Efficient Video Breakup Detection and Verification

Automatic quality assessment for audiovisual media is an important task for several steps of the media production, delivery and archiving processes. In this paper we focus on the semi-automatic quality inspection of videos and propose a novel algorithm for the detection of severe visual distortions, commonly termed as `Video Breakup'. In order to enable the ecient human interaction with quality analysis results we present the extended `Quality Summary Viewer' application, which enables the user to eciently verify the detected events. Moreover it allows the user to quickly grasp the frequency and strengths of these visual impairments in the content and allows for its overall quality appraisal. Evaluation on a huge, challenging dataset shows, that our algorithm is able to detect up to 97.3% of the events annotated by domain experts. Depending on the application specific needs, we can reach a false detection rate of only 0.1-1.5 per minute.