Watermarking Stack of Grayscale Face Images as Dynamic Multi-Objective Optimization Problem

Face images have been used in many access control applications to recognize individuals. Protecting face biometric templates against unauthorized digitization and manipulations is the main challenge in such applications. Intelligent watermarking addresses these challenges using optimization techniques to find the optimal embedding locations to maximize both watermark quality and robustness. Watermark quality measures the distortion resulting from watermark embedding, and robustness represents the resistance to different manipulations on watermarked image. The computational complexity of optimizing embedding for large stack of high resolution grayscale face images is high. In this paper, we propose to handle the overall optimization for stack of face images as one dynamic multi-objective problem rather than series of static problems. It is proposed to use multi-objective optimization to optimize simultaneously watermark quality and robustness in dynamic environment using incremental learning techniques. Experimental results show significant computational complexity reduction to find near optimal solutions for watermark optimization.