Video abstraction inspired by human visual attention models

This paper describes an application of three different state-of-the-art human inspired visual attention models to video abstraction. Two types of video abstractions, video skim and key frame extraction, are performed over three different genres of videos. Qualitative and quantitative results are reported based on user studies and statistical tests. A comparison is made with human made video abstraction set to benchmark the current analysis. We report some abstraction similarities at scene level showing that all three models are successful in capturing semantic content despite having anatomical differences. However different models are more suitable for different genres of videos.