Abstract

The efficiency of motion compensated prediction in modern video codecs highlydepends on the available reference pictures. Occlusions and non-linear motionpose challenges for the motion compensation and often result in high bit ratesfor the prediction error. We propose the generation of artificial referencepictures using deep recurrent neural networks. Conceptually, a referencepicture at the time instance of the currently coded picture is generated frompreviously reconstructed conventional reference pictures. Based on theseartificial reference pictures, we propose a complete coding pipeline based onHEVC. By using the artificial reference pictures for motion compensatedprediction, average BD-rate gains of 1.5% over HEVC are achieved.

 

Quick Read (beta)