Automatic fingerprint identification and authentication with specialization in identical twins differentiation (GEMINI)

• Main Authors: Castaneda, Kristin Angela G., Cu, Rachellle Khriseth S., Ruivivar, Carlo G., Takahashi, Satomi T.
• Format: text
• Language: English
• Published: Animo Repository 2009
• Subjects: Identity (Philosophical concept); Identification > Automation; Human beings; Computer vision; Pattern recognition systems"
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/14486
• Institution: De La Salle University

Biometric refers to technologies that measure and analyze human body characteristics in order to identify specific people through certain unique characteristics. Traditional methods of individual recognition are prone to fraud, hence, the creation of a more secure way of verification comes in demand. This automatic identification and authentication method based on fingerprints can prevent instances of deception in this genre--addressing the accuracy issue especially in the case of identical twins who has the maximum similarity between their fingerprints. In this case, image processing techniques are used to achieve an automatic fingerprint identification and authentication system with image enhancement as well as fingerprint reconstruction capability in order to consider every minute fingerprint feature. Binarization-based approaches are employed in order to enhance fingerprint images. Through the Hough transform, these images are to be oriented in a specific angle for uniformity. Since it is inevitable for fingerprint images to be imperfectly acquired, morphological operations are observed for image reconstruction. Careful extraction of features is done by scanning 3x3 windows for terminations and bifurcations--involving post-processing activities afterwards. Matching is minutia-based with consideration of orientation, whose outputs after cross-correlation of vector values, would generate acceptable values for a successful match. Fingerprint enhancement and reconstruction helps boost system performance by improving the similarity values rated at 88.867%. The threshold obtained from training is at 82% and this becomes the basis for matching. Identification's success rate is equal to 78.33%. The system's overall success rate is 89%, which generally shows sufficiency in differentiation.