IBM's Marvel: A Breakthrough in Video Search Technology

IBM researcher John Smith recently demonstrated a computer system called Marvel: MDEG-7 Multimedia Search Engine, which uses statistical techniques to analyze relationships between colors, shapes, patterns, sounds, and other content in video footage. This capability allows users to search for specific frames within large video collections, potentially revolutionizing the way TV news producers, intelligence analysts, and home movie editors work with video content.

Key Takeaways:

• IBM's Marvel system uses statistical techniques to identify relationships between visual and audio content in video footage.
• The system can analyze 70 scenes of rocket launches and rank 50 of them among the top 100 results of a search.
• The system has been trained on over 140 concepts extracted from several years of ABC News and CNN broadcasts, provided by the University of Pennsylvania.
• The Marvel system can identify individual shots within a video by labeling the footage.
• This technology has significant potential for applications in TV news production, intelligence analysis, and home movie editing.

The IBM Research team, led by John Smith, aims to improve the current state of the art in video search technology, which relies on searching for keywords in video files.

Statistics:

• The Marvel system has been trained on over 140 concepts extracted from several years of ABC News and CNN broadcasts.
• The system can analyze 70 scenes of rocket launches and rank 50 of them among the top 100 results of a search.
• 50% of the top 100 search results in the Marvel system correspond to specific rocket launch scenes.

Sources:

• [IBM]
• [John Smith, IBM Research senior manager]
• [University of Pennsylvania]
• [ABC News]
• [CNN]