Nowadays, there is an oversupply of multimedia content, such as video, which is one of the most significant type. A great amount of video content can be found online as well as on the devices of almost every user. It is a great challenge to create a simple overview of these type of content. A trivial solution to this problem is to create a summarized clip, which aims to demonstrate the original video and summarize its content.
The method described in this paper is capable of analyzing and processing videos without any meta-information for the reason to create a summarized clip without human interaction. The key of this method is to find and select similar segments to join them as a new video.
The first step of the implementation is the video segmentation (shot detection), so I have developed a simple threshold based method for this task. This is followed by HSV histogram calculation and SURF descriptors extraction. Finally, clustering algorithm was executed to arrange the shots to groups by similarity. One shot from every group is marked as the most significant, and the union of these is the final summarized clip.
A prototype was implemented in python, using the openCV third-party library.