Biometric watermarks based on face recognition methods for authentication of digital images

Because making digital images secure runs into the substantial challenge of owner authentication, many security schemes based on cryptography, steganography and watermarking technology include biometric recognition methods. To follow on these studies, this paper describes a combination of facial images with watermarking technology to automatically authenticate digital images owners/users. In the proposed methodology, biometric face recognition methods such as principal component analysis and eigenfeature regularization and extraction produce vectorial representations of facial images. These vectors are used as copyright watermarks, in a few common watermarking schemes, and are tested for identification purposes after they are extracted. Initially, watermarking algorithms are studied with some arbitrary cover image, and also the most robust algorithm is tested for different cover images of particular subjects. The strength of this paper is finding relationships between the original and extracted biometric data using neural networks instead of the most common, simple measures such as correlation coefficients or distance metrics. The NN subject identification is performed directly, as there is no need to reconstruct facial images after the watermarks are extracted, compute templates for particular subjects, or seek a suitable distance metric. What is more, the presented study includes a performance comparison of two machine learning methods, frequently used for face recognition, and of a few popular watermarking algorithms. Very promising identification results were obtained in many considered experiments, even those involving attacks on watermarked images. Copyright (c) 2014 John Wiley & Sons, Ltd.