Performance Evaluation Protocol for Video Scene Detection Algorithms

In this paper, we address the problem of performance evaluation (PE) of video scene detection algorithms. An objective and comprehensive protocol for performance evaluation and comparison of different scene detection algorithms is proposed. Unlike traditional precision-recall measures, the proposed PE measures treat false alarms unequally, and take the extent of incorrectness of the wrongly detected scene boundaries into consideration, which can generate more accurate and more reasonable PE results. The protocol has been applied not only on comparing different scene detection algorithms but also on determining the most appropriate thresholds for a given scene detection algorithm.