Towards a Model Based Paradigm for Efficient Coding of Context Dependent Video Material

This paper proposes a generalised framework for model based, context dependent video coding, based on exploitation of characteristics of the human visual system. The system utilises variable quality coding, based on a map which is created using context dependent rules. The technique is demonstrated for a specific video context, namely open signed content. Consequently a model for gaze prediction in open signed content is developed, based upon motion and shot changes. The framework is shown to achieve a considerable improvement in coding efficiency for the given context.