slide1
Download
Skip this Video
Download Presentation
Exciting Media Limsoon Wong Institute for Infocomm Research

Loading in 2 Seconds...

play fullscreen
1 / 33

Plan - PowerPoint PPT Presentation


  • 209 Views
  • Uploaded on

Exciting Media Limsoon Wong Institute for Infocomm Research . Plan. I will discuss some of the advances on the handling and processing of native media New things that you can do with texts New things that you can do with images New things that you can do with audio

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Plan' - Sharon_Dale


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
slide1

Exciting Media

Limsoon Wong

Institute for Infocomm Research

slide2
Plan
  • I will discuss some of the advances on the handling and processing of native media
    • New things that you can do with texts
    • New things that you can do with images
    • New things that you can do with audio
    • New things that you can do with video
take search engines to the next level
Search engines are getting less useful than before

too many hits

not all relevant

not “organized”

[Hierarchal clustering from BIGontheNet] take Google to a new level – Yahoo! Finance

[p-zoom] may have a leg up on… its competitors – Tech Web News

by

“I like BigOnTheNet\'s and Groxis\'s web search categorization technologies a lot” – Chris Shipley on WashingtonPost.com

Take search engines to the next level
intelligent information extraction improve safety
Extract chemical safety information from Materials Safety Data Sheets (MSDS)

Check for conformance to standards

Benefits

OHD: Check 100% of MSDS (currently < 10%) with same manpower

Chemical Suppliers: Savings in distribution of MSDS as it is online

End users of chemicals : Better quality MSDS, improved safety

MSDS (in variety of formats)

searches

End-users

(English-

speaking)

Chemical

suppliers

MSDS

(verified)

MSDS

Knowledge Workbench

Knowledge

encoding

OHD

Intelligent information extraction,improve safety
make computers easier to use
Abstraction of image content allows interpretation and matching in semantic space

Visual query language allows specification of what and where

Make computers easier to use

Search photos by

visual keywords

how is it done what s needed to get it done9
Trained visual keywords for semantic detection and summarisation

Automatic indexing using such keywords

Faces :

Crowd :

Buildings :

Foliage:

How is it done? What’s needed to get it done?
let machines perceive as we do

Higher PSNR!

New perceptual metric perceives correctly.

Comparison with MOS (Mean Opinion Score)

Pearson Correlation

Spearman Correlation

PSNR metric

0.66

0.69

New metric

0.83

0.81

Let machines perceive as we do
  • Perceptual visual quality according to characteristics of human vision
  • Outperform metrics in ITU-T VQEG test
  • Adoption in video coding results in efficiency & quality improvement (other systems make compromise betw. the two)

Better consistency

Better accuracy

let machines listen as we do

Automatic speech recognition

Business Logic

PSTN

Text-to-speech

Speech enhancement,

Noise reduction

Multilingual

voice mining,

Speech & dialogue

processing

Text categorization,

Natural language

processing

Semantic

outputs

voice

Let machines listen as we do
to build a mobile audio industry analogous to the graphics industry
To build a mobile audio industry… analogous to the graphics industry
  • Synthesis-directed analysis of sounds
    • how would you model a lion’s roar?
  • Algorithms for synthetic sound generation
  • Tools for sound model automation and support
  • Cross-platform audio synthesis engine with a small footprint and low compute requirements
make health monitoring less hazardous
Mobile and non-invasive health-monitoring devices for home use

Potentially large market for densely populated areas with low medical facilities & personnel

Aging population in developed countries means more need for automatic continuous monitoring to lower overall medical costs

Passive health monitoring devices less of health hazard and allow long-term usage

Sound-based automatic detection & classification of medical anomalies, e.g., Long term fetal heart sound monitoring

Make health monitoring less hazardous
make home videos more fun
Make home videos more fun
  • Select video frames
  • Cut to music
  • Decide on transitions

Turn home video into high

quality MTV automatically

prevent drowning save lives
Drowning Early Warning System

tracks people in dynamic aquatic conditions

intelligently detect water crises situations

Prevent drowning, save lives
improve sophistication of our media industry

Tagged to camera h/w

  • costly
  • done once at source
  • “Intrusive” ads
  • pops out during play!
  • Enhanced tennis TV
  • auto-tracking
  • super-resolution
  • Software detection:
  • performed any-time
  • demographic Ads
  • cheap
  • Non-intrusive insertions
  • detects non-play segment
  • non-interfering insertion
  • Tennis TV
  • manual
Improve sophistication of our media industry

