Evaluating interactive search in videos with image and textual description defined target scenes

For non-expert users, basic video players with simple controls are often the tools of choice when searching for specific scenes within videos. Basic actions like play, fast-forward, fast-reverse, positioning using the seeker-bar are familiar and are often preferred over advanced retrieval tools when it comes to interactive search and browsing. In order to get a deeper insight into how users approach such tasks with a simple player, we analyzed the search behavior of two groups of 17 users in two different setups. In both setups the users had to locate an approximately 20 seconds long video sequence within an hour long news video. In the first setup, the users were presented with pictures uniformly sampled from within the target scene as well as a textual description. In the second setup, only the sampled pictures were provided for each target. We asses the impact, in terms of search behavior, of identifying target scenes through sampled images, respectively sampled images and textual descriptions.