摘要：How to avoiding the disadvantage of the pixel-wise image process methods has been a open problem. An object-wise method was proposed to extract the water body of remote sensing image, which is locally excitatory globally inhibitory oscillator networks (LEGION). This oscillation network could bind an object based on the similar property. The oscillation network in this paper used the normalized difference water index (NDWI) to encode the binding of pixels of water body. Because of the good coherence of spectrum of water body on the remote sensing image, the pixels of the water surface have strong couple weight and can overcome the globally inhibition in the network. Additionally the special spectrum property also was considered, it could reduce the disturber of the other ground object which has the similar spectra property. The proposed method combined both pixel-wise and object-wise scale. Thus, the proposed extraction method was robust to noise, and could keep the detail information of water body. The outputs were according to the human perception mechanism and can be used in the higher image analysis and reorganization. The results of experiment validate the effective of the proposed method.
