International Journal on Advanced Science, Engineering and Information Technology, Vol. 11 (2021) No. 6, pages: 2488-2494, DOI:10.18517/ijaseit.11.6.12888

Deep Learning-based Video Summarization

Myoungchan Seo, YoungJin Suh, Kyuman Jeong


With the development of communication technology, many different kinds of media transmission have become popular. Among various media, video is the most popular media these days. However, users need to spend much time watching the whole video content. Due to the characteristics of video media, many users tend to playback video content quickly or even stop watching in the middle. Some websites provide summary images by capturing only important frames of video content, which is called a video summary. Users can shorten the viewing time by only watching the summary results. In particular, it is highly useful because content such as news articles or speeches can be delivered and utilized quickly. Since video summarization is a labor-intensive task, there is an increasing demand for research on automation techniques. In this paper, an automated process to solve the temporary problem of existing video summary techniques is proposed. The proposed method improves the existing video summarization methods that have been performed manually through human labor by developing artificial intelligence technology that can effectively perform content delivery using video summary automation. In the preprocessing process, the information transfer unit is partitioned using optical flow. In the following process, CNN (Convolutional Neural Network) is used as an in-depth learning method for feature extraction. The results show the efficiency of the proposed algorithm, and some future work will be given in the end.


Deep learning; video summarization; scene extraction; convolutional neural network; optical flow.

Viewed: 1271 times (since abstract online)

cite this paper     download