Watermarking schemes to secure the face database and test images in a biometric system

Abstract

This paper attempts to solve the integrity issues of a compromised face biometric system using two watermarking schemes. Two new blind watermarking schemes, namely S1 and S2, are proposed to ensure the integrity of the training face database and of the test images, respectively. Scheme S1 is fragile spatial-domain based and scheme S2 works in the discrete cosine transformation (DCT) domain and is robust to channel noise. The novelty of S1 lies in the fact that it is lossless and the ratio of watermark bits to the size of the host image is 2.67, while S2 has better robustness than existing blind watermarking schemes. The performance of both schemes is evaluated on a subset of the Indian face database and the results show that both schemes verify the integrity with very high accuracy without affecting the performance of the biometric system.