Multiscale Image Segmentation and Classification with Supervised ECHO of High Spatial Resolution Remotely Sensed Imagery

This paper presents a new method of supervised extraction and classification of homogenous object(ECHO),aiming to enhancement the multiscale homogeneity in a local neighborhood of high resolution remotely sensed imagery.This method fused multiscale spectral and spatial information using a series of homogeneous regions such as 2×2,4×4 and 8×8 window sizes.Experiment proved that the proposed method outperforms the pixelwise MLC and the single scale ECHO method,with Washington DC data set obtained by HYDICE sensor and Beijing data set obtained by QuickBird.