Hybrid Object-based Change Detection and Hierarchical Image Segmentation for Thematic Map Updating

A hybrid object-based change detection (OBCD) method incorporating a hierarchical image segmentation strategy and cross-correlation analysis (CCA) is described and demonstrated. The proposed hybrid OBCD method was used to update an existing thematic map derived from Landsat-5 and -7 imagery (circa 2000), with imagery consisting of markedly different sensor specifications (Landsat-2, circa 1976). The proposed hierarchical image segmentation strategy successfully constrained change objects within existing land cover boundaries, avoiding the production of "sliver objects," an issue related to other image segmentation strategies used in OBCD. In combination with the CCA method, the hybrid OBCD method is capable of generating change thresholds for individual land-cover classes, providing a mechanism to limit the amount of spurious change detected. Two change threshold methods were tested: (a) change threshold values based on two standard deviations, and, (b) an unsupervised threshold method. No statistically significant difference was found between these threshold methods.