Improving color based video shot detection

Video shot detection plays a fundamental role in video access/analysis. Among many shot detection schemes, color based schemes seem to be the most widely used ones which can be applied to videos of different domains. However color based techniques are not always very efficient. For example, over-detection (high false-alarm rate) may occur, especially for difficult videos such as home video which often contain significant object motion and illumination change. In this paper we propose a new learning method to improve the color-based shot detection schemes by employing a better similarity measure obtained through a minmax optimization procedure. This new measure is not tied to a particular feature such as color, hence it can be applied to video browsing, indexing based on other features as well