Land Cover Detection with Unsupervised Clustering and Hierarchical Partitioning
An image segmentation technique relying on spatial clustering related to the single linkage approach has been put forward recently. This technique leads to a unique partition of the image domains into maximal segments satisfying a series of constraints related to local (α) and global (ω) intensity variation thresholds. The influence of such segmentation on clustering separability was assessed in this study, as well as the threshold values for segmentation maximising the cluster separability. The CLARA clustering method was used and the separability among clusters was calculated as the total separation between clusters. The clustering was applied to: (i) raw data; (ii) segmented data with varying α and ω parameters; and (iii) masked segmented data where the transition segments were excluded. The results show that the segmentation generally increases the separability of the clusters. The threshold parameters have an influence on the separability of clusters and maximising points could be identified while the transition segments were not completely included in one single cluster. The constrained connectivity paradigm could benefit land cover types/changes detection in the context of unsupervised object-oriented classification.
Unable to display preview. Download preview PDF.
- Kaufman, L., & Rousseeuw, P. J. (1990). Finding groups in data: An introduction to cluster analysis. New York: Wiley.Google Scholar
- Soille, P. (2007). On genuine connectivity relations based on logical predicates. In 14th International Conference on Image Analysis and processing, Proceedings (pp. 487–492). 14th International Conference on Image Analysis and Processing, Modena, Italy, Sep 10–14.Google Scholar