Ultra high definition video saliency database

Visual attention plays an important role in image and video processing. Nowadays, high definition (HD) techniques have been widely used. And ultra high definition (UHD) is becoming more and more popular. However, existing researches in visual attention mainly focus on relatively low resolution videos or images. There is very limited studies in visual attention of UHD videos. In this paper, we built a Ultra High Definition (4k) Video Saliency Database. Using this database, we explored the characteristics of visual attention related to ultra high definition videos. A concept of aggregation maps (AGM) for videos is put forward to better analyse the characteristics of visual attention of videos. Through the experiment, we find that there exist fairly strong correlations between the video resolution and visual attention behaviors. We also find that people tend to focus on the center of videos of relatively low resolution. The database will be make publicly available at gvsp.sjtu.edu.cn soon.