关键词: 多光谱图像分类, 多相位水平集, Parzen窗, 密度估计, 光谱异质

Abstract: The characteristics of spectral variations and details disturbances bring great disadvantages to remote sensing image classification.To solve the problem，this paper presents a novel multi-phase level set method based on non-parametric density estimation technique.Firstly，Parzen window non-parametric density estimation technique is introduced into the level set framework so as to improve accuracy of probability density estimation.In addition，a new energy term is introduced using textures derived from Gabor filter，in order to strengthen texture analysis of multi-spectral imagery.Experimental comparisons and analysis show that the proposed model performs an effectiveness of alleviating those problems mentioned above，in the case of little priori knowledge of study area.

Key words: multispectral image classification, multiphase level set, Parzen window, density estimation, spectral inhomogeneity