A Novel Visual Assisted Technique for Video Summarization Using Attention-Based Key-Frames Selection

The advancements in digital videos led to new demands and novel research opportunities. Many tools and methods are developed over time for efficient video browsing. One of the widespread methods is to extract keyframes to generate summary of the video that present the complete message of the video. The video summarization uses visual attention model, in which noteworthy frames are extracted as a keyframe based on human attention model. It is a workable solution for video summarization, however high computational cost limits its application in practical scenario. In this paper, we propose a novel keyframe extraction (KFE) method based on improved visual attention model. The computational cost and expensive calculations are minimised by utilising dynamic visual highlights that are based on the temporal gradient as an alternative of the conventional optical flow procedures. Secondly, introducing an effective method by means of a discrete cosine conversion is utilised for the static visual salience. Moreover, static and dynamic visual attention system of measurements are combined through a nonlinear weighted fusion method. The practical outcomes show that proposed method is efficient and high standard for the video summarization.