A method for the automatic verification of handwritten signatures
is described. The method relies on global and local features that
summarize aspects of signature shape and dynamics of signature production.
We demonstrate that with the addition of our local feature based
on stroke direction coding, the performance of signature verification improves
significantly. The improvement is comparable with that of more
sophisticated algorithms, which require much more computer resources.
The current version of the program needs about 150 bytes to store a
signature model and has 3% equal error rate. At the 1% false rejection
point, the addition of the stroke direction information to the algorithm with
only global features reduces the false acceptance rate from 13 to 7.5%.
Prior to extraction of stroke direction information, signatures are normalized
for position, size, and orientation using their Fourier transform. Such
a scheme can also be useful in signature smoothing and compression.