EU Commission promotes the development of new image compression method codec which aims to increase quality and reduce data volume in video streaming
The codec is designed to imitate the neuronal processes of the human eye and thus reduce the file size of images by a factor of ten while maintaining the same quality. This technology is being developed by London-based AI startup Deep Render. German lives
Berlin(20 Jun 2022)
Global consumption of Internet data has grown by more than ten times prior usage. Image and video data comprises more than 85% of all global internet traffic, and this trend will continue due to more and more streaming. The development will be reinforced by new offerings such as 4K, 5K, virtual and augmented reality streaming and cloud gaming. It will not be possible to cover the further demand for Internet capacity with network expansion alone.
Better compression for less energy consumption
A simpler, cheaper and faster solution to expand capacity is more efficient compression of image and video files. This is because better compression reduces file sizes, storage space and bandwidth requirements. It would also save energy and CO2 during storage and also during transmission. However, progress in video compression technologies has stalled since the last decade.
Deep Render has been developing a highly innovative approach to video compression not based on previous compression technologies or codecs. “Built from scratch, we have re-invented the entire domain of compression around modern frameworks, creating a radically new class of compression methods. We combine artificial intelligence, machine learning, statistics, and information theory, in a non-linear approach to video compression that mirrors the neurological processing of the best video compressor known, the human eye — aka, Biological Compression”, says Chri Besenbruch, Deep Render CEO.
The goal of the project is to develop a complete demonstration video codec ready for commercialization for video-on-demand (VoD), with compression efficiency 80% better than the best existing codecs (AV1/H265).
Market-ready Codec as target
“For this project, our objectives are to develop a market-ready codec with a compression efficiency of 80%, decoding times of 16.6 ms/frame (60fps) on an iPhone 13 and a memory footprint of 3MB. The increased compression efficiency means for the same quality of video our equivalent file size will be 20% of the other codec's compressed existing file size, freeing up 80% of previously used internet bandwidth”, says Besenbruch. File size reduction will have a direct impact on energy usage and CO2 production by a similar amount.
This new codec will be tested with Contentflow's streaming software to identify and resolve any integration issues. The Contentflow live streaming platforms covers all aspects of live streaming. Thus, livestreams of events (conferences, discussions, general meetings) can be implemented just as well as complete conferences or trade fairs or media coverage of breaking news. Well-known customers such as ARD Aktuell (tagesschau), the Online Marketing Rockstars or Messe Berlin use the software. "Our goal from the beginning was to offer the highest quality in livestreaming. With the codec, we hope not only to be able to further increase quality, but also to reduce the required data volume for our customers at the same time," says Sebastian Serafin, CEO of Contentflow.
TU Vienna will use its technical expertise in computer vision and machine learning to extend the new codec for four use cases, each with specific challenges. These will be medical imaging, satellite imaging, virtual stereo reality and autonomous cars.
About Deep Render
Deep Render Ltd. was founded in 2017 as a spin-off from the Department of Computing at Imperial College London. The AI startup aims to use its patented technology to develop the next generation of media compression algorithms. More about Deep Render at Tech Crunch. More information: https://deeprender.ai/
Contentflow is a startup that specializes in software for live streaming. With its SaaS solution, Contentflow covers all aspects of live communication. The company's German customers include the Tagesschau, Messe Berlin or Online Marketing Rockstars, while in the USA customers from the entertainment sector and esports rely on the solution. More information: www.contentflow.de
About Computer Vision Lab (CVL)
The Computer Vision Lab (CVL) at TU Wien is devoted to both basic and applied research in the field of Computer Vision or in other words: We are teaching computers to “see”. Computer Vision comprehends the natural world through images, image sequences, and films and reconstructs its properties, such as shape, illumination, and color distributions. The theoretical backbones at CVL are amongst others Image Processing, Feature Extraction and Object Recognition, Document Analysis, 3-D Computer Vision, Multi-Spectral Imaging (MSI), and Machine Learning (Neural Networks, Adaptive Methods, etc.) which the CVL employs in applied research in the areas of Cultural Heritage Applications, Ambient Assisted Living (AAL), Medical Imaging, Industrial Vision, or Image Compression. More information: https://cvl.tuwien.ac.at/