Leveraging the Cuboidal Partitioning for Low Complexity CTU Structure Prediction in Versatile Video Coding

As the capacity of communications systems is currently being outpaced by the increasing demand for access to audio-visual services, the versatile video coding (VVC) standard has emerged. It shows around 40% bit rate savings over the predecessor codec, the high efficiency video coding (HEVC) standard. However, this coding gain is obtained with a significant increase in the computational complexity associated with VVC. The cuboidal partitioning algorithm has shown capabilities in modeling both the global and local commonality of a video frame and discovers a set of non-overlapping rectangular regions, known as cuboids, over the current frame. Since cuboids are highly homogeneous and are computationally efficient, in this paper, it is proposed to employ their intrinsic homogeneity information in predicting the coding tree unit (CTU) structure of VVC. Experimental results show that up to 10.5% complexity reduction for intra coded 4K resolution video sequences, with an increase in bit rate by 1.4%.