A Coupled Co-Occurrence Matrix/Multi-Scale Segmentation Method to Extract Water from High Resolution Remote Sensing Image

Abstract

This study developed a coupled co-occurrence matrix/multi-scale segmentation method to improve extraction precision of water from high-resolution remote sensing images. Two images of Kunming city (subject A & B) were obtained from Quick Bird image gallery, pre-processed by co-occurrence matrix, and then multi-scale segmented based on inherent geometrical and geographical attributes. Water encompassed by the ring roads of the city was extracted via object-oriented information analysis with successfully removal of all shadows. Results showed that water extraction precisions had significantly increased for both subject A (68.6% → 95.2%) and B (63.0% → 92.3%), indicating superior performance of the proposed method in extracting water from complex urban environment.