HIGH RESOLUTION REMOTE SENSING IMAGE SEGMENTATION BASED ON MULTI-FEATURES FUSION

摘要

High resolution remote sensing images contain richer information of spatial relation in ground objects than low resolution ones, which can help describe the geometric information and extract the essential features more efficiently. However, the handling difficulties due to the relative poorer spectral information, represented by phenomena of different objects with the same spectrum and the same object with the different spectrum, may cause the spectrum-based methods to fail. Besides, the inherent geometric growth in processing of traditional methods caused by growing pixels always leads to longer processing time, poorer precision, and lower efficiency. Combining the spectral features with textural and geometric features, we proposed a novel kernel clustering algorithm to segment high resolution remote sensing images. The experimental results were compared with mean shift and watershed algorithms, which validated the effectiveness and reliability of the proposed algorithm.