Robust Image Watermarking Based on ICA-DCT and Noise Augmentation Technique

In this paper, We present a blind image watermarking method based on DCT domain and independent component analysis (ICA). First of all, an approximation of cover image is created using ICA. We divide approximation image to equal 8x8 blocks and then apply DCT transform to each block. We use spread spectrum (SS) method for embedding the watermark bits. In our method a group of random bits are generated and embedded into high frequency coefficients of each block using SS method. For finding the watermark bits, we denoise watermarked image in order to augment the noise that was embedded into cover image. Then similar to embedding process the approximation image is created from the augmented noise image using ICA, and two group of random bits are generated again, and correlation of high frequency coefficients of the watermarked approximation image with the groups of random bits is computed. We find that, higher robustness and better imperceptibility is obtained because of using approximation image and spread spectrum technique. We test our method against some attacks such as additive Gaussian noise, Salt and Pepper noise, Speckle noise, JPEG and JPEG2000 compression and Average filter. Results show that, our method have a high robustness against Gaussian noise and compression, and good imperceptibility measure like PSNR for the watermarked images. Sensitivity to geometric attacks and need to more times for calculating ICA are two limitations of this method.