Fast semi-blind color image watermarking scheme using DWT and extreme learning machine

Abstract

In this paper, a semi-blind watermarking of digital image based on Extreme Learning Machine (ELM) in DWT domain is proposed. The fourth level (LL4) low frequency sub-band coefficients are used for embedding the watermark. The machine is tuned iteratively and used for training and predicting the sub-band coefficients. The target fourth level sub-band coefficients are augmented by the quantized fourth level sub-band coefficients which are set as an input data-set to train the machine. A random key determines the starting position of the coefficients where the watermark is embedded. A binary watermark is embedded in the blue channel of four colored host images. This watermarking scheme strengthen the robustness towards popular interferences on images. The results of simulation clearly proves that the recovered watermark from signed and attacked images are similar to the embedded watermark. The time spans for training, embedding and extraction are computed and they show real time behavior (millisecond time spans) so that the proposed scheme is suitable for developing real time watermarking applications.