Coding Unit Partitioning Using Refined Stacked – Convolutional Neural Network with Scaled Exponential Linear Unit

High-Efficiency Video Coding (HEVC) is a significantly used video coding standard with primary encoding challenges in the searching process.Video encoding is a challenging task due to high complexity and time consumption in Coding Tree Unit (CTU) partitioning.The Refined Stacked -Convolutional Neural Network with Scaled Exponential Linear Unit (RS-CNN with SELU) method is developed to minimize complexity and time consumption of encoding process.Initially, the blocks are partitioned by using Quad Tree (QT) based portioning method.The partition of the Coding Unit (CU) is performed by using the RS-CNN with SELU method, which minimizes the overall complexity and quickens the process of encoding.By using the refined stacking layers in the CNN method, it learns complete feature representation which is essential for correct partitioning of CU.The SELU activation function is carried out in the training process of RS-CNN that maintains stable activations and leads to quick convergence for accurate CU partition.The RS-CNN with SELU method attains Bjontegaard Delta Bit Rate (BD-BR) of 2.569% on JCT-VC dataset and 3.320% of BD-BR on UVG dataset which shows effective performance than Densely Connected Convolutional Neural Network (D-CNN).