Digital Watermarking of Stereoscopic Images Using Depth Image Based Rendering and Machine Learning

Abstract

Digital Watermarking of Stereoscopic Images Using Depth Image Based Rendering and Machine Learning The increasing acceptance of 3D-TV in recent years has attracted attention of the research community and industry towards the technology. Depth Image Based Rendering (DIBR) is the latest advancement in 3D-TV and free-view television. With the widespread of 3D content, copyright protection is becoming a serious concern for maximizing the profits and preventing illegitimate acquisition and distribution of the 3D media content. Digital watermarking is one such promising technique to tackle this problem. This thesis presents a comprehensive literature study of application of digital watermarking techniques, i.e., the three different paradigms for 3D images in particular and 3D-TV system in general. These paradigms include robust, fragile, and reversible watermarking for 3D system. In addition, this work also presents an improved reduced reference image quality assessment for 3D images along with a comprehensive literature survey of the field. Firstly, this thesis proposes a robust watermarking technique for 3D-TV. The proposed technique is based upon intelligent parameter selection using Genetic Algorithm (GA) in an improved spread spectrum based watermarking system. In order to enhance robustness, another layer of security of the watermark is added using Bose–Chadhuri–Hocquenghem (BCH) coding. Experimental results and comparison with state of the art technique demonstrate the effectiveness of the proposed technique to structure the watermark for high robustness in the presence of a number of hostile attacks. The second phase of this research proposes a fragile watermarking scheme for 3D-TV system for the purpose of content authentication. The proposed technique is based upon implicit watermarking of a number of other un-watermarked coefficient from a dependence neighborhood in a DCT based watermarking schemes. Such a scheme is robust against common counterfeiting attacks such as collage attack, cover up attack, transplantation, and vector quantization attack. In addition, the propose scheme is also capable of localizing the tempering of the cover work. x In the third phase of this research work, a reversible watermarking technique for DIBR 3D-TV is developed. The DIBR 3D-TV or free-view TV is one of the most promising technique in multimedia world. So, the protection of these valuable contents is an important concern in the world of digital processing. This study exploits the interpolation scheme by applying Genetic Programming (GP) based intelligent reversible watermarking technique. Previously, the empirical solutions are not that much effective and use hit and trial strategies for selecting optimal space for watermark embedding. The proposed scheme achieved considerable watermark capacity as well as Structure Similarity Index Metric (SSIM) and Peak Signal to Noise Ratio (PSNR), compared to existing state of the art contemporary schemes. The existing methods performed well against GA and Particle Swarm Optimizer (PSO), but show relatively less performance compared with Tian’s method. However, the PSNR value in the proposed method outperforms against Tian’s scheme. The proposed algorithm is tested on different standard publicly available 3D datasets. It is observed that the proposed scheme gives relatively better results compared with state of the art techniques. Last phase of this research work proposes an enhanced reduced reference image quality assessment technique. Reduced reference image quality assessment (RR-IQA) technique does not require the presence of the original image for assessing the quality of a degraded image. This work proposes an intelligent method for reduced reference image quality assessment based on Reorganized Discrete Cosine Transform (RDCT). The GA is used to compute optimized estimation of Generalized Gaussian Distribution (GGD), which then approximates the coefficient distribution in RDCT domain. Experimental results validate that such an intelligent estimation produces far superior results as compared to conventional empirical estimation methods as presented in literature. The comparative analysis of the proposed technique with a number of contemporary techniques present in the literature demonstrate the generalization capability and effectiveness of the proposed technique as compared to prior works.