Artificial Intelligence based algorithm for image quality enhancement
Streaming Video Content accounted for 63% of Internet Traffic as of 2017 and future traffic volumes forecast significant increase as a result of more internet users, more smart devices along with the demand for higher resolution content from HD to 4K or 8K UHD. The challenge of the project was to deliver high quality video experience to the user by increasing the compression and decreasing the resolution of the source stream. This will have the potential to massively reduce global internet traffic, improving the user experience and benefitting the World at large.
Artomatix goal was to develop a more efficient method to deliver high-resolution content to the user while minimising the network traffic and associated infrastructure costs. So the challenge was to enable data to be transmitted / broadcast at a low resolution and then upscaled and enhanced in real-time on the user’s device by applying “AI at the edge” to deliver that compelling user experience.
By exploiting the capabilities offered by new AI approaches and applying CNNs to image upscaling, Artomatix was able to develope an efficient high-quality hardware independent upscaling engine which can be licensed to and exploited by multiple platforms such as the TheoPlayer™ multi-platform players.
FED4SAE has enabled Artomatix to further a collaboration with Intel; receive support from CEA and receive advice on the management of innovation and the got-to-market strategy from Blumorpho.
Artomatix have built a stripped-down version of its image-enhancement NN making it compatible with resource-limited hardware such as the Intel ® Movidius™ Myriad VPU in the Intel Neural Compute Stick hardware with almost unperceivable quality loss. Thanks to the collaboration with CEA, Artomatix is now operating an industry-standard-based image and video compression testbench for enhanced-images quality assessment and NN training.
Artomatix have investigated the impact of applying Artificial intelligence approaches to enable high quality video upscaling images from 720K to HD 1024K and above. They have identified and validated the best approaches and compared these results to existing upscaling approaches using SSIM etc. They have applied and validated these approaches in a number of domains from Gaming (synthetic images), cartoons to static images used in web content (furniture).
In 2020 Artomatix is now part of the Artomatix is now part of the Unity family (€60m acquisition) who have committed to growing the investment resulting in significant and headcount growth in Europe.