PROJECT TITLE :

Inner and Inter Label Propagation: Salient Object Detection in the Wild - 2015

ABSTRACT:

During this paper, we propose a completely unique label propagation-based methodology for saliency detection. A key observation is that saliency in a picture will be estimated by propagating the labels extracted from the most sure background and object regions. For many natural pictures, some boundary superpixels serve as the background labels and therefore the saliency of different superpixels are determined by ranking their similarities to the boundary labels primarily based on an inner propagation scheme. For pictures of complicated scenes, we have a tendency to additional deploy a three-cue-center-biased objectness measure to pick out and propagate foreground labels. A co-transduction algorithm is devised to fuse both boundary and objectness labels primarily based on an inter propagation theme. The compactness criterion decides whether the incorporation of objectness labels is critical, so greatly enhancing computational efficiency. Results on five benchmark information sets with pixelwise accurate annotations show that the proposed technique achieves superior performance compared with the most recent state-of-the-arts in terms of various evaluation metrics.