Optimizing Video Compression & Decompression Using Deep Learning Networks

Over recent years, the necessity for video streaming has grown significantly. People have incorporated the habit of watching videos and sharing them over electronic devices very often into their daily lives. This activity has made the process of storing a video and transferring it much more challenging for service providers, and in turn, increased the need for rugged and robust algorithms that can compress videos. One of the latest video compression standards, High-Efficiency Video Coding (HEVC), also known as H.265, supports resolutions up to 8192Å∼4320Å, including 8K UHD. While maintaining a high-quality image, HEVC reduces the storage reduction by 50% as it encodes video at the lowest possible bit rate. In this paper, Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN) aid in compressing the video frames, which are already known to give a better compression than JPEG. Further, the video frames are enhanced using Image Processing Filters (Guided, Bilateral and Gaussian filters). The standard Moving Picture Experts Group (MPEG) algorithm is applied to these video frames in the final stage. A significant data set of diverse videos are given as input to the training data to minimize error and distortion on a single video frame.