Automatic verification system for seal imprints on Japanese bankchecks

We propose an automatic seal imprint verification system composed of a seal imprint extraction stage and a verification stage. It is assumed that each seal imprint is affixed on a bankcheck which has a background pattern and a signature. The extraction stage to separately extract the seal imprint and the signature from a background consists of 3 steps: the transformation of RGB color bankcheck image into HSV color space, the extraction of seal imprint and signature by clustering method in HSV color space, and isolated noise and baseline elimination. The seal imprint verification stage following the extraction stage is based on a method using global and local features. An experiment has been performed to examine the performance of the proposed system. Forty genuine imprints and forty-eight forged imprints on real Japanese bankchecks with various background patterns and signatures were used in the experiment. The experiment achieved as low a false-rejection error rate (Type I error rate) of 7.5% and a false-acceptance error rate (Type II error rate) of 0%.