Efficient Content Processing for Adaptive Video Delivery

Abstract

Delivery of video using Adaptive Bit Rate (ABR) streaming requires creating multiple representations of the same input video at different profiles encoded at different bitrates, frame rates and/or spatial resolutions. Selection of these profiles is currently done in a static fashion. We present experimental data to show that the static selection of profiles is not optimal given the wide variability in video content and provide recommendations for dynamic profile selection. Multiple profiles require that the same input video stream be processed and encoded or transcoded several times. This requirement increases the computational complexity of an ABR transcoder by many folds compared to conventional legacy delivery of a single stream video. We present a cooperative transcoding technique that can substantially reduce the processing requirement of generating multiple representations as well as produce a uniform quality of user experience across the multiple representations. Two different scenarios are addressed: generation of multiple bitstreams at different bitrates with the same spatial resolutions, and generation of streams at different bitrates and resolutions. Details on savings in computational complexity of up to 70% using co-operative transcoding is presented along objective video quality measurements based SSIM.