Segmentation-assisted classification for IKONOS imagery

A segmentation-based method is presented for classification of multispectral imagery from IKONOS satellite. Three different types of subimages pertaining to natural environment were used from IKONOS image to test the segmentation-based classification approach. Initially multispectral threshold values were obtained by global thresholding. Based on these threshold values, segments were grown in the image. The segmented image obtained by this step was further refined by merge score criteria. The refined segmented image obtained from above procedure was subjected to Gaussian maximum likelihood and minimum distance to means classifications. The classification results have shown that the proposed approach yielded statistically significant, different and better results than the conventional per-pixel classifiers.