Superpixel Tracing on Graphs
Processing streaming video is a challenging task due to the quantity of data involved. Considering a million pixels per frame, a video generated at 30 frames per second (fps) would generate 30 million pixels every second. In order to get a handle on the complexity of the problem, pixels are grouped into consistent image regions called superpixels in every frame. This reduces the problem from one of processing million pixels to processing a few thousand consistent image regions a.k.a superpixels. The primary goal of this research investigation was to group superixels belonging to a certain object of interest, thereby visually tracking the object over time. This can be applied to the segmentation of consumer video, as well as bio-microscopic imagery like Electron Micrographs. The main idea was to group superpixels as nodes in a hypergraph and perform graph diffusion to estimate a probability for every superpixel as belonging to the objects being tracked. The system could scale gracefully to track a large number of objects without a noticeable increase to run-time complexity.
Relevant Publications
Scalable Tracing of Electron Micrographs by Fusing Top Down and Bottom Up Cues using Hypergraph Diffusion
Vignesh Jagadeesh, Min Chi Shih, B.S.Manjunath, K.Rose
Medical Image Computing and Assisted Intervention, MICCAI 2012.
Multiple Structure Tracing in 3D Electron Micrographs
Vignesh Jagadeesh, Nhat Vu and Bangalore S. Manjunath
Medical Image Computing and Assisted Intervention, MICCAI 2011.
Vignesh Jagadeesh, Min Chi Shih, B.S.Manjunath, K.Rose
Medical Image Computing and Assisted Intervention, MICCAI 2012.
Multiple Structure Tracing in 3D Electron Micrographs
Vignesh Jagadeesh, Nhat Vu and Bangalore S. Manjunath
Medical Image Computing and Assisted Intervention, MICCAI 2011.