Abstract: In this paper, the manual digitization method and computer automatic interpretation method were integrated, and the traditional method of glacier interpretation and the objectoriented image feature extraction method were applied to extract the glacier boundary in headwaters of the Kanas River. Through the application, it was found that: (1) Human interpretation could be the best tool for extracting highlevel information from satellite imagery for many glacier types; (2) Classification methods based on threshold band ratios or normalized band differences like NDSI were proven to be accurate and as the robust methods for interpreting clean glacier ice, but they may lead to errors for the glaciers covered by surface moraine; (3) The objectoriented method could be used to extract the surface features, and its accuracy was high in extracting glacier information, especially the disturbance of cloud and snow cover could be restricted; (4) The classification method of automatic threshold decision tree needs to be further studied; (5) An interpretation of 40 glaciers in headwaters of the Kanas River was carried out, the various methods were compared, and the values of precise glacial boundary were derived. The results showed that the glaciers in headwaters of the Kanas River were holistically in a retreat trend, and their area was reduced about 23.43  km² during the period of 1959-2011.

Key words: glacier interpretation, object oriented image feature extraction method, traditional method, glacier boundary, headwaters of Kanas River