Saliency detection on sampled images for tag ranking

Image saliency contributes to rank the unordered tags extracted from social media, but the existing saliency detection methods can hardly efficiently handle massive images in tag ranking. In this paper, we focus on improving the efficiency of saliency detection methods by applying them on the sampled images with suitable resolutions. We extensively investigate the influence of image resolution to saliency detection performance of the typical methods, and summarize a sampling strategy for different categories of salient object detection methods. Furthermore, we validate the effectiveness of the sampling strategy by applying the salient object detection methods on the sampled images with the selected resolutions in tag ranking. The experimental results show that sampling can significantly improve the efficiency of the existing salient object detection methods without obvious loss in effectiveness.