Hierarchical Segmentations with Watershed and Waterfall
The watershed transform is a popular image segmentation procedure from mathematical morphology used in many applications of computer vision.
We have been working on the watershed and waterfall transforms and their use in the hierarchical segmentation of 2D or 3D images. Earlier results include novel tree-based waterfall algorithms and hierarchical segmentations for which we designed an Image Partition Forest (IPF) data structure.
Our latest results on a new parallel watershed algorithm are designed specifically for GPU implementation. The algorithm constructs paths of steepest descent and reduces these paths into direct pointers to catchment basin minima in logarithmic time, also crucially incorporating successful resolution of plateaux. Our GPU implementations of the watershed algorithm are available to download for research purposes. Please contact us for instructions on how to extract the contents.
Selected Publications
-
Two Tree−Based Methods for the Waterfall
S. M. Golodetz‚ C. Nicholls‚ I. D. Voiculescu and S. A. Cameron
In Pattern Recognition. Vol. 47(10). Pages 3276−3292. October, 2014.
Details about Two Tree−Based Methods for the Waterfall | BibTeX data for Two Tree−Based Methods for the Waterfall | Link to Two Tree−Based Methods for the Waterfall
-
Simpler Editing of Graph−Based Segmentation Hierarchies using Zipping Algorithms
Stuart Golodetz‚ Irina Voiculescu and Stephen Cameron
In Pattern Recognition. Vol. 70. Pages 44−59. October, 2017.
Details about Simpler Editing of Graph−Based Segmentation Hierarchies using Zipping Algorithms | BibTeX data for Simpler Editing of Graph−Based Segmentation Hierarchies using Zipping Algorithms
-
Path reducing watershed for the GPU
Varduhi Yeghiazaryan and Irina Voiculescu
IEEE Winter Conf. on Applications of Computer Vision. 2018.
Details about Path reducing watershed for the GPU | BibTeX data for Path reducing watershed for the GPU | Link to Path reducing watershed for the GPU
Principal Investigator
People
Share this:
