Scalable mobile quality assessment for User-generated Video

Abstract

Meerkat, Twitter's Periscope and Facebook.Live are examples of the recently evolved trend for mobile live video broadcasting. Platform providers want to offer video streams in an appealing and high quality manner. A variety of quality assessment algorithms exist inspecting different modalities and offering varying accuracy and runtime behavior. This work proposes an adaptive selection of a quality assessment algorithm for mobile generated video in order to balance real-time constraints and accuracy of an algorithm. In addition, we leverage mobile devices' capabilities to conduct video quality assessment on their own. Therefore, the processing location of an assessment is determined adaptively and executed on a central server or on any of the recording mobile devices. We show that compared with the state-of-the-art quality estimation the number of real-time quality assessments is increased by more than 25%.