Image Database Retrieval
NTT: Visit
1/7/99

Overview of IDB Meeting
Motivation from MIT ...
Discuss current and related work
Flexible Templates
Complex Features
Demonstrations
Related NTT Efforts
Discussion of collaboration
Future work
Dinner

Motivating Scenario
Image Databases are proliferating
The Web
Commercial Image Databases
Video Databases
Catalog Databases
“Find me a bag that looks like a Gucci.”
Virtual Museums
“Find me impressionist portraits.”
Travel Information
“Find me towns with Gothic architecture.”
Real-estate
“Find me a home that is sunny and open.”

There is a very wide variety of images...

Search for images containing waterfalls?

Search for
   cars?

What makes IDB hard?
Finding the right features
Insensitive to movement of components
Sensitive to critical properties
Focussing attention
Not everything matters
Generalization based on class
Given two images
Small black dog & Large white dog
(Don’t have much in common…)
Return other dogs

Overview of IDB Meeting
Motivation from MIT ...
Discuss current and related work
Flexible Templates
Complex Features
Demonstrations
Related NTT Efforts
Discussion of collaboration
Future work
Dinner

Complex Feature Representation
Motivated by the Human brain…
Infero-temporal cortex computes many thousand selective features
Features are selective yet insensitive to unimportant variations
Every object/image has some but not all of these features
Retrieval involves matching the most salient features

Slide 10

Slide 11

Resolution is reduced at each step…

Not every feature is useful for a query

Normalization of Signature Space

Distance/Similarity Measure

Incorporating Negative Examples
Diagonal approx. of Fisher’s Linear Discr.
Weight features highly if:
Variance of pos and neg is greater than
Variance of pos alone

Interesting Pattern Recognition Properties
The statistics of the data is non-gaussian
Data is 45,000 dim. but highly redundant.
PCA can be used to reduce dimension
But,  retrieval performance deteriorates (??)
** Non-gaussian data!
Retain only those features which are kurtotic
45,000 down to 5,000
** Performance improves!
Kurtotic features are those which are unusual
Distinct, interesting.
Kurtotic features require fewer bits

Slide 18