Accurate Visual Features for Automatic Tag Correction in Videos

We present a new system for video auto tagging which aims at correcting the tags provided by users for videos uploaded on the Internet. Unlike most existing systems, in our proposal, we do not use the questionable textual information nor any supervised learning system to perform a tag propagation. We propose to compare directly the visual content of the videos described by different sets of features such as Bag-Of-visual-Words or frequent patterns built from them. We then propose an original tag correction strategy based on the frequency of the tags in the visual neighborhood of the videos. Experiments on a Youtube corpus show that our method can effectively improve the existing tags and that frequent patterns are useful to construct accurate visual features.