DCT Coefficients Modelling for Image Watermarking

Abstract

It is important to select the best well-modelled coefficients of discrete cosine transform (DCT ) to satisfy the requirements of watermarking algorithm. In this paper, we proved that the low ac DCT coefficients are well modelled by using the generalized gamma distribution. A new estimator based on moment method was proposed to estimate their parameters, and the goodness-of-fit (χ2)is used to prove its effectiveness. The modelled coefficients were chosen to perform the watermark embedding process. To increase the security of our scheme, the chaotic maps have been used. Model of the probability edensity functions for the DCT watermarked images is derived, and the detector is developed based on the maximum likelihood. Simulation results proved that the proposed method has better imperceptibility, highly sensitive to the key and robust against JPEG attack when the moment method is used. With our method ewe can still detect the watermark for JPEG compression when the compression ratio is over 28.5 for all attacked watermarked images.