Object-based symmetric difference for land surface segmentation scale parameter optimisation

Abstract

Although multi-resolution segmentation (MRS) has been shown to delimit geomorphological boundaries more accurately than other available segmentation procedures, the selection of the appropriate parameters remains a challenge. The optimisation of scale has largely been focussed on deriving suitable scale parameters (SPs). Local variance (LV) has been shown to exploit spatial autocorrelation for identifying suitable SPs. However, further investigation into SP selection for land surface segmentation is necessary. This paper demonstrates how symmetric difference (SD) can be used to detect suitable SPs when segmenting continuous land surfaces. Comparisons with an established suitable SP selection technique (ESP 2) are made and, finally, it is concluded that SD provides a simple and effective way of selecting suitable SPs for land surface segmentation.