GLACIER INFORMATION EXTRACTION 
BASED ON MULTI-FEATURE COMBINATION MODEL 
J.M. Gong * *, X.M. Yang? T. Zhang *, X. Xu*, Y.W. He 2 
* State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Science and 
Natural Resources Research, CAS, 100101 Beijing, China - (gongjm, yangxm, zhangt, xux, heyw)@lreis.ac.cn 
Commission VI, WG VI/4 
KEY WORDS: Multi-feature Combination Model; Image Segmentation; Remote Sensing; Glacier Landform; Information 
Extraction; Feature Description 
ABSTRACT: 
As a typical landform class of Qinghai-Tibetan Plateau, glacier is widely distributed in alpine terrain. However, field measurement is 
impossible in those areas because of complex terrain and adverse weather. At first, on the basis of analyzing the features of glacier 
image spectrum, object shape, spatial relations and environment distribution including terrain and climate, this paper combines and 
develops the existing feature description algorithm of object-oriented method. Secondly, we build a series of combined extraction 
models for glacier landform by using high resolution remote sensing images and DEM data. At last, based on object-oriented method 
and combined extraction models, this paper tests glacier landform extraction in Qinghai-Tibetan Plateau study area of Western 
Mapping Project. Results demonstrate that the multi-feature combination model is feasible. The researches introduce a new approach 
to remote sensing auto-extraction of glacier information which is difficult to measure in the field. Moreover, the paper explores some 
new ideas in the researches of monitoring glacier ablation and climatic change. 
1. INTRODUCTION the rich information of shape, texture and context is treated as 
noise, resulting in the phenomenon of misjudgment and 
In recent years, the fact that the world's glacier is accelerating misclassification in image interpretation process (Chen, 2006; 
ablation has caused great concerns of native and international Tan, 2007). Compared to the traditional methods of image 
scholars (Aizen, 2007; Noferini, 2009; Scherler, 2008; Wolken, analysis, object-oriented approach, which primarily produces a 
2006; Yao, 2006). With an area of more than 2.5 million km?, ^ certain criterion of polygon objects composed of homogeneous 
an average elevation of 4500m above, Qinghai-Tibet Plateau is pixel cluster by image segmentation, is used in glacier 
located in Eurasia, which has unique alpine climatic information extraction of multi-feature combination. Further, 
characteristics of Qinghai-Tibet Plateau. As a typical landform we can extract varieties of landform classes based on the 
landscape of the plateau, glacier is widely distributed in the analysis of object features, including spectrum, shape, texture 
alpine region. According to statistics of glacier catalogue in and spatial relations. This paper attempts to find a suitable 
2004, the glaciers area of Qinghai-Tibet Plateau covers about combination model to describe glacier features and achieve the 
47000 km’, accounting for more than 80% total glacier area of purpose of glacier information automatic extraction, which 
China. Permafrost area is about 1.5 million km”, accounting for mainly includes the following steps: image segmentation, 
more than 60% total area of Qinghai-Tibet Plateau (Pu, 2004). feature description and glacier information extraction of multi- 
Glacier ablation of Qinghai-Tibet Plateau has great research feature combination model. 
value to the worldwide climate change. Therefore, the 
monitoring of glacier change is an important topic of current 2. GLACIER IMAGE SEGMENTATION ALGORITHM 
global change research. In the process of glacier ablation, 
retreat and thinning of glacier result in various types of glacial Image segmentation is a critical step of information extraction 
landforms. Based on remote sensing techniques, this paper ^ based on object-oriented method, in which its segmentation 
explores to find a quick automatic extraction method of glacier ^ quality has a direct impact on image analysis accuracy. Image 
information, which has great research significance to segmentation is a process that image is expressed as a number 
monitoring glacier ablation and global climate change. of region set , which fulfils the homogeneity standard including 
spectrum, shape and other features description while meeting 
Solely based on a single gray level or spectral information, ^ the heterogeneity standards among the adjacent regions 
traditional remote sensing image analysis methods often focus (Definiens, 2007). Starting from the pixel, the smaller 
on gray-level statistical characteristics of image and calculate homogeneous objects gradually merged into a large 
its variance, mean and other statistical parameters to achieve the homogeneous image object by using region merging approach 
purpose of image analysis. However, because the interrelated of bottom-up. Usually, in accordance with the different research 
information of spatial characteristics contained in image is ^ purpose, image segmentation approaches are generally 
ignored, those image analysis methods often limit the accuracy classified as three categories based on edge, region, and the 
of information extraction, even make wrong judgments since mixture of the previous two. 
  
* Corresponding author: Jianming Gong, E-mail: gongjm(lreis.ac.cn. 
128