以具有降低失真效果的資訊隱藏與驗證新技術做資訊的保護與安全傳輸

Abstract

隨著軟硬體的進步，數位資訊可以容易地製作、修改、與廣泛地傳播。因此，資訊安全已經成為近年來最重要且具挑戰性的問題之一。本論文致力於解決數位資訊的保護與安全傳輸的問題。 保護各種不同型態的數位資訊的一種方式是將其藏入影像中。以此方式，降低隱藏所造成的失真是主要的考量之一。為了設計不可視的資訊隱藏方法，在本研究中提出了三個降低失真的技術。首先，第一個技術是基於本研究中所提出的可藏度的新概念，它用來評估在原影像添加秘密資訊後的影像品質，只有可藏像素用來隱藏資料。另外兩個技術是以一個整體的角度使隱藏所造成的失真達到最小。其中之一是基於錯誤校正編碼，在所要隱藏的資料中加入可校正的錯誤，使其具有將隱藏所造成的失真達到最小的效果。這些可校正的錯誤可以被偵測到並校正，所以能夠還原所隱藏的資料。另一個技術是基於提供許多碼持有者做為用來隱藏資料的候選，最佳的碼持有者使得隱藏所造成的失真最少，被選擇用來隱藏資料。另一方面，也提出了五個新的資料添加與四個新的驗證技術以解決許多資訊隱藏與驗證的應用問題。 更明確地，我們提出了兩個對於黑白與調色盤影像的資訊隱藏方法，以及兩個黑白與調色盤影像的驗證方法。在所提出的資訊隱藏方法中，不僅一般的數位資訊能夠在不造成可見的失真的情況下，隱藏在黑白與調色盤影像中，而且也能夠驗證所隱藏的資訊。除此之外，一些設計需求之間的交易，包括影像品質與隱藏量，和驗證的正確性與驗證資訊的可攜度，皆被指出且也提出相對平衡這些需求的方法。 再者，我們提出了一個將黑白影像隱藏到雜訊影像的新方法。這個理由是在做資訊隱藏時，未必需要找一個有意義外顯影像。除此之外，基於錯誤校正編碼，我們提出了一個通用的資訊隱藏方式。數位資訊的隱藏和驗證是利用錯誤校正編碼的錯誤校正與錯誤偵測能力。 另一方面，我們提出了一個新的利用數位簽章的內容驗證方法。概念是由影像區塊中抽取顏色與線條的特徵，這些特徵對於內容保持的操作會保持不變，每個影像區塊的可信度可以藉由比較區塊的特徵與相對紀錄在數位簽章內的特徵得知。 以上所提出的方法的效能都已經由廣泛地理論分析與實驗做評估，結果顯示他們對於實際的應用皆可行。

With the advance of hardware and software technologies, digital information can be easily created, seamlessly manipulated, and widely distributed. Therefore, information security has become one of the most critical and challenging problems in recent years. This dissertation is dedicated to solving the problems of protection and secure transmission of digital information. One way to protect various types of digital information is to hide them in images. In this approach, reduction of embedding distortion is one of the major concerns. To design imperceptible data hiding methods, three techniques for distortion reduction are proposed in this study. First, the first technique is based on the new concept of embeddability proposed in this study, which is defined as a measure for checking the quality of stego-images yielded from embedding secret data in cover images. Only embeddable pixels are used to embed data. The other two techniques are proposed for minimizing embedding distortion from a global view. One is based on the use of error-correcting coding to create correctable errors corresponding to the data to be hidden, with the effect of minimizing embedding distortion. These errors can be detected, corrected, and so extracted to recover the hidden data. The other technique is based on providing multiple code holders as candidates for embedding data. An optimal code holder which results in the least embedding distortion is selected to embed data. On the other hand, also proposed are five new techniques for data embedding and four new techniques for solving various digital information hiding and authentication application problems have been developed. More specifically, two methods of data hiding in binary and palette images, and two methods of authentication of binary and palette images are proposed. In the proposed data hiding methods, not only can general digital information be embedded into binary and palette images without introducing noticeable distortion, but also can the credibility of the hidden information be verified. Besides, some tradeoffs between design requirements, including that between image quality and embedding capacity as well as that between the accuracy of authentication and the portability of authentication information, are addressed and solutions for balancing these requirements are proposed. Furthermore, a new method for embedding binary images into cover noise images is proposed. It is argued that there is no need to find a meaningful cover image for data hiding. Instead, a cover noise image which has larger embedding capacity can be used to hide a secret binary image. Nevertheless, a general approach to data hiding based on the employment of error-correcting coding is also proposed. Both embedding of digital information into images and authentication of the results are achieved by the error-correcting as well as error-detection abilities of an error-correcting scheme. On the other hand, a novel content-based authentication method based on the use of digital signatures is proposed. The concept is to extract visual color and edge features, which are invariant to content-preserving manipulations, from image blocks, as digital signatures. The authenticity of each image block can be verified by comparing its features with the corresponding ones recorded in the digital signature. The performances of all the above proposed methods have been evaluated extensively by theoretical analyses and experiments. The results indicate that they are feasible for practical applications.