Foreground Object Segmentation with Objectness Measure
Zhu Junguang · 2016
This paper proposes a novel model to address the problem of image segmentation with objectness measure.Recently, many objectness measures are proposed, which aims to generate candidate windows to localize the possible objects in the image.Consider to combine this useful object location piror into a foreground segment model.Specifically, a Conditional Random Filed model is constructed on superpixels graph, and it efficiently incorporates objectness measure, color distribution and appearance similarity.Expermental results on a extended GrabCut dataset demonstrate that the proposed model can yield a foreground object segmentation of better quality.