On Extreme Learning Machine for Watermarking of an Images in Discrete Wavelet Transform Domain

In this paper, an implementation and efficacy of Extreme Learning Machine (ELM) algorithm for watermarking of an images in Discrete Wavelet Transform (DWT) domain has been demonstrated. ELM is a regularization algorithm works based on the concept of generalized single-hidden-layer feed forward neural networks (SLFNs) with different activation functions likes rbf, sine, sigmoid and hardlim in hidden nodes in unified environment framework. In this learning method, the parameters of hidden nodes like the input weight and bias value of additive nodes are randomly selected based on input data samples. This algorithm developed for batch learning is extremely good and has better generalization performance. Except from selecting the number of hidden nodes, no other learning parameter is selected here manually. Detail performance and efficacy of this algorithm is tested on watermarking purpose on color images in Discrete Wavelet Transform (DWT) domain. The results show that watermarking scheme based on ELM is very robust and imperceptible and produces better generalization performance against common image processing attacks.