Self-adjustive DE and KELM-based image watermarking in DCT domain using fuzzy entropy

With advances in machine learning and development of neural networks that are efficient and accurate, this paper explores the use of kernel extreme learning machine (KELM) to develop a semi-blind watermarking technique for grey-scale images in discrete cosine transform domain. Fuzzy entropy is employed for selection of the blocks where the watermark bits are to be embedded. A dataset formed from these blocks is used to train KELM. The nonlinear regression property of KELM predicts the values where watermark bits are embedded. Self-adjustive differential evolution (SeAdDE) controls the strength of the scaling factors finds their optimal values. The adaptiveness of differential evolution (DE) helps in self-adjustment and varies the DE parameters to explore best solutions. This saves time as the manual hit and trial method for finding the appropriate parameter values is avoided. The scheme presented shows robustness against various attacks like histogram equalisation, resizing, JPEG compression, Weiner filtering, etc. and still also retains the quality of the watermarked image. Thus, the proposed technique can be used as a solution to ensure authenticity via watermarking.