Video encoding according to the latest standard HEVC and google encoder VP9 require a large number of operations per encoded frame, with a large increase when compared to the previous standard, H.264/AVC. This trend of each new generation of video encoder demanding more computational resources than the previous generation is expected to apply to the successor of HEVC, as inferred from previous published works by the proponent which describe several enhancements to HEVC that most likely will be part of the next generations of encoders. These problems are not exclusive to video encoding and also apply to video transcoding, for example from H.264/AVC to HEVC. Recent studies have shown that it is possible to reduce drastically the video encoding and transcoding complexity through the incorporation in the encoder/transcoder of fast mode decision algorithms designed using offline Machine Learning techniques. Unfortunately the same studies also showed that the methods described are sensitive to changes in video data characteristics. Clearly some kind of online adaptation/learning has to be added to the mode decision algorithms to make them more robust to these changes, much in the way of what has been proposed in . The goal of this project is to investigate the use of Machine Learning techniques with online adaptation capabilities in the control of video encoders and transcoders, aiming at building coding mode predictive models to speed up the overall encoding/transcoding operations, with the ability to adapt to changes in the characteristics of the video signals being encoded or transcoded. The first approach to be followed will involve modifying previous machine learning aided encoder/transcoder mode decision methods proposed by the project leader and its collaborators to provide them with continuous learning abilities. Time allowing, other solutions will be sought, with the same aim of reducing the number of computations needed to encode or transcode digital video, in the context of the HEVC encoding standard.
|Start Date: 01-04-2017|
|End Date: 01-03-2018|
|Team: Luis Alberto da Silva Cruz, Sérgio Bampi|
|Groups: Multimedia Signal Processing – Co|
|Local Coordinator: Luis Alberto da Silva Cruz|