Developed by IBM
Extensive use of human computer interaction

Data Model

Database population

Arbitrary number of image objects are identified by the user using semi-automatic tool.

Feature calculation

Image query

The user navigates and refines the query until satisfied.

Content Similarity Retrieval

Semi-Automatic Object Identification

-more acurate than automatic object identification
-outlines the object in the image
-floodfill the local region with the similar pixel

Automatic Object Identification

-partition based
-region-based "Query-by-painting" for color

Similarity Measures

All of these are user adjustable.  This measure is taken by the weighted sum of the Euclidean distance.

• Color
• Texture
• Shape
• Scene Sketch

Query by example

Direct query:  The user specifies the color/shape/texture/sketch or combination of these.

Feature palettes or pickers

Query by sketch

Query-by-painting

Results and Query Refinement

Ranked set of thumnail images that may be browsed.