关键词: 协同视觉显著性, 公共显著目标, 深度学习, 浅层特征, 端到端模型

Abstract:

Co-saliency detection is based on the human visual attention mechanism to capture common and salient objects in a group of related images, co-saliency is widely used in co-segmentation, object detection and other fields. The existing methods of co-saliency detection are summarized and experimentally evaluated. Firstly, according to the difference of features, all methods are divided into two categories：one is the traditional methods with low-level features, the other is the deep learning methods with deep-level features. Secondly, on the basis of the different ways of obtaining inter-saliency and building models, the two kinds of methods are introduced and theoretically analyzed. Then the state-of-the-art methods are subjectively and quantitatively evaluated in the two public datasets. Finally, the existing methods are qualitatively summarized, the existing problems are analyzed in the present research, and the future work is prospected.

Key words: co-saliency, common salient object, deep learning, low-level feature, end-to-end model