A method is described to extract an index from a binary character imag
e for database storage and retrieval. The index is hierarchical in nat
ure, and different parts can retrieve character image clusters of diff
erent sizes. A hierarchy of Kohonen neural networks is used to generat
e the clusters on which the indexing method is based. The application
is the identification of the font from a sample of printed characters.
This is particularly important for forensic document examiners who ma
y have to match a document of unknown origin with the equipment that m
ay have produced it. A computerised system is expected to considerably
reduce the time required for each document search.