Feature fused multi-scale segmentation method for remote sensing imagery
文献类型:会议论文
作者 | Chen, T.Q.1; Liu, J.H.1; Wang, Y.H.1; Zhu, F.1; Chen, J.2; Deng, M.2 |
出版日期 | 2016 |
会议日期 | 2015-08 |
会议地点 | Guangzhou, China |
页码 | 741-744 |
英文摘要 | Image segmentation is a key step in Object-Based Image Analysis (OBIA) for High Spatial Resolution Remote Sensing Imagery (HSRI). Previous segmentation method used image spectrum and structure information only, so that it is difficult to obtain good segmentation results. A novel segmentation method based on multiple feature fusion and multi-scale region merging was proposed in this paper. Firstly, initial over-segmentation objects were obtained by marked watershed transform. Then, the spectrum, texture, shape, area, and edge intensity information were combined to build the merging cost of adjacent objects, and all objects were mapped to adjacent graph nodes with their merging cost. Finally, the nodes in Region Adjacency Graph (RGA) were merged to get multi-scale segmentation result. Experiment results indicate that the proposed method can produce nested multi-scale segmentation results with accurate boundaries, and achieve a better performance compared to Mean-shift and ENVI segmentation method. © 2016 Taylor & Francis Group, London. |
产权排序 | 1 |
会议录 | Advances in Energy, Environment and Materials Science - Proceedings of the International Conference on Energy, Environment and Materials Science, EEMS 2015 |
会议录出版者 | CRC Press/Balkema |
学科主题 | 数理科学和化学 |
语种 | 英语 |
ISBN号 | 9781138029316 |
源URL | [http://ir.opt.ac.cn/handle/181661/27877] |
专题 | 西安光学精密机械研究所_遥感与智能信息系统研究中心 |
作者单位 | 1.Laboratory of Remote Sensing and Intelligent Information System, Xian Institute of Optics and Precision Mechanics of CAS, Xian, China 2.Department of Geo-informatics, Central South University, Changsha, China |
推荐引用方式 GB/T 7714 | Chen, T.Q.,Liu, J.H.,Wang, Y.H.,et al. Feature fused multi-scale segmentation method for remote sensing imagery[C]. 见:. Guangzhou, China. 2015-08. |
入库方式: OAI收割
来源:西安光学精密机械研究所
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。