Towards a standard protocol for the evaluation of temporal video segmentation algorithms

This paper proposes a general framework to evaluate and compare several temporal video segmentation algorithms. The problems that must be solved to confront di erent methods summarise as gathering a common content set to test the methods, building a reference segmentation, establishing the rules to match the results with the reference, and providing a quality measure. Some solutions to these problems are given in this study and are applied for evaluating di erent methods developed in various contexts. The paper concludes by presenting results obtained on practical tests.