Fast Block Size Decision for HEVC Encoders with On-the-Fly Trained Classifiers

High Efficiency Video Coding (HEVC) introduced flexible block partitioning structures that increased significantly compression rates in comparison to previous standards. However, such features resulted in a non-negligible increase in computational cost as well. To accelerate this complex partitioning process, this paper proposes a method that halts the usual rate-distortion optimization employed in Coding Unit size decision by a set of decision tree classifiers, which are trained on the fly according to the current video sequence characteristics. The classifiers are built during the encoding process by the C5 machine learning algorithm, which was chosen based on an extensive analysis that compared several algorithms in terms of decision accuracy and training complexity. Experimental results show that the strategy is capable of building accurate models and decreases the HEVC encoding time in 34.4% on average, at the cost of a compression efficiency loss of only 0.2%.