eCognition already has several basic segmentations for bottom-up object primitives, like chessboard (squares), quadtree, some superpixel. But it would be useful to have hexagonal segmentation supported.
There are multiple applications built on sampling as hexagons (as opposed to square 'chessboards'), as they have better access to meaningful neighborhood information (edge to edge, no corners), more uniform connectivity of objects, higher packing density, better representation of curves (than squares). As a additional bottom-up unit, it would be very helpful for building new custom segmentation algorithms, and allow for reproduction of multiple published sampling and methodological methods based on hexagonal lattices.
Was this article helpful?