Multispectral image segmentation using Gabor filtering and local homogeneity analysis with application to MSG

Image segmentation is an essential stage in many computer visions tasks. In the last decades, many segmentation methods based on the gray levels of the texture image have been proposed. However, the use of the spectral information improves the segmentation accuracy of the spectral textures. This paper describes a multispectral image segmentation method, which consists of two steps. In the first step, the Gabor filtering is applied to the spectral components of the texture images. By using Gabor analysis, the multispectral texture images can be segmented in multiples scales and orientations. To get an initial segmentation of the image, the obtained spectral textures features are first combined and then classified using the FCM algorithm. In the second step, local homogeneity analysis is carried out using Deng´s criterion for “good” segmentation. The homogeneity analysis is applied to the extracted regions to get the final result of the segmentation. The proposed method has been tested successfully using synthetic and Meteosat Second Generation (MSG) satellite images.