Tamper Detection in Multimodal Biometric Authentication Systems Using Fragile Fingerprint Watermarking and Convolutional Neural Networks

الملخص

The rapid growth of multimodal biometric authentication employment to
protect information and services has attracted significant attention toward
securing the vulnerabilities in these systems. One of the main techniques that
are used to improve the security of these systems is digital watermarking. The
use of digital watermarking allows the biometric authentication to recognize
the authenticity of the images communicated among the different parts of the
system, or stored in a database. Fragile watermarking can also be employed
for tamper detection, in addition to the authenticity recognition task. In this
study, a fragile digital watermarking technique is proposed, which uses the
fingerprint image as a watermark on face images. The proposed method
combines the Discrete Cosine Transform (DCT) and Least Significant Bit
(LSB) watermarking technique. This combination allows the compression of
the watermarked image, as it manipulates the DCT version of the cover
image, while maintaining the fragility of the watermarking, using the LSB
technique. Moreover, the method also encrypts the fingerprint image, using
Arnold Transformation, to add another layer of security to the biometric
authentication system. The evaluation results show that the proposed method
has outperformed the state-of-the-art methods existing in the literature, as it
highly maintains the information in both the fingerprint and the other
biometric images, so that, both can be used in the authentication process
without the need to communicate them separately.