Segmentation-Based Outlier Elimination
Segmentation-Based Outlier Elimination
Despite the consistency checks in the reverse matching and the redundancy from using two ADS stereo pairs in,
the SGM result might still contain some outliers, e.g. in challenging areas such as water or trees. It has been
proposed by Hirschmüller (2005) – and integrated into various implementations of SGM since – to assume that
small isolated patches, which significantly differ in height (or disparity) from their neighborhood, are most likely
errors. For the required segmentation, neighboring pixels in which disparities differ by more than 1 are considered
part of different segments. Small segments are then removed from the disparity map; the threshold in the XPro DSM
Extractor is 50 pixels in area.