Robust Image-Adaptive Watermarking Scheme with Strength Factor Analysis

Abstract

Towards the end of 20th century digitalization had started to spread in all forms of the technology related areas. Around the same time internet had also begun to turn the whole world into a global village. Though digital form of data had many benefits over its analog counterpart like ease of storage and transmission, still a lot of multimedia content providers were skeptical of it due to the fear of probable copyright violation by the high quality duplication possible with the digital data. As necessity breeds the invention, watermarking techniques came in the forefront to tackle this problem. In watermarking a secret and invisible data is hidden in the digital content. This secret data could be a logo or some other information for correctly identifying the original owner of the content. At the time of conflict the original owner could easily prove his ownership by extracting the hidden information from the digital content. A lot of research work carried out in the watermarking field, emphasized the use of perceptually significant areas of the image for robust watermarking. HVS models facilitated the use of certain frequencies and chromatic components for embedding the watermark, towards which the human eyes have low sensitivity. Many image-adaptive techniques espousing these two doctrines of robustness and imperceptibility have been proposed. Following the footsteps of the preceding researchers, the presented work is an effort to provide better and less complicated robust watermarking techniques by applying image-adaptive measures. In the available literature there is no study to suggest if any one size of block segmentation in image could provide better robustness than the other size of block segmentation for the same watermarking technique. In the presented work three sizes of block segmentations (8×8, 16×16 and 32×32) have been used in the proposed DWT based semi-blind image-adaptive watermarking technique. By comparing the results of these block segmentations it was found that 8×8 sized block segmentation is most predisposed to augment the robustness of an embedded watermark. Dynamic strength factor is used in the presented technique for better control over the quality of the watermarked image. By using the dynamic strength factor watermarked image of any desired quality can be obtained. In this thesis all the watermarked images have been kept at PSNR value of 45 dB, because PSNR value of 45 dB is an acceptable measure of very good visual quality for the images. The value of dynamic strength factor for a particular block depends v on the standard deviation of the low frequency DWT coefficients belonging to that block. In this way the image-adaptivity propagates to the very basic building blocks of the image and hence helping to make the watermarking technique more robust. The proposed technique outperforms many techniques in literature by providing better robustness of the extracted watermark when exposed to various watermarking attacks. A hybrid watermarking technique which uses DWT, SVD and DCT is proposed next. Dynamic strength factor in this technique is derived from the standard deviations of the DCT coefficients belonging to the selected blocks for watermark embedding. When exposed to the common watermarking attacks this hybrid technique shows better performance than the DWT based technique and two established techniques of literature. In the quest to find an even better robust watermarking another hybrid watermarking technique is proposed which uses DWT, SVD and WHT. Three local properties of the image are used to derive three dynamic strength factors of the proposed technique. Out of the three versions of the proposed technique using three strength factors, one version that uses entropies of the watermarking blocks provides the best results. This version of the proposed technique called as DWT-SVD-WHT (entropy) shows much better performance than the previously proposed two techniques and many of the other watermarking techniques of literature. So this could be termed as the most robust watermarking technique presented in this thesis. As the proposed techniques are semi-blind in nature, side information generated in the watermark embedding is used in the extraction of watermark. Security of the side information has been provided by an innovative method. This method ensures that even if side information gets intercepted illegitimately, there is only less than half percent chance that it can be used to detect and destroy the watermark. An encoding method has also been used which provides an additional security to the side information.