Mustafa S., Abdul Karim (2015) A study on data embedding in Universal Domain / Mustafa S. Abdul Karim. PhD thesis, Universiti Malaya.
Abstract
Conventionally, data embedding is a feature dependent process, where a feature of the host is modified to insert a payload while satisfying some properties. This dependency limits the interchangeability among data embedding methods. In other words, the applicability of a conventional data embedding method is restricted to certain types of signal. This restriction is observed in the most of the surveyed methods. Hence, a universal data embedding method applicable to any digital signal is nonexistent, albeit such method can potentially be applied in applications where feature extraction is technically challenging. For example, a cloud storage receives various multimedia contents. In addition, some contents are encrypted or compressed, which complicates the feature extraction process for data embedding purposes. In this study, the conventional data embedding methods are surveyed and evaluated in terms of interchangeability (Chapter 2). The problem of limited interchangeability is overcome by the proposed concept of universal data embedding, which is realized by a novel parser referred to as universal parser (Chapter 3). This parser segments the host signal into partitions of unified length referred to as IC’s (Imaginary Codewords). Theoretically, it is shown that the entropy (and hence redundancy) of IC’s changes based on the length utilized in the segmentation process. Thus, the defined redundancy is replaced by payload using four proposed methods. The first method, uREADS (Chapter 4), is based on mapping IC’s to GRC (Golomb-Rice Codewords). Then, GRC’s are modified to accommodate external information. However, uREADS has inconsistent and low carrier capacity. These problems are overcame by the second method, urDEED (Chapter 5), which applies a similar mapping to GRC’s, but with different way to handle the side information. urDEED offers consistent average carrier capacity of 0.169 bpb (bit per bit), and it is universally applicable to any encrypted signal. However, the mapping of IC’s to GRC’s involves the sophisticated processes in handling the side information. Also, this method is not applicable to high entropy signal. To overcome this problem, DeRand (Chapter 6) is proposed, which is based on histogram mapping. DeRand achieves a carrier capacity up to 0.4 bpb while being able to control the distortion in high entropy hosts such as random signals. The conventional concept of data embedding is further generalized to the novel concept of data fusion in (Chapter 7). Here, unlike conventional data embedding, which implies the processing of two signals only, namely, the host and the payload, the proposed data fusion can conceptually fuse two or more signals. A novel DSC (Dual Semantic Code) is proposed as a mean to realize data fusion, where each DSC codeword can accommodate two independent data simultaneously. The proposed data fusion achieves scalable carrier capacity, which can be further traded-off with file-size. All in all, the discussion in (Chapter 8) shows that the proposed data embedding methods are universal and superior to the conventional methods in terms of interchangeability. In addition, the proposed methods preserve file-size and they are reversible. Also, data fusion and DeRand offer scalable distortion and scalable carrier capacity.
Actions (For repository staff only : Login required)
![[img]](http://studentsrepo.um.edu.my/style/images/fileicons/application_pdf.png)
