A digital watermarking technique based on ICA image features

Abstract

A novel digital watermarking technique based on ICA image features is proposed in this paper. This new watermarking technique is provided for both high-quality visual imperceptibility and robust & effective watermark detection. An adaptive-transform approach is employed in this technique, which is different from the conventional DCT or Wavelet transformations. The learned image-adaptive ICA features with localized, oriented and band-pass characters represent similar properties exhibited by the primary and secondary visual cortexes in human vision system (HVS). It enables a powerful masking effect to hide extra information into images with very little visual changes to human eyes. The embedding and detection of watermarks on ICA coefficients whose distribution is super-Gaussian in nature are found to be effective and robust even when only a classical spread-spectrum method is used. Additionally, the adaptive watermarking on suitable images and image regions is achieved implicitly owing to the merit that the ICA bases are automatically learnt from images. The experiments of the blind image watermarking system demonstrate its advantages on good image quality and robustness under various attacks such as image compression, geometric distortion and noises, in comparison with some conventional methods. © Springer-Verlag 2004.