Fast, gradual, super-voxel-based spatiotemporal video segmentation method

摘要

The two-step architecture of the supervoxel-based spatial temporal video segmentation method ensures speed and scalability. The computationally intensive first step uses a highly efficient super-voxel segmentation method. The second step is done over pre-grouped super-voxels, hence has much lower temporal and spatial complexity. The progressive segmentation scheme deployed in the first step enables segmenting huge input volume part by part, without loading all the data into the memory which may be infeasible. At the same time, the progressive segmentation is able to effectively prevent seam artifacts, leading to segmentation results virtually identical to those of whole volume processing.