Optimal shot boundary detection based on robust statistical models

The paper presents a novel statistical framework for optimal shot boundary detection based on minimization of the average detection error probability. The required statistical functions are modeled using a robust metric for visual content discontinuities (based on motion compensation) and by taking into account the knowledge about the shot length distribution and visual discontinuity patterns at shot boundaries. Major advantages of the proposed framework are its robust and by far sequence-independent detection performance, as well as the possibility for simultaneously detecting different types of shot boundaries.