Color- and texture-based image segmentation for improved forest delineation

This paper concentrates on the delineation of forest boundaries from aerial images with focus on spatially contiguous and reproducible results for the Swiss National Forest Inventory. Because of the poor performance of common edge models to extract natural vegetation boundaries, this paper presents a combined method of image segmentation and wavelet-based texture features for the delineation of forest. The selected J-measure-based segmentation method has been found to be useful to produce initial segmentation results, but lacks a semantic concept for forest vegetation. To overcome this conceptual limitation, the combination with wavelet transformation gives access to additional texture features and leads to a robust approach to obtain proper forest boundaries. Preliminary results are encouraging regarding the better agreement compared with maximum-likelihood classification results.