Segmentation and classification of remote sensing images using confident marker selection

A new method for segmentation and classification of agricultural remote sensing images is proposed. The method is based on region growing algorithm from confident selected markers. We propose a straightforward spectral-spatial method to choose the most reliable classified pixels in order to define suitable markers. We use probability estimates which are obtained from support vector machine (SVM) classification for computing entropy criterion of each pixel in order to determine spectral confident pixels. Erosion technique is used to extract spatial confident pixels of initial classification map which are the pixels far from the borders (spatial boundaries). By combining these procedures, confident markers are defined and by construction of a minimum spanning forest (MSF) from these markers the final spectral-spatial classification map is obtained. Experimental results show that the proposed method improves dramatically classification accuracy in comparison with pixelwise classification.