Foreground segmentation based on selective foreground model
Foreground segmentation often fails when the background and foreground exhibit similar colour distributions. In this reported work, the general foreground model as a set of latest sequential segmentations is challenged, and it is asserted that the latest samples are not always best suited for foreground modelling. Proposed is a selective foreground model composed of the best suited samples chosen from all historical segmentations based on colour histogram similarity, with the time order of segmentations being ignored. Experiments demonstrate that the selective foreground model is efficient in solving the colour similarity problem.
