Contact Point:
Prof. Alberto Del Bimbo ( )
University of Florence
  

Technical Contact Points:
Ing. Stefano Berretti ( ) and Prof. Pietro Pala ( )
University of Florence
  

Type of Software:
DEMO
  

Descriptive Keywords:
Content based image retrieval; color based image retrieval; shape based image retrieval.
  

Potential Use and Applications:
Images are being created and employed in a wide range of application domains. The development of techniques to enable retrieval by content of images assumes an ever increasing relevance in all those area where it is necessary to find similar images in large databases.
  

General Description:
At the Media Integration and Communication Center of the University of Florence, a prototype system for content based image retrieval has been developed. The basic idea of content based image retrieval, is to allow users to find images in a large repository by providing to the system an image as example of the the main characteristics the user is looking for. The query image is described according to a set of representative features which capture salient color and shape information of the image. The goal for the retrieval system is to find images similar to the query image in terms of the extracted features.
    

Technical Description:
The content based image retrieval demo comprises two separate modules: feature extraction; query.
  
The feature extraction module is used to process the images to be archived. The processing can automatically extract an image descriptor based on color information. Instead, shapes of relevant objects in the image must be manually outlined by the user. Shape representation are then automatically extracted by the system considering only the shape contour outlined by the user.
  
The query module is based on a client/server architecture. On the client side a Java applet is used as query interface. This permits the user to draw a query image using a query by sketch paradigm. Alternatively, the user can select one of the images in the database as a query. On the server side, the remote server communicates with the applet by using a Java socket connection. Search engines for color and shape similarity are then called on the server machine as separate applications. These search engines are written in C/C++. Though the original version of the demo system was written for a UNIX operating system, a porting on the Windows OS has been also successfully tested.
  

Required User Skills: 
No particular skills are required to use the demo. Just a normal level of practice in interacting with software applications is needed. A short learning phase is necessary to use the search modality which uses the query by sketch paradigm.
  

Pre-Requisites for Installation:
The content based image retrieval demo has been tested on three different platforms: Windows, Linux and Unix IRIX. On the server side, the installation of a Java runtime environment is needed. On the client side, the query interface runs as an applet in a web browser.
  

Conditions of Use:
Contact the authors.