Fraunhofer-GesellschaftNowadays, the world's Internet traffic is predominantly made up of video streaming. In order to support the Internet's limited transmission capacities, videos are compressed during encoding. Reducing video file sizes comes at an expense, however, when disseminating large volumes of streaming video via the Internet, resulting in a loss in quality. Live streaming, in particular, has special requirements with regards to the video transmission process, and can benefit from intelligent approaches in compressing content in a more efficient manner.
Our FAMIUM Deep Encode solution utilizes artificial intelligence methods for automating per-title encoding for Video on Demand and live streaming. During a live playout, we analyze the video and collect existing playback metrics in order to predict the upcoming optimal ladder. With the prediction, encoding settings are then adapted accordingly.
Video analytics is conducted on a per-scene basis in order to adapt settings to the current scene. In comparison to traditional encoding solutions, average storage and transfer volumes are reduced by 30%. This large decrease leads to significant cost savings in the long-run, and improves the overall Quality of Experience for the end user. This FAMIUM Deep Encode solution is codec and format agnostic, and the quality prediction (VMAF) is based on several extracted unique video characteristics.