Learned video processing for high quality content delivery

In the evolving context of creation, production and distribution of media content, machine learning algorithms are used more often. In particular it has been shown that video enhancements can be much more efficiently performed when deep learning is utilised. This project will address high quality video creation and delivery also taking into account how the data is managed, processed and structured in that process. The main aims are both to deliver algorithms as well as to understand what the "black boxes" of machine learning are doing. This project will look into ensuring the effectiveness and interpretability of data-driven technologies applicable in the context of high quality data distribution scenarios.

Description of Work
Review state-of-the-art techniques for quality enhancement based on deep networks and their interpretability for application on video content.
Establish the limitations of existing approaches for a selection of representative use cases.
Investigate novel quality enhancement strategies for various types of datasets.
Implement reusable components in a framework that can perform various quality enhancement types.
Perform extensive testing to evaluate algorithmic approaches in the context of broadcast applications.
Build a prototype system to demonstrate the developed techniques applied in a broadcast application.
Assess the performance of the developed system and simplify the selected approaches, for more efficient implementations.