Existing

pains

Non-intrusive virtual contents insertion

how is it done what s needed to get it done22
How is it done? What’s needed to get it done?
  • Robust Scene Modeling and Camera calibration
    • Given a 2D court model of 3D scene that camera is capturing, identify 3D object positions robustly and accurately
  • Super-Resolution Image Reconstruction from Video
    • Given a low resolution image sequence of an object far away from the camera, reconstruct a larger resolution image sequence
    • This is essentially an ill-posed problem, but we can apply domain info such as motion, pose, etc, to seek a good solution
  • Robust Object and Landmark Detection
    • Real-time
    • Geometric invariance
  • Deployment and Application Constrains
    • Real-time
looking even further

Looking even further...

Media in 2010

according to

Institute for the Future

dimensions of entertainment activities
Dimensions of entertainment activities
  • The event
  • The process
  • Popularization of research
  • Practice & performance level
  • Entertainment spaces
  • Entertainment tools
  • Sharing & social communication
  • Consuming entertainment
  • Creating entertainment
  • The work of entertainment
key shifts shaping new entertainment
Key shifts shaping new entertainment
  • Mass to Personal
    • consumers will appropriate mass media tools for their own personal expression
  • Packaged to Self-generated
    • consumers will create the entertainment experiences they engage in
  • Episodic to Persistent
    • entertainment experience will be ongoing & will have no clear starting and stopping point
  • Virtual to Embodied
    • information, images, and experiences will be embedded into physical objects & physical world
mass to personal
Eg: Blogging---how the Web evolves from a publishing medium to a personal creativity tool

Blogs are written online, using tools accessed thru a Web browser, combining simplicity and immediacy of instant messaging, & broad accessibility of Web sites

Most personal blogs are essentially online diaries, sometimes devoted to specific subjects. They are short, updated frequently, and full of hyperlinks, & crafted to look like snapshots of what their authors are thinking and creating

Mass to personal
packaged to self generated
Eg: Fantasy sports leagues

In these leagues, individuals act as managers of professional sports teams. They pick players in live drafts & develop rosters for playing their team against other fantasy teams in their league

The real experience of Fantasy Leagues is the interaction among the various player managers—the player trading, email, & competition

The digital world becomes a vibrant place for all sorts of interactions like trading, creation or production, & spectating

Packaged to self-generated
episodic to persistent
Eg: Massive multiplayer role-playing games

Sony’s EverQuest has over 400,000 players, 70,000 of which are online at any given time, slaying dragons, traveling to cities, or trading with each other. Individual players acquire property, talents, and social identities over time

Players’ actions permanently affect the virtual world: a player’s property will exist even after they’re gone

Runs 24/7 & when players log off, game continues to evolve and grow, driven by social behavior of thousands of players that log in each day

Episodic to persistent
virtual to embodied
Eg: Geocaching adventure game played worldwide

Founder creates a cache which holds a treasure, & posts real world location coordinates on a Web site

Seekers retrieve coordinates to cache site on the Web & use a GPS device to get within 20 feet of it

Then they use clues, hunting skills, special real-world skills (hiking, scuba diving, etc) to find cache

Finally, they sign logbook belonging to cache, recover treasure, & leave a treasure for next seeker

Virtual to embodied
in summary
In summary
  • The new “entertainment” media in 2012 will be oriented around personal media that
    • are generated by consumers rather than packaged and distributed by providers
    • provide persistent experiences that do not disappear with a switch of a button but linger over the course of daily life
    • are touch points in the physical environment that embody entertainment and set forth a new relationship among consumers, entertainment, and their broader daily life activities
implications
Implications
  • Focus on leveraging distinctive points of view
  • Co-create continuous experiences with consumers
  • Think of ways to target players who personalize mass events
  • Develop tools and processes for fusing physical & virtual campaigns
  • Focus on promoting user customization
  • Design for participants to “sell to their friends”
  • Develop products for persistent experiences
  • Include digital experience in physical products
slide33
title: Exciting Media
  • abstract: I will discuss some of the advances on the handling
  • and processing of native media. In particular, I
  • will look at
  • - new things that you can do with video
  • - new things that you can do with audio
  • - new things that you can do with images
  • evening of 6th april
ad