This is the course project of CPSC 554M, which is done with the collaboration with Matthew Fong and Xueqin Zhang.

Goal

As more and more videos are being consumed over the web, organizing and searching for videos becomes more difficult. Videos in at-a-glance search spaces are generally represented by a title, a description, and static thumbnail to show the contents of the video. In the particular case where a user is looking for a video they have seen before, this information may not be enough to trigger a user’s memory of the video. We introduce two techniques to represent video in a personalized way that is unique to the user through the analysis of their viewing behaviour, as well as explicit bookmarking for representative thumbnail selection. We performed a formal evaluation on the techniques compared with the traditional YouTube style representation and found users were faster in selecting the video using our methods, and also preferred our method.

player

The video playing/viewing interface

filmstrip

A filmstrip consists one or several static/dynamic thumbnails, used to quickly skim through the whole video and jump to the segment of interest.

thumbnail

A thumbnail is static or dynamic images extracted from videos to represent and summarize the videos.

Control group

Youtube thumbnail interface

Lo-fi prototype

player 1 player 2 filmstrip
thumbnail 1 thumbnail 2

Me-fi prototype

player prototype medium-fi prototype of player and filmstrip

Hi-fi prototype

player prototype mixed thumbnail automatic thumbnail
video selection interface

Video

Final report

Slides