Motion-based, multi-stage video segmentation with motion boundary refinement
AssigneeHewlett-Packard Development Company, L.P.
InventorAndrew J Patti
Life of this patent
Estimating the term…
Abstract
A motion-based, multi-stage segmentation of a video frame is provided. A first segmentation stage identifies motion boundaries in a plurality of sub-blocks of the video frame. The motion boundaries are identified by computing an occlusion-insensitive localized motion cost and minimizing a MAP-MRF energy based on the localized motion cost. A second segmentation stage refines the motion boundaries by adding a color cost to the localized motion cost in the MAP-MRF energy.
Claims19
A method for performing a motion-based, multi-stage segmentation of a video frame, the method comprising:
- identifying motion boundaries in a plurality of sub-blocks of the video frame in a first segmentation stage, the motion boundaries identified by computing an occlusion-insensitive localized motion cost on a 3-level image map and minimizing a MAP-MRF energy based on the localized motion cost;
- and
- refining the motion boundaries in a second segmentation stage by adding a color cost to the localized motion cost.
A non-transitory computer-readable medium having instructions stored thereon executable by a processor to:
- form a 3-level image map for a video frame, the video frame divided into sub-blocks;
- compute a primary motion for at least one sub-block based on the 3-level image map;
- determine a secondary motion for the at least one sub-block;
- compute an occlusion-insensitive localized motion cost for the at least one sub-block involving the primary and secondary motions;
- determine a set of motion boundaries for the at least one sub-block based on the localized motion cost;
- and
- refine the set of motion boundaries for the at least one sub-block based on color information of the at least one sub-block.
A computing system, comprising:
- shared memory resources;
- and
- a processor communicatively coupled to the shared memory resources, wherein the processor executes instructions stored in at least one of the shared memory resources to:
- form a graph to minimize a MAP-MRF energy defined by an occlusion-insensitive localized motion cost computed on a 3-level image map in a first segmentation stage to identify a set of motion boundaries for a video frame;
- and
- refine the set of motion boundaries based on a color cost added to the localized motion cost in a second segmentation stage.
Dependencies are inferred from the claim text.