Fast CU Splitting And Computational Complexity Minimization Technique By Using Machine Learning Algorithm In Video Compression

Abstract

Video compression is frequently used for reducing the original data in order to store or communicate the data through a transmission line. All modern video codecs use the flexible block partitioning at the time of rate-distortion (RDO) process. As a result, these codecs minimize the bits but has a immense time complexity. So, computational time complexity is one the main problems for video encoders and decoders. In this work, we proposed a machine learning approach called gradient descent algorithm for fast CU splitting to reduce the time complexity of the video encoder by using the concept of skip criterion. Experimental results show that on average 44.18% encoding time can be saved over HEVC (High Efficiency Video Coding). On the other hand, on average, 1.34% bit rate is increased as a trade-off.