Image saliency prediction by learning deep probability model

doi:10.1016/j.image.2019.08.002

Image saliency prediction by learning deep probability model

2019-10-01

3

PRE

AI

DOI：10.1016/j.image.2019.08.002

Original

Original request for help

Save

Abstract

En 中文

Image saliency prediction is an indispensable technique in computer vision, such as video surveillance, image semantic segmentation. However, it is still a challenging task due to the low-resolution images or inaccurate predicted salient regions. With the development of convolutional neural networks (CNNs), deeply-learned feature-based methods have shown impressive performance. Thus, in this paper, we develop a deeply-learned probability model for image saliency prediction. Specifically, a series of candidate regions are generated based on BING feature, where these regions are considered as the salient regions that attract human visual attention. Subsequently, these candidate regions are ranked sequentially based on the designed ranking algorithm. Afterwards, the high-quality patches are fed into CNNs to extract deep representation of the entire image. Finally, a probability model is learned based on deep representation for image saliency prediction. To testify the performance of our method in predicted high-quality salient regions, we also conduct IQA experiment and experimental results show the effectiveness of our method.

Keywords:

Image saliency prediction

Probability model

Deeply-learned feature

Image saliency prediction by learning deep probability model

Abstract

AI Summary

Journal

Organization