Lossless Video Compression Using Reinforcement Learning in UAV Applications

This research examines a technique for lossless pixel-by-pixel image compression that can be learned by using reinforcement learning based image compression (RLIC). It is shown that a small neural network with 59K parameters can learn an auto-regressive dependency and outperforming other recently learned lossless image compression systems in terms of the effectiveness of compression. It was demonstrated that the network could achieve state-of-the-art standards of effectiveness in data compression.