Connectivity thresholds and data transformations for sample supervised segment generation

Image analysis techniques based on mathematical morphology principles such as attribute filters and constrained connectivity shows promise for specific remote sensing applications; such as the identification or delineation of urban structures in VHR optical data. This could be attributed to the flexibility of these techniques to employ a range of morphological and spectral attributes, with controlling values, in segmenting imagery. In this work a sample supervised image analysis approach is investigated whereby the controlling values of attributes, and in the case of constrained connectivity, the controlling parameters, are modeled as a multidimensional search problem. The search landscape is defined via a spatial accuracy metric observing both over and under segmentation. This method is extended with the addition of data transformations, allowing for higher segmentation accuracies. Preliminary results are given comparing accuracies of this approach and a traditional segmentation method, comparing different data transformation functions and presenting some search method profiling.