Abstract
This paper develops a new method based on intelligent feature selection for aerial image segmentation. In order to select the best feature for segmentation we considered different regions of interest like floods and roads, taking into account information about color and texture. The features investigated derived from the inter-spectral co-occurrence matrices between the color channels RGB and HSV and from LBP approach. The proposed algorithm consists of two phases: the learning and the segmentation, respectively. In the learning phase we proposed a selection algorithm of features for each region of interest based on trace of confusion matrices. In the segmentation phase we proposed a classification algorithm based on a voting scheme for each selected feature. The experimental results were done on aerial images taken by a fixed-wing type UAV.