基于支持向量机的无序图像有序化研究  

Research on unordered image ordering based on support vector machine

作　　者：钱赛男 李英成[1,2,3] 朱祥娥 刘晓龙 QIAN Sainan;LI Yingcheng;ZHU Xiang’e;LIU Xiaolong(Chinese Academy of Surveying and Mapping,Beijing 100036,China;Key Laboratory for Aerial Remote Sensing Technology of Ministry of Natural Resources,Beijing 100039,China;Beijing Engineering Research Center of Low Altitude Remote Sensing Data Processing,Beijing 100039,China)

机构地区：[1]中国测绘科学研究院,北京100036 [2]航空遥感技术自然资源部重点实验室,北京100039 [3]北京市低空遥感数据处理工程技术研究中心,北京100039

出　　处：《测绘科学》2020年第2期111-116,共6页Science of Surveying and Mapping

基　　金：国家十三五重点研发计划项目(2017YFB0503004).

摘　　要：为提高大规模图像的分类效率及准确性,解决大规模无序图像的有序化问题,该文通过学习引用自然语言及机器学习的理论知识,提出基于视觉词袋模型与支持向量机的无序图像有序化方法,该方法主要流程为:首先利用词袋模型对输入的训练图像数据集构建视觉单词向量;然后利用二分类器支持向量机对生成的视觉单词进行训练,得到训练好的分类器后;再输入待检测图像数据集进行预测;最后得到每一幅图像与之相对应的具有连通性的图像,最终实现无序图像的有序化。通过该文的方法,能够较为准确地、快速地确定大规模图像之间的相互关系。实验表明,该方法显著地提高了效率,较为准确地确定具有连通性的图像,为多视图匹配,三维重建等其他应用提供了良好的数据支持。In order to improve the classification efficiency and accuracy of large-scale images and solve the problem of ordering large-scale unordered images,this paper proposed the unordered image ordering method based on bag of visual word model(BoVW)and support vector machine(SVM)by learning the theoretical knowledge of natural language and machine learning.The main process of the image ordering method is:firstly a visual word vector was constructed for the input training image data set by using the word bag model,and then the binary classifier support vector machine was used to train the generated visual words to obtain the trained classification.After the device,the images data set was input to be detected for prediction,and finally the connected image corresponding to each image was obtained,and the ordering of the unordered image was realized.Through the method of the present invention,the mutual relationship between large-scale images can be determined more accurately and quickly.Experiments showed that the method significantly improved the efficiency,and more accurately determined the images with connectivity,which provides good data support for other applications such as multi-view matching and 3 D reconstruction.

关 键 词：无序图像 有序化 视觉词袋模型 K-MEANS 图像分类 支持向量机 

分 类 号：P237[天文地球—摄影测量与遥感]