;FENG Jun;LIU Jianni(School of Information Science and Technology, Northwest University, Xi′an 710127, China;Xi′an Key Laboratory of Paleo-Bioinformatics, Shaanxi Key Laboratory of Early Life and Environments, Department ofGeology/State Key Laboratory for Continental Dynamics, Northwest University, Xi′an 710069, China)

机构地区：[1]西北大学信息科学与技术学院,陕西西安710127 [2]西北大学地质学系/大陆动力学国家重点实验室陕西省早期生命与环境重点实验室西安市古生物信息学重点实验室,陕西西安710069

出　　处：《西北大学学报（自然科学版）》2022年第4期590-601,共12页Journal of Northwest University（Natural Science Edition）

基　　金：国家自然科学基金面上项目(62073260);西北大学古生物信息学创新团队项目(2019TD-012)。

摘　　要：使用古生物化石图像检索古生物三维模型是一种建立和管理古生物三维数据库的有效方式,而直接以化石图像或复原图像检索古生物三维模型精确度较低。针对此问题,该文提出一种基于多视图和显著性图分割模型相结合的古生物三维模型检索方法。首先,在三维模型投影的基础上,对投影图像和检索图像使用显著性分割网络进行主体分割得到主体特征,然后,利用主体特征进行匹配得到相似投影图像对应的三维模型。在自建的古生物数据集上进行验证,该文所提方法最近邻准确度为0.845,F值为0.782,折损累计增益为0.856,各类别平均准确率为0.754。该方法可以有效地适应古生物数据集,并且准确率相对较好,能够辅助古生物三维模型的检索。Using paleontological fossil images to retrieve paleontological 3D models is an effective way to establish and manage paleontological 3D databases,while retrieving paleontological 3D models directly from fossil images or restored images is less accurate.Aiming at this problem,a 3D model retrieval method of paleontology based on the combination of multi-view and saliency map segmentation model is proposed in this paper.That is,on the basis of the projection of the 3D model,the saliency segmentation network is used to segment the projected image and the retrieved image to obtain the subject features,and then the 3D model corresponding to the similar projection images is obtained by matching the subject features.Validated on the paleontological dataset established in this paper,the nearest neighbor accuracy of the method proposed in this paper is 0.845,the F value is 0.782,the cumulative gain of impairment is 0.856,and the average accuracy of each category is 0.754.The method in this paper can be effectively adapted to the paleontological dataset,and the accuracy rate is relatively good,which can assist in the retrieval of paleontological 3D models.

关 键 词：三维模型检索 多角度 深度神经网络 主体分割 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]