Based on region mapping paradigm.   Blobs are relatively homogeneous regions of color and texture.  User sees the system's blob representations.  Claims of 28-62% precision on first 50 hits.

Home Page:

http://http.cs.berkeley.edu/~carson/blobworld/

Chad Carlson:

http://http.cs.berkeley.edu/~carson/

Can recognize certain objects based on pre-defined body plans.  A body plan is a collection of grouping hints that are based on geometric, color, and texture constraints for a specific object, e.g. a  horse.   Claims 10% recall but only 0.4% false hits.

Body Plans Overview:

http://http.cs.berkeley.edu/~daf/people.html

David Forsyth:

http://http.cs.berkeley.edu/~daf/

Margaret M. Fleck:

http://www.cs.hmc.edu/~fleck/

Designed to handle realistically large databases.  Offers colorand texture analysis with integrated textual searching in a single graphical user interface.  Allows user-defined concepts.  Claims 93% precision and 64% recall for searches that combine color with text.

Chabot:

http://www.cs.berkeley.edu/~ginger/aboutChabot.html

Virginia Ogle:

http://www.cs.berkeley.edu/~ginger/

This is a new version of Chabot.

Home Page:

http://elib.cs.berkeley.edu/cypress/

This is a list of other image matching projects and web sites.

List of Links:

http://dbweb.csie.ncu.edu.tw/~ljr/idbm/idbm_links.html

Breakthrough system that performs a form of object recognition that is used to construct image scene descriptions.  A thesaurus module is used to generate text tags from low-level primitives.  The system also interprets whole/part relationships.  Claim 65% precision for a 1200 image database.

Thornsten Hermes:

http://www.informatik.uni-bremen.de/~hermes/home.html

Features an advanced attribute-based model.   Users start with text search, then browse images according to appearance and shape characteristics (things) or texture mapping (stuff).  Iconic (similarity) matching captures faces. Claim high system performance with 200 images.

Home Page:

http://vismod.www.media.mit.edu/vismod/demos/photobook/

Alex Pentland:

http://pentland.www.media.mit.edu/people/pentland/

Online image library that offers royalty free digital stock photography.

Home Page:

http://www.photodisc.com/

Identifies and names people in images.   Captions that are associated with news photographs are used as collateral information.  Spatial and other relationships are extracted from these captions.   These help the system determine the location of individuals in a particular image.   Claim 65% success with a database of 50 images.

Home Page:

http://www.cedar.buffalo.edu/Piction/

This system is available for download in a 90 day trial.  QBIC is designed for large image databases.  The system accepts both still images and video as input.  The system also supports search by color, texture, user sketches, iconic matching, and basic text strings (keywords).  Although QBIC returns ranked results, recall is low.  At this time, QBIC is the only commercial content-based image retrieval system.

Home Page:

http://wwwqbic.almaden.ibm.com/

Virage demo relies on visual scanning by the user.  System supports similarity search. User starts with image thumbnails.  Search parameters can be changed to emphasize color, texture, and composition.  Lacks system performance evaluation.

Home Page:

http://www.virage.com/index.html

Visual Information Retrieval:

http://www.virage.com/market/vir.html

Visual Information Retrieval Demo:

http://www.virage.com/market/virdemo.html

Media Manager and Browser:

http://www.virage.com/market/mediamanager.html

Description

Home Page:

http://disney.ctr.columbia.edu:8021/VisualSEEk/

Description

Home Page:

http://www.dreamlabs.com/~webland/c_p/ip97/ip97s301.htm