Characters Verification via Siamese Convolutional Neural Network

In the printing and carving industries, it is necessary to check whether printed outputs or carved wares are missing or etched through comparing the drawings. Traditional approaches and identification methods can’t be used for this application where the number of character categories are not determined, and where the character may be unique designed by manufacturer. Driven by the one-to-one matching pattern, we propose an end-to-end dual input network for automatic comparison, which uses convolutional neural network to extract features from the scanned images which collected from printed matters. Then, we convert the corresponding drawing to the vector of the same dimension to calculate distance and the match/mismatch result. Experiments show that our method can effectively solve the problem of character comparison with many types, and at the same time propose an automated comparing program for the industrial imprinting process.