The information identification is the most difficult in the application of high-resolution remote sensing images.This paper focused on the method of QuickBird image's information extraction.Six kinds of land cover types: road

water

forest

agriculture

nuke and residence in the study area

Zhuzhou

Hunan

were identified by visual interpretation

supervised classification and non-supervised classification respectively

and the accuracy is 98.2%

72.64% and 60.71% correspondingly.At the same time

pre-processed images of QuickBird

ETM+ and TM were identified by supervised classification and non-supervised classification respectively and the accuracy of QuickBird image classification is lower than that of ETM+ and TM.This showed that resolution's improvement couldn't raise the accuracy of classification by the method of traditional classification.This paper avoided the traditional classification based on pixel-to-pixel and applied new method of classification based on object's gray character and shape characters.QuickBird image based on pixels was changed into new image based on objects first by segmentation

then models of measuring objects' area

perimeter

length

width

length/width

rectangle and roundness were built.Six kinds of land types were classified again by computer stimulating visual interpretation in the study area

the composite accuracy of classification is up to 91.6%

this showed that the method based on objects is a very effective way to improve the accuracy of classification of QuickBird image.