Simplified geometry-adaptive block partitioning for video coding

Geometry-adaptive block partitioning (GEO) can greatly enhance video coding efficiency but at the expense of significantly increased computational complexity. Instead of proposing fast searching algorithm for encoding only, this paper proposes to reduce the size of partitions for both the encoder and the decoder. The proposed scheme only searches partitions recognized as most valuable partitions, which is derived by analyzing different GEO partitions from the contribution to the coding efficiency. On the one hand, by only searching limited number of partitions, the encoder computation burden is much alleviated and the decoder needs to handle fewer cases. On the other hand, restricting the number of candidate partitions causes fewer overhead bits. Results obtained in the experiments show that compared to the original GEO, the proposed scheme can achieve similar coding efficiency at much lower complexity.