A Study on Object Image Watermarking and Palette Image Steganography 摘要

The current rapid development of information technologies makes provision of commercial multimedia services on Internet. However, in rerum natura, digital multimedia is easy to be copied, re-scaled, rotated, transformed, and modified using handy processing software without much effort. This fact has derived an urgent demand for reliable and secure copyright protection for digital multimedia. Watermarking is the technology used for copy control and media identification and tracing. Most proposed watermarking methods are designed for the protection of digital images. Theses methods embed a short message (a watermark) in the image without affecting the usability but that can be detected using dedicated analysis software.
Watermarking techniques have been widely discussed and developed in recent years. However, studies which focus on watermarking techniques those can resist geometrical attacks are still not quite mature. Geometrical attacks like rotation, scaling, and transforming will miss-position the location of watermark embedded. There are some previous works working on this problem for watermark re-synchronization. We use the geometrical features of object shape in a digital image to achieve the goal of watermark re-synchronization to obtain quite remarkable imperceptibility and robustness.

In this dissertation, we aim to design some information hiding techniques for protecting images and private messages of communication on the Internet. Four information hiding related schemes are proposed. They are (1) Geometric attack free self-synchronized object image watermarking schemes (2) Palette-based object image watermarking scheme (3) Palette-based image Steganography scheme, and (4) Hierarchic texture classification based on statistical steganography.
First of all, to achieve the goal of watermark re-synchronization and embed digital watermark into an object image, geometric features of an object image like the length of principal axes, eigenvectors, and orient of mass center are use for subdividing the object shape. Using those geometric features for shape subdividing makes re-synchronization of watermark detection possible. In our study, different shape subdivision methods, including self-similarity, parallel segmentation and recursively principal axes, are proposed for object watermark embedding. And the geometric attack resistance schemes for object images can be applied to different image formats, including raw and palette-based images. Unlike the true color images and gray level images, palette-based images have discrete color palette. To embed watermark into limited colors and hold the imperceptibleness and robustness is a challenging work. An algorithm for palette-based object image watermarking is proposed here. These proposed methods are easily implemented and can obtain comparable NC values to against outside geometrical attacks such as rotation, scaling and translation.
For the issue of steganography, an iterative method of palette-based image steganography scheme is proposed to improve the existed methods. Better quality of stego- images can be produced using the proposed method. Besides, we also proposed a hierarchic texture classification using statistical steganography technique to obtain robust texture features for texture classification. Experimental results show that the proposed methods can achieve high accuracy rate and also work well even the query textures are distorted by noise corruption or compressed.

近年來由於網際網路的發達，使得現代人可以輕易的自網際網路下載與傳送各式影像，或是經由影像軟體，將影像放大、縮小和切割，甚至隨意修改影像的內容。為了落實保護智慧財產權的目的，數位浮水印技術，透過編碼的方式，將資料轉成數位訊息，再利用浮水印嵌入的技術，將浮水印藏匿至數位影像、媒體中。由於在嵌入浮水印時，即考慮到可能遭受到的攻擊與破壞，因此，即使有人非法盜用或利用影像處理軟體進行竄改，皆可經由浮水印偵測/擷取技術，取出浮水印，驗證數位影像、媒體之所有權。

在過去幾年當中，浮水印技術的研究已逐漸受到重視。然而，一般的浮水印技術，只能抵抗普通的影像處理，對於旋轉、縮放、座標轉換等幾何攻擊，還未能做到有效的保護，主要的原因在於浮水印藏入與取出位置的同步定位的問題，由於藏入浮水印後的影像在經過幾何攻擊後，浮水印資料所在位置難以掌握，使得所藏入的浮水印無法正確被抽取或是偵測出來。目前一些相關的研究如藏入看不見之固定樣式的圖案或是利用特徵點所組成的三角形來定位浮水印位置等方法，然而這些方法仍有改善的空間；本研究的主要目的是利用影像中物件形狀的特性來作為同步定位的參考，並藉由形狀的切割之改良式拼湊法以發展可抗各類幾何攻擊之數位浮水印方法。
此論文提出兩種利用物件主軸以及主軸方向所定位之空間浮水印方法：(1)利用物件形狀切割之浮水印方法以及(2)抗幾何攻擊之調色盤影像浮水印方法。首先，為達到同步定位之目的，利用物件形狀以及幾何特性探討各種取得補丁(patch)之物件形狀切割方法；應用於各種切割方法中的幾何特性包括主軸長度、質心的座標以及特徵向量等；運用這些特性，所嵌入之數位浮水印便可抵抗幾何攻擊。這樣的抗幾何攻擊之數位浮水印法亦可適用於其他影像格式，例如調色盤影像。有別於灰階以及全彩影像，調色盤影像具有顏色不連續的特性，要在有限色盤的限制下兼顧浮水印之不可見性(imperceptible)以及強固性(robustness)是具有挑戰性的工作，本論文中提出利用物件形狀的同步定位應用在調色盤影像浮水印之演算法。另外，影像或照片之變造也是數位浮水印被廣為探討之議題之一，運用物件形狀以及幾何特性，本論文亦提出一個可以抵抗物件切割與張貼(cut-and-paste)攻擊之技術，實驗結果說明所提出的方法具有優良的效果。
另外在資料隱藏(steganography)技術方面，本論文也提出兩種方法：具高品質之調色盤影像資料隱藏方法以及利用統計資料隱藏技術於紋理分類的方法。我們利用疊代方式尋找調色盤中較不重要的顏色，將空出的顏色以最有資料隱藏價值的顏色取代，持續疊代的結果，可以改進常用的調色盤資料隱藏方法，產生更高品質的偽裝影像。最後我們利用統計資料隱藏技術中抗攻擊的特性，轉換成為紋理影像的特徵用於紋理影像的辨識，實驗結果說明所提出的方法仍可有效辨識經雜訊或是壓縮攻擊過之紋理影像，具有高度的實用性。