Verificación de Firmas Usando Transformada de Gabor

This paper proposes an off-line signature verification system with fairly good detection capacity against expert forgeries. The feature extraction stage of the proposed system, estimates the coefficients of the Gabor Transform in each local region of the signature image, and extracts the positions of relevant coefficients. These provide local information about frequency and orientation of the signature image texture. Using the extracted features, the proposed system adapts a back-propagation multiplayer neural network with 9-9-2 architecture for each signer. The proposed system was evaluated using 30 genuine signatures and 20 expert forgeries for each signer. The computer simulation results show a 90% overall success, with a lower computational complexity.