Description

When Object Based Image Analysis (OBIA) is applied in the context of damage mapping (DM) a key role still plays the initial segmentation’s quality since it is the entry point for the following image analysis. The less the initial segmentation needs to be improved in the follow-up process the faster and the more accurate the results will be available. A typical segmentation strategy in OBIA is to start with a knowledge-free pre-segmentation (e.g. hexagon segmentation or super-pixels) and to aggregate the initial results to more meaningful objects. The latter is often knowledge based either by explicit rules described by an experiences operator or based on rules describing the semantics of the objects to be segmented.

However, in the past not much systematic research has been spent on analyzing different pre-segmentation methods for their suitability in the context of DM. Especially the aspect of repeatability and robustness plays an important role in DM.

Objectives and Methodology

In this research the following major questions should be addressed:
- What impact can pre-segmentations have on the follow-up image analysis? Is there a positive impact at all? What particular analysis steps do benefit from a pre-segmentation, and how?
- To what degree do appropriate pre-segmentations increase the robustness and transferability of already established OBIA methods in DM?
- Which pre-segmentations are appropriate and for which application context in DM (damage mapping, landslide mapping and monitoring, floods, wild fires etc.)?

Further reading