Parameter-based Affine Intra Prediction of Screen Content in Versatile Video Coding

Widespread adoption of communication tools, including screen sharing and video conferencing, is creating an increasing need for better compression of artificially generated (screen) content. This type of content is typically characterised by distinct features, including sharp edges and repeating patterns, and as such it benefits from ad-hoc coding tools designed specifically to exploit such features. Among these tools, the emerging Versatile Video Coding (VVC) standard includes Intra Block Copy (IBC), a successful technique whereby reconstructed portions of the current frame are copied as predictions for a given block of samples. However, IBC can only represent translational motion and as such, it cannot predict geometrical transformations common in screen contents. In this paper, a novel tool to improve the efficiency of VVC screen content coding is proposed extending the functionality of IBC. The tool exploits the fact that screen content contains predictable transformations of contents including zooms, rotations and stretches, which may be useful to improve IBC predictions. By appropriately defining affine transformations that are applied on the IBC reference samples, and compacting the information necessary to represent these transformations, higher compression efficiency can be achieved at limited impact in complexity. Extensive experiments showed that the proposed solution outperforms state-of-the-art schemes, achieving consistent coding gains, up to 5.4% for specific screen content sequences.