International Journal of Computer Applications (0975 – 8887) Volume 110 – No. 4, January 2015 4 Survey on Image Watermarking Schemes using Adaptive Soft Computing Techniques Amita Goel Department of Computer Science, Teerthanker Mahaveer University Moradabad,India Anurag Mishra Department of Electronics, Deendayal Upadhyay College, University of Delhi, New Delhi,India ABSTRACT During last few years, many soft computing techniques have been employed for image watermarking. These are more into delving the issue of optimization of visual quality of signed images and robustness of the embedding algorithm. The used techniques either operate in adaptive or learning mode, especially those using Artificial Neural Networks or in non adaptive analytical mode such as ones based on Fuzzy logic. Several researchers have also worked on this problem using hybrid and evolutionary algorithms. This research survey especially deals with the image watermarking techniques which rely on adaptive soft computing techniques. The results of gradient descent based Back propagation Network (BPN algorithm, Radial Basis Function Neural Network (RBFNN algorithm and a newly developed Single Layer Feed forward Neural Network (SLFN algorithm commonly known as Extreme Learning Machine (ELM) used to carry out watermarking in uncompressed grayscale images are compared. These techniques are compared for different images and the comparison is based on the visual quality of signed images, the watermark detector response coefficients such as similarity correlation and normalized correlation parameters and the robustness studies. Time complexity issue is also examined to establish the use of watermarking process on a real time scale. It is concluded that the ELM algorithm gives a reasonable generalized behavior in terms of computation of these parameters as compared to its other counterparts. It’s fast training in milliseconds and subsequent embedding and extraction makes it suitable for developing watermarking application on a real time scale. General Terms Image Watermarking, Soft Computing Techniques Keywords BPN, Radial Basis Function Neural Network, ELM 1. INTRODUCTION Watermarking of grayscale images is an advanced area of research in the signal processing domain. It is frequently used for copyright protection, owner identification and content authentication and authorization. The content authentication and authorization specifically helps consumers to:  Find how, when and where the copyrighted content is being consumed  Protect the digital content from any possible unauthorized use  Link consumers to and from the content that is relevant to their interests  Provide a way for users to easily identify the source of content by communicating copyright information and identify the appropriate channels for licensing or purchasing  Organize and manage content in a digital asset management system  Identify the source of leaks when confidential content inadvertently or intentionally makes its way onto the Internet Figure 1 depicts a typical image watermarking system [15]. Figure 1: A typical image watermarking system To this end, several robust image watermarking schemes have been developed [1-4]. Broadly, these schemes belong to pure signal processing domain in which the algorithms specifically deal with pixel coefficients in transform domain using standard mathematical formulations. Another branch of algorithmic evolution deals with the development of watermarking schemes using standard soft computing techniques. This is particularly due to the fact that the problem of watermarking of images is now converged as an optimization problem. The twin parameters – visual quality of the signed images and the robustness of the embedding scheme are often found to be mutually exclusive, keeping the embedding capacity as constant. This is more because, the embedded content is far less than the size of the host signal (grayscale image). If visual quality is enhanced, it is often carried out at the cost of robustness which makes the signed images vulnerable to signal processing attacks, thereby easily removing the watermark. On the other hand, if robustness is increased beyond a certain limit, it can only be done at the cost of visual quality of the signed images which makes the signed content hardly of any practical use. To optimize these parameters, adaptive soft computing techniques are Transformed Image + Signed Image Watermark Generator IDCT / IDWT DCT / DWT Original Image