Prediction and Parameter Coding for Non-rectangular Block Partitioning

In the first half of 2019, digital video data was reported to make up 60% of the total downstream volume of traffic on the internet. With new, bandwidth-intensive video applications on the rise, such as immersive video or cloud-based gaming, and established applications moving to increased spatial and temporal resolution, the need to compress the video data with higher efficiency is evident.

A fundamental principle in video coding is the segmentation of every picture into rectangular blocks of pixels. The available methods of prediction and coding of a video coding scheme are then applied to these blocks individually.

This thesis proposes geometric block partitioning as an additional coding tool, to better adapt to the underlying signal characteristics. In this scheme, a rectangular block is partitioned by a straight line into two segments. The pixels associated with each segment are then predicted using motion compensation techniques.

The main contribution of this work is a fully developed, low-complexity geometric block partitioning coding tool (GEO) that provides up to 1% of improved coding efficiency on top of the state-of-the-art, without significantly increasing the decoding runtime. It is particularly useful for coding natural, moving video content in motion with distinct object boundaries (people, animals, or cars), as well as for coding of screen content.

GEO has been presented to the Joint Video Experts Team (JVET) of ISO/IEC JTC1/SC29/WG11 (MPEG) and ITU-T SG 16 (VCEG), the world's leading standardization group for video coding technology, and has been accepted into the working draft of its newest video coding standard in development, Versatile Video Coding (VVC).