The 3-stages clustering algorithm for visual patterns to find structural 
coefficients describing image properties is presented. 
Structural coefficients and statistical characteristics of visual fragments 
are proposed for pattern classification and image searching in the databases.
An approach based on three stages clustering algorithm.  For  structuring 
coefficients were used the following definitions: MC - a microcluster number 
(input clusters for 1st stage); C — clusters by a shape of rectangle 
(output clusters from 1st stage and input clusters for 2nd stage);  
CR –  regions (output clusters from 2nd stage).    
Eight features to store and retrieve images were taken as pattern keys.
There has been developed an experimental program package   with the user 
interface  which controls all operation stages: image input, monitoring
 of control parameters, output of clustering options (full, specified), 
restored image and full report of results.