Machine Learning Identifies Thousands of Places And Objects In Video, Automatically Adds Metadata Via Deep Content Analysis
December 21, 2015 – Oulu, Finland – Valossa™ today announced finalization of a $650k seed round with Butterfly Ventures (Oulu-based Micro VC) and several Finnish angel investors, as well as new details concerning its groundbreaking AI- and voice-powered deep content video search technology. The company’s flagship platform is a descriptive search engine (Valossa Search) for video, which enables natural, verbose and flexible querying for voice-controlled movie services and entertainment platforms.
Valossa’s technology is capable of analyzing video streams in real-time to identify more than one thousand concepts (e.g. places and objects) from any video stream. This enables automated scene descriptive metadata creation for content production, real-time video content discovery applications and contextual advertising.
Valossa technology actually understands the contents of video files themselves through a combination of natural language processing and pattern recognition AI. Designed for service and content providers, Valossa can reach down into their video content, identify it and make it searchable. Valossa Search brings natural query understanding to a level never before seen in commercial on-demand movie services.
Example queries understood by Valossa Search include:
- Sean Connery in red pants
- scifi movies about space battles and laser guns
- Eastwood protecting the president
- comedy in Hawaii
- nonviolent princess movies
- romantic comedy movies involving career issues and family
Valossa Smart Search can also easily identify conditional and chained voice commands (a feat not currently possible with Apple® Siri® and other voice-powered search assistants). An example of this would be:
- Query 1: find me epic history movies
- Query 2: only the ones with large battles
New features are also in development – soon, users will be able to find the most recent topical movies using queries such as “latest scifi movies involving time travel”.
Valossa Smart Search is based on the company’s patent-pending, unique deep content models that are constructed from various descriptive movie data sources from all over the Web. These models are then ranked using the company’s proprietary natural query engine.
The engine is capable of finding the best matching results to elaborate searches thanks to years of university research on multimedia information retrieval and AI that is now part of Valossa. The company’s technology demo site at www.whatismymovie.com currently contains data on more than 45,000 English language movies. It demonstrates a level of excellence in search accuracy that exceeds on-demand movie searching by ten-fold performance (results are based on crowdsourced, real user experiments).
Totally cross-platform with Mac® OS, Microsoft® Windows® and Linux®, Valossa’s server-side technology enables virtually identical functionality to the newly announced Apple® TV OS for any system provider, plus more. Valossa also integrates with all leading voice platform vendors, such as Amazon®, Apple, Google® and Microsoft®.
Valossa Automated Metadata for Keywords, Places, Objects and Emotions:
In addition to Valossa Search, Valossa’s technology platform also has the capability for video stream analysis of live TV and other videos. This produces holistic video content descriptions at the scene level as well as descriptive keyword annotation for video overviews.
Valossa uses the most cutting-edge computer vision, machine learning and NLP (Natural Language Processing) to obtain this Deep Content Metadata from several modalities, including analyzing each video frame as well as automatic multilingual keyword extraction from existing metadata.
Valossa’s technology for scene level searching of DVB broadcasts has been successfully in use since 2010 with an online TV search service in Finland. The system summarizes broadcasted TV programs by detecting uniquely descriptive concepts from broadcasted data – a Finnish language technology demo is at www.kuukkelitv.fi/mediaseina.
Lastly, Valossa is finalizing new technologies that will integrate into its platform that include analyzing TV transcripts for emotional sentiment and detecting emotions from human faces.
The IP for all these new algorithms and services has been developed through several different research projects at the University of Oulu with support from the Finnish Funding Agency for Innovation (Tekes) and the Academy of Finland – all IP is owned by Valossa.
Valossa Labs Oy is a new Finnish startup that includes a development team of Ph.D.’s, graduate students and M.Sc.’s from the University of Oulu in Finland, specializing in NLP, AI and video search algorithms. The company has raised a $650k seed round and is seeking business partnerships and additional investors as it launches its technology platform for commercial use. Details at www.valossa.com and @valossainc.