Deep Learning Approach to Video Compression

Abstract

Video streaming requirement has increased exponentially and video currently consumes 75% of the internet traffic. Due to which video streaming and storage is a huge challenge for service providers. Image and video compression algorithms rely on codecs which are encoders and decoders that lack adaptability. Due to the advent and advances in Deep Learning these issues can be solved. This paper proposes a method for video compression using neural networks that outperforms the H.264/AVC video coding standard as measured using Multi-Scale - Structural Similarity Index (MS-SSIM).The neural network model proposed is a multi-layer architecture consisting of two parts i) Encoder and ii) Decoder. The training of the two parts of the model happens together and during test time the encoder and decoder are separated to be used as just like any another compression encoding and decoding modules. The entire model’s purpose was to try and capitalize on the temporal and spatial dependencies between frames of a video.