Image-Adaptive Watermarking Under Geometrical Attacks

Abstract

With the evolution of digital forum and internet telecommunication, multimedia essentialities are becoming more notable and acceptable boost in people's daily life. However, this ascendancy of digital contents of media proves to be a major drawback in terms of copyright contravention and unauthorized use of information content. Digital watermarking has expeditiously progressed during the past two decades, which can provide methodical solutions for protection of proprietorship, identification of trademark, corroboration of data legitimacy, tracing of copies of digital content, reprint and ingress controls, etc. Digital image watermarking is an approach to enclose an undisclosed signal (watermark) within a specific image (cover image) and extricate the untold message at the terminus, thereby fortifying the image from signal filtering attacks during the channelling process. The proposed study concentrates on the research topic of robust image watermarking technique for protection of copyright and imperative authentication under structural and coincidental attacks.The key specifications of any watermarking technique are robustness, capacity, imperceptibility and security. While the expertise of safeguarding the watermark under several attacks is called as robustness, the foretelling of information implanted in the selected cover work is known as capacity. The security refers to potentiality of the watermarking technique to endure intentional complicity. Though many researchers embraced these stipulations, but an insubstantial balance was achieved between the desired requirements. Hence, an image-adaptive approach was suggested to achieve the favourable characteristics by adapting the features of the digital image to recover the hidden watermark in the best possible manner.The available literature demonstrated limited statistical approach in image-adaptive watermarking technique to accomplish the extraction of the watermark under various attacks. The main intend of this research is to develop robust and coherent image-adaptive watermarking techniques. Discrete Wavelet Transform (DWT) is used on (8 ×8) size chunks of the host image, wherein the low coefficients of frequency (LL) are modeled using Gaussian distribution to achieve high Peak Signal to Noise Ratio (PSNR) values of watermarked images with minimum BER (%) of recovered watermarks. Different statistical moments of central tendency have been used to hide the watermarks using a flexible strength factor, which were later extracted using a decoder employing Maximum Likelihood (ML).With the advent of time, need for multiple watermarking has also proliferated. Therefore, the next proposed technique tendered to insert two watermarks into LL and LH sub bands using v DWT with high PSNR values. Kurtosis is used to insert the watermark bits utilizing an adjustable strength factor. It helps to recover the watermarks with fewer errors. The statistical decoder helps to reclaim both the watermarks by making use of the perspective of Gaussian modeling with minimum Bit Error Rate (BER in %) and Gradient Mean Structural Deviation (GMSD) values. Though many researchers suggested the use of more than one transform for robust watermarking, again a limited success was achieved against common and geometric attacks. The third proposed technique offered to use a combination of DWT and Fast Walsh Hadamard Transform (FWHT) for obtaining good PSNR values. This hybrid technique outplayed the existing techniques against many geometric attacks using the ML decoder. The values of Normalized Cross Correlation (NCC) and Structural Similarity Index Measure (SSIM) too approached the desirable value of unity utilizing this approach.Another improved hybrid technique has been proposed to embed two watermarks simultaneously using the transforms DWT and FWHT for high PSNR. Five Hybrid Strength Factors based upon mean, standard deviation, entropy, etc were used for inserting the watermarks. The statistical decoder is again used to recover the watermarks with increased robustness and high imperceptibility. The values of Image Quality Assessment (IQA) parameters, such as, PSNR, NCC, SSIM, GMSD, UIQI and BER (%) too proved good efficacy against other state of art techniques.The utility of all the proposed techniques lies in real time applications of tele-medicine. Alike the natural images, different sized watermarks are used to hide the information of patients in medical images too. All the proposed techniques are semi blind in nature. Thus, side information is communicated to the receiver for the successful recovery of the watermarks. The security of the side information is also ensured. The study may be broadened for colored images through the utilization of other transform domains too in future.