Efficient Algorithms for Embedding Digital Watermark in Curvelet Domain

Abstract

Due to the popularity of the internet and its increasingly easy access to digital multimedia, many powerful tools are available for editing digital media without the loss of quality. So, authentication and intellectual property rights of digital media are critical issues. The Intellectual Property Rights (IPR) of the author are required to be protected i.e. Copyright Protection. The number of solutions to this problem, from encryption to watermarking, is growing every year. Many authors are working in the field of watermarking to protect the author ownership. Among these watermarking algorithms, some algorithms are better than the others in terms of basic watermarking requirements like invisibility, robustness and computational cost, etc. In this thesis, we propose six watermarking techniques based on curvelet transform, which can be applied to different problems. The main aim of these techniques is to use the properties of curvelet transform and show that these are more suitable for hiding the watermark in more robust and invisible manner. The first technique proposed is a non-blind watermarking technique for embedding a watermark in different scales of curvelet transform domain. The quality of extracted watermark of curvelet domain embedding technique is compared with wavelet domain at different number of decomposition levels. This technique is an application to a grayscale image. The second technique is also a non- blind watermarking technique. This technique utilizes the property of HVS as the edge part of an image is less sensitive. Imperceptibility of the watermark is high if it is inserted into the texture part rather than smooth part. Accordingly, in this application, information is embedded into the texture blocks of the cover image by using curvelet transform. This technique can be applied to protect the author ownership for gray scale images. The third technique proposed is a semi-blind watermarking technique of embedding the color watermark using curvelet coefficient in the RGB cover image. This algorithm uses the bit plane method and the blue color plane of the cover image for embedding the watermark. The most significant bit (MSB) plane of watermark image is embedded into the selected scale and orientation of the curvelet coefficients of the blue channel in the cover image. The fourth algorithm combines the technique of cryptography and watermarking for application to color images. This technique demonstrates a secure, robust and semi-blind watermarking technique for a color image by using Bijection mapping function and curvelet domain. The fifth and sixth techniques proposed here are blind digial image watermarking techniques. These techniques are more suitable for the application of copywrite protection, forgery detection and security. The fifth technique embeds a random sequence i.e., watermark, into the curvelet transform of the color image. To make this technique more secure and robust, the luminance of the cover image is used. The luminance part of the image is transformed into curvelet domain and clustering approach is used to embed the watermark into the selected scale and orientation of curvelet domain. This technique is used to detect tampering in an image. First the technique identifies whether an image is tampered or not by comparing the embedded watermark with extracted one. Second, if image is tampered it locates the tampered region of the image. The sixth technique is applicable to a variety of medical images. Here, it embeds the patient’s information used as watermark into curvelet domain of ECG signal. Nowadays, with the help of wearable medical devices, it is easy to monitor a patient even from remote locations. Patient’s information is broadcast to the hospital servers over wireless or wired media without any security. An electrocardiogram (ECG) is simply a representation of the electrical activity of the heart muscle as it changes with time. ECG can provide useful information and remains a crucial element for the assessment of cardiac patients. Embedded watermark into an ECG signal is a challenging work, because any change in ECG signal will affect the diagnosability of ECG. In this technique, it is emphasized that embedding a watermark as image is more robust as compared to text or numbers. In this technique, the QRS complex attributes of ECG are preserved so that embedding patient information does not affect the diagnosability. This algorithm is also applicable to other medical images e.g. EEG, CT scan, MRI, X-Ray etc. In addition, the thesis provides hypothetical analysis for the performance and practicability of all the above-said techniques. To evaluate the performance of proposed techniques, performance evaluation metrics such as PSNR, NC, SSIM, BER and MSSIM are used. For sixth technique, the embedding domain is 1D signal, so the performance of technique is evaluated by all above said metrics as well as by PRD and KL as additional metrics. Also, we present experimental results to authenticate the hypothetical explanation and comparison of results for all the algorithms with existing popular technique.