Optimized Spread Spectrum Watermarking for Fading-Like Collusion Attack with Improved Detection Performance

This paper proposes an optimized spread spectrum image watermarking scheme robust against fading-like collusion attack using genetic algorithms (GA) in multiband (M-band) wavelet decomposition. M-band decomposition of host data offers advantages in better scale-space tiling and good energy compactness. This makes watermarking robust against frequency selective fading-like attack gain. On the other hand, GA would determine a threshold value for the selection of host coefficients (process gain i.e. the length of spreading code) used in watermark casting along with the respective embedding strengths. We then consider colluder (random gain) identification as multiuser detection problem and use successive interference cancelation (SIC) to accurately detect weak (low gain) colluder and innocent users too. A weighted correlator is developed first where weight factors are calculated through training/learning using a neural network. A closed mathematical form of joint probability of error is developed with the objective of minimizing the probability of false detection i.e. a colluder is identified as an innocent user and an innocent user is identified as a colluder, for a set of colluders involved. Simulation results show the relative performance gain of the weighted correlator and SIC scheme compared to the existing works.