Poster abstract details

Max-tree and alpha-tree are hierarchical representations of image which are highly efficient in the detection of objects and segmentation in many application fields. However, the advance of image acquisition techniques has lead to a larger database of images with higher resolution, dimension and dynamic range, which makes the construction and the interpretation of max-tree and alpha-tree of those images less feasible. Here, we introduce a scalable max-tree and alpha-tree algorithm, which constructs a scale space of trees with respect to the resolution, the dimension and the dynamic range. The proposed method creates a scale space of the image that is easy to interpret, and provides a straightforward way for efficient parallel implementation since coarser scales could split a tree into independent subtrees that can be used in parallel construction of finer scales. The proposed method is tested on FDS data and remote sensing data to show the construction time and the resulting scale space.