In this paper, a system of two separate phases for
signature recognition and veri"cation is developed. At
"rst the signature database is described, then preprocessing
steps and feature extraction for the recognition process
are discussed. A suitable combination of global and
local features is used to produce more distinctive and
e!ective features by combining the advantages of both.
A recognition technique is developed based on a multistage
classi"er. In which a preclassi"cation stage for a
group of similar slant signatures is applied in the "rst
stage. Then, a recognition scheme is applied to resolve
individual identi"cation within a group. In the second
stage, the distances between the global feature vector of
the input sample and the mean of each class in the group
are computed and compared sequentially, in order to
select the best three candidates. Finally, in the third stage,
the local center points are used to choose the best candidate
or to decide that the signature cannot be recognized.
This is decision based on the corresponding threshold of
each candidate class.