A time-efficient optimization for robust image watermarking using machine learning. (15th June 2018)

• Record Type: Journal Article
• Authors: Abdelhakim, Assem Mahmoud; Abdelhakim, Mai
• Is Part Of: Expert systems with applications. Volume 100(2018)
• Page Start: 197
• Page End: 210
• Publication Date: 2018-06-15
• Subjects: Meta-heuristic optimization -- Machine learning -- Robust image watermarking -- Discrete Cosine Transform -- Artificial Bee Colony -- K-Nearest Neighborhood; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2018.02.002
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: An efficient approach for optimizing image watermarking is proposed. Machine learning is utilized to solve the optimization time consumption problem. New robust watermarking method is presented. Embedding parameter is optimized in terms of watermarking quality and robustness. The K-NN regression method is used for predicting the optimum value. Abstract: Watermarking is utilized for securing multimedia data exchange. The techniques used to embed a watermark typically require optimizing scheme parameters often through applying meta-heuristic optimization techniques. Although meta-heuristic techniques have been widely used due to their performance enhancing capability, they are not suitable for time sensitive applications due to their large time consumption. In this paper, a time-efficient optimization based on machine learning is proposed to find the best embedding strength parameter for robust image watermarking in terms of both the imperceptibility and robustness. First, a watermark embedding scheme is designed in the Discrete Cosine Transform domain, which provide a proper robustness against common watermarking attacks. Then, a training process is performed through selecting set of images upon which the watermarking is applied and optimized using the Artificial Bee Colony algorithm. Observation data is collected, which includes the optimum strength values along with the feature vectors that represent the training images. The features, of an image, are extracted by calculating the optimization fitness function at different values of the embedding strength. Finally, new set of images are chosen to be watermarked using the optimum embedding parameters that are predicted through the K -Nearest Neighborhood regression method. Experimental results show that the proposed method consumes considerably less time to evaluate the optimum solutions compared to using meta-heuristic optimization. Meanwhile, the error between the optimum and predicted optimum solutions has negligible impact on the optimization objectives.