Salient object detection by combing eye fixation prediction and semantic segmentation

Abstract

Salient object detection is a significant part of a great deal of vision tasks and a large number of saliency detection measures have been presented of late, and omong them are complementary reciprocally. A majority of the works in computer vision concentrates on the next two particular assignments of saliency: eye fixation prediction and salient object segmentation. In this paper, we divides saliency detection into two sub tasks: salient object localization and segmentation, which can achieve more precise results of significant object detection, and also can obtain more accurate segmentation results directly. The results are validated on publicly available ECSSD and PASCAL datasets, which are superior to current state-of-the-art methods.