JVET-AJ0098 Non-EE2: OBIC with improved sampling strategy and scaling weights [Y. Liu, Y. Huo (Transsion), G. Li, J. Chen, L. Luo, H. Guo, C. Zhu (UESTC)]

This contribution proposes an improved sampling strategy for the occurrence-based intra coding (OBIC) mode. In OBIC mode, the intra prediction mode of a block is determined according to the occurrence of the intra modes in the block’s spatial neighborhood. The proposed method utilizes a sampling strategy to expand the number of adjacent spatial neighborhood blocks and non-adjacent spatial neighborhood blocks. Furthermore, the area-related amplitudes are scaled with different weights before the cumulative sum of occurrence values in the Histogram of Occurrence.

The performance of the proposed method over ECM-14.0 is reported as follows {for Y, U, V, EncT, DecT, EncVmPeak, DecVmPeak}:

AI { -0.02%, 0.03%, 0.04%, ET 102.8%, DT 101.8%, 99.6%, 100.0%}

RA { xxx%, xxx%, xxx%, xxx%, xxx%, xxx%, xxx%}

Two aspects: Sampling and scaling.

Encoder and decoder run time are claimed to be unreliable, but there is probably some additional processing performed at encoder

Gain is not sufficient to be included in EE – further study to improve it is needed.