QQ客服 官方微博 反馈留言

	Water extraction in high resolution remote sensing image based on hierarchical spectrum and shape features
	Li, Bangyu (1) ; Zhang, Hui (1) ; Xu, Fanjiang (1)
	2014
会议名称	35th International Symposium on Remote Sensing of Environment, ISRSE 2013
会议日期	April 22, 2013 - April 26, 2013
会议地点	Beijing, China
通讯作者	Li, B.(bangyu@iscas.ac.cn)
中文摘要	This paper addresses the problem of water extraction from high resolution remote sensing images (including R, G, B, and NIR channels), which draws considerable attention in recent years. Previous work on water extraction mainly faced two difficulties. 1) It is difficult to obtain accurate position of water boundary because of using low resolution images. 2) Like all other image based object classification problems, the phenomena of "different objects same image" or "different images same object" affects the water extraction. Shadow of elevated objects (e.g. buildings, bridges, towers and trees) scattered in the remote sensing image is a typical noise objects for water extraction. In many cases, it is difficult to discriminate between water and shadow in a remote sensing image, especially in the urban region. We propose a water extraction method with two hierarchies: the statistical feature of spectral characteristic based on image segmentation and the shape feature based on shadow removing. In the first hierarchy, the Statistical Region Merging (SRM) algorithm is adopted for image segmentation. The SRM includes two key steps: one is sorting adjacent regions according to a pre-ascertained sort function, and the other one is merging adjacent regions based on a pre-ascertained merging predicate. The sort step is done one time during the whole processing without considering changes caused by merging which may cause imprecise results. Therefore, we modify the SRM with dynamic sort processing, which conducts sorting step repetitively when there is large adjacent region changes after doing merging. To achieve robust segmentation, we apply the merging region with six features (four remote sensing image bands, Normalized Difference Water Index (NDWI), and Normalized Saturation-value Difference Index (NSVDI)). All these features contribute to segment image into region of object. NDWI and NSVDI are discriminate between water and some shadows. In the second hierarchy, we adopt the shape features to remove more shadows. The water polygons are generated by vectorization algorithm after water segmentation, and then some shape parameters (Compact, Critical Point and Symmetry) are considered to remove shadow. To evaluate the performance of the proposed method, we collect several Quick Bird images at 0.61-m resolution which are acquired in May 2009 at GUANGZHOU province of China. The proposed method is compared with four other methods in water extraction, including pixel-based segmentation by NDWI, Mean-sift segmentation by NDWI, and SVM with different channels. Experimental results show that the proposed method can increase extraction accuracy and reduce the influence of shadows.
英文摘要	This paper addresses the problem of water extraction from high resolution remote sensing images (including R, G, B, and NIR channels), which draws considerable attention in recent years. Previous work on water extraction mainly faced two difficulties. 1) It is difficult to obtain accurate position of water boundary because of using low resolution images. 2) Like all other image based object classification problems, the phenomena of "different objects same image" or "different images same object" affects the water extraction. Shadow of elevated objects (e.g. buildings, bridges, towers and trees) scattered in the remote sensing image is a typical noise objects for water extraction. In many cases, it is difficult to discriminate between water and shadow in a remote sensing image, especially in the urban region. We propose a water extraction method with two hierarchies: the statistical feature of spectral characteristic based on image segmentation and the shape feature based on shadow removing. In the first hierarchy, the Statistical Region Merging (SRM) algorithm is adopted for image segmentation. The SRM includes two key steps: one is sorting adjacent regions according to a pre-ascertained sort function, and the other one is merging adjacent regions based on a pre-ascertained merging predicate. The sort step is done one time during the whole processing without considering changes caused by merging which may cause imprecise results. Therefore, we modify the SRM with dynamic sort processing, which conducts sorting step repetitively when there is large adjacent region changes after doing merging. To achieve robust segmentation, we apply the merging region with six features (four remote sensing image bands, Normalized Difference Water Index (NDWI), and Normalized Saturation-value Difference Index (NSVDI)). All these features contribute to segment image into region of object. NDWI and NSVDI are discriminate between water and some shadows. In the second hierarchy, we adopt the shape features to remove more shadows. The water polygons are generated by vectorization algorithm after water segmentation, and then some shape parameters (Compact, Critical Point and Symmetry) are considered to remove shadow. To evaluate the performance of the proposed method, we collect several Quick Bird images at 0.61-m resolution which are acquired in May 2009 at GUANGZHOU province of China. The proposed method is compared with four other methods in water extraction, including pixel-based segmentation by NDWI, Mean-sift segmentation by NDWI, and SVM with different channels. Experimental results show that the proposed method can increase extraction accuracy and reduce the influence of shadows.
收录类别	CPCI ; EI
会议录出版地	Institute of Physics Publishing
语种	英语
ISSN号	17551307
内容类型	会议论文
源URL	[http://ir.iscas.ac.cn/handle/311060/16517]
专题	软件研究所_软件所图书馆_会议论文
推荐引用方式 GB/T 7714	Li, Bangyu ,Zhang, Hui ,Xu, Fanjiang . Water extraction in high resolution remote sensing image based on hierarchical spectrum and shape features[C]. 见:35th International Symposium on Remote Sensing of Environment, ISRSE 2013. Beijing, China. April 22, 2013 - April 26, 2013.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

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。