Efficient coding unit classifier for HEVC screen content coding based on machine learning

The Video Coding Joint Collaboration team (JCT-VC) has been working on an emerging standard for screen content coding (SCC) as an extension of high efficiency video coding (HEVC) standard known as HEVC-SCC. The two powerful coding mechanisms used in HEVC-SCC are intra block copy (IBC) and palette coding (PLT). These techniques achieve the best coding efficiency at the expense of extremely high computational complexity. Therefore, we propose a new technique to minimize computational complexity by skipping undesired modes and retaining coding efficiency. A fast intra mode decision approach is suggested based on efficient CU classification. Our proposed solution depends on categorizing a CU as a natural content block (NCB) or a screen content block (SCB). Two classifiers are used for the classification process. The first one is a neural network (NN) classifier, and the other is an AdaBoost classifier, which depends on a boosted decision stump algorithm. The two classifiers predict the CU type individually and the final decision for CU classification depends on both of them. The experimental results reveal that the suggested technique significantly decreases encoding time without sacrificing coding efficiency. The suggested framework can achieve a 26.13% encoding time reduction on average with just a 0.81% increase in Bjontegaard Delta bit-rate (BD-Rate). Furthermore, the suggested framework saves encoding time by 51.5% on average for a set of NC sequences recommended for standard HEVC tests with minimal performance degradation. The proposed strategy has been merged with an existing methodology to accelerate the process even further.