The Real Problem of Bridging the Multimedia “Semantic Gap”

Loading...

Flash Player 9 (or above) is needed to view presentations.
We have detected that you do not have it on your computer. To install it, go here.

0 comments

Post a comment

    Post a comment
    Embed Video
    Edit your comment Cancel

    1 Favorite

    The Real Problem of Bridging the Multimedia “Semantic Gap” - Presentation Transcript

    1. The Real Problem of Bridging the Multimedia “Semantic Gap” WWW 2007 John R. Smith Senior Manager, Intelligent Information Management IBM T. J. Watson Research Center Contact: jrsmith@watson.ibm.com
    2. Multimedia and the Web
      • Multimedia on the Web has arrived in a big way:
        • Video is estimated to be 60% of web traffic by volume
        • 70,000TB (or 101M hours) original TV & radio content produced annually; much is finding its way to Web
        • 130B video streams will be served by 2010
      • Standards can have a role, but what exactly is the problem?
        • Playback – not today
        • Search?
        • Filtering?
        • Piracy prevention?
        • Advertising?
      Let’s address search
    3. Multimedia Standards and the Web No shortage of standards, metadata schemas, languages Too little understood about user needs and meanings related to the content Digital Content Chaos Metadata Chaos Semantic Chaos Formats Meanings
    4. Need for Machine Tagging … Professional Cataloging and Social Tagging Not Enough!! Manual Cataloging – By Professionals Social Tagging – By Users Automated Tagging – By Machine Popularity Digital item Personal content Web, Deep archives, raw footage High-value content, hit-TV shows, movies “ Long tail” “ Bridging Semantic Gap”
      • Costly
      • Human resource intensive
      • Cannot keep up
      • Controlled vocabularies & standard taxonomies
      • Higher quality
      • Example: Fox, CNN, BBC, Broadcast TV
      Cons Pros
      • Ambiguity
      • Uncontrolled vocabulary
      • Synonyms
      • User driven
      • Emergent folksonomies
      • Serpendipitous browsing
      • Examples: Del.icio.us and Flickr
      Cons Pros
      • Requires training of models
      • Lower quality than manual tagging
      • Lower human cost
      • Domain & data driven approach to semantics
      • Example: Marvel, Informedia, TRECVID concept detection
      Cons Pros
    5. The Challenge
      • The “semantic gap” is a well-know problem in multimedia research (#1 on top-10 list)
      • Or so we thought …
      • The problem has been posed as classifying and searching multimedia content from low-level audio-visual features
      • But, most work to-date has been on the wrong side of the problem
    6. Current State
      • Yes, machine learning technologies are an important tool
      • But, not being applied effectively to improve video search
      • Problem is lack of understanding of required semantics
        • Indoors/Outdoors ? Sunsets ? Cityscapes ?
      • Today’s approaches are ad hoc:
        • Researchers don’t have domain insights
        • Content/search providers don’t know full capability of today’s multimedia analysis technologies
        • Nobody has training data at required scale to effectively bridge semantic gap across full breadth and depth of multimedia semantics on Web
    7. So, What do we do? Leverage Digital Masses to Create the Key Missing Resource – Media Net
      • Since video search is visual, the semantic spaces should be defined visually as well
      • Create large multimedia knowledge-base with exemplar content representing all semantic concepts relevant for search
      • Allow semantics space to evolve from end-user perspectives (across sports, entertainment, news)
      • Allow technology to focus on extracting the relevant semantics – truly providing the needed data-driven approach for bridging the multimedia semantic gap
      … … … Sports Court Scene Players Game Scene Basketball News Crowd Scene

    + jrs21jrs21, 3 years ago

    custom

    2472 views, 1 favs, 0 embeds more stats

    WWW-2007 Panel Position:
    - Since video search is v more

    More info about this document

    © All Rights Reserved

    Go to text version

    • Total Views 2472
      • 2472 on SlideShare
      • 0 from embeds
    • Comments 0
    • Favorites 1
    • Downloads 150
    Most viewed embeds

    more

    All embeds

    less

    Flagged as inappropriate Flag as inappropriate
    Flag as inappropriate

    Select your reason for flagging this presentation as inappropriate. If needed, use the feedback form to let us know more details.

    Cancel
    File a copyright complaint
    Having problems? Go to our helpdesk?

    Categories