25th January 2018
At last year’s user meeting, NMR introduced its ReCAP project. One year on and Alex Buchanan from NMR gave us an update of the project. ReCAP stands for Realtime Cotent Analysis & Processing. It is a project co-funded by the European Union’s Horizon 2020 programme for research and innovation. The end goal is to commercialise an affordable, scalable and flexible software application for Machine Learning.
The project is being run in partnership with ToolsOnAir, Joanneum Research Institute and Nablet.
Use cases include content enrichment, i.e. automatically adding descriptive and technical metadata to content in production and post-production workflows. It can also be used for compliance, such as ensuring broadcast content is appropriate for the audience, time of day etc. It uses descriptive algorithms, such as face and sentiment recognition, as well as Automatic Speech Recognition (ASR), logo and object recognition, Optical Character Recogniton (OCR) and more.
The final ReCAP product will deliver time-stamped, real-time descriptive analysis of video and audio content, including objects, logo, face detection, and content duplication. It will also provide automatic detection of quality impairments, such as lost/frozen video frames, block dropouts, visible macroblocking, and estimation of sharpness and noise, as well as potentially automatic quality improvement, interpolation of image regions or entire frames, denoising, and block artifact suppression. Additional ReCAP benefits include archive enrichment, such as identifying objects, locations, duplicate or similar content; identifying logos and on-screen text (OCR); and automatic speech-to-text (ASR) to search for content.
When it goes to market, the resulting solution will initially be deployed as an on-premise appliance. However, the aim is for it to be made available as software for deployment on customer infrastructure, whether that is on-premise, in the cloud, or in a hybrid cloud environment.
Alex talked us through the facial recognition. The system doesn’t of course know who the face belongs to but NMR and its partners are spending a great deal of time on the training interface and user experience, ensuring that the system is able to group together faces in a logical way that makes it easy to apply metadata that will be applied to all instances of that same face.
NAB 2018 will see the first public preview of the software and this will be followed by private beta testing with the product launch due in Q3. As with all technology in this space, this will of course be followed by ongoing product development. Third party integrations will also be a key feature, enabling it to combine the best in breed for an all-encompassing solution.
If anyone is attending BVE next month, NMR’s Neil will be going into a great deal of detail around ReCAP on Wednesday 28th, at 11:45 in the Techflow Theatre.