Comparing classification metrics for labeling segmented remote sensing images

Image segmentation and labelling are the two concep-tual operations in image classification. As the remote sens-ing community uses more powerful segmentation proce-dures with spatial constraint, new possibilities can be ex-plored for labelling. Instead of assigning a label to a sin-gle observation (pixel), whole segments of image are la-belled at once implying the use of multivariate samples rather than pixel vectors. This approach to image classi-fication also offers new possibilities for using a priori in-formation about the classes such as existing maps or object signature libraries. The present paper addresses the two is-sues. First a labelling scheme is presented that gathers ev-idence about the classes from incomplete a priori informa-tion using a “cognitive reasoning ” approach. Then, five dif-ferent metrics are compared for the label assignment and are combined through a voting scheme. The results show that very different results can be obtained depending on the metric chosen. The metric combination through voting, be-ing a suboptimal approach does not necessarily provide the best results but could be a safe alternative to choosing only one metric. 1