強軔數位影像浮水印之偵測效能增進設計

Abstract

近年來數位浮水印研究的興起， 主要導因於多媒體資料廣泛而快速地傳布， 而引發數位智產權保護的須求。 在這份論文裡， 我們針對影像數位浮水印的強軔度， 偵測效益的穩定度， 浮水印容量， 浮水印的隱藏性四項浮水印重要的設計議題之間的關係做深入的研究。 我們提出兩種不同型態的數位浮水印方案, 兩者皆包含了理論分析, 實現與測試。 在本論文的第一個部份, 我們設計一個根據不同影像與不同攻擊方式而選擇嵌入浮水印有效係數的最佳化程序。 這組係數能增加浮水印的強軔度與偵測的可靠性， 同時亦能達成隱藏浮水印的目標而維持影像品質。 就某種程度上而言， 我們是在探討在假設已知可能的攻擊方式以及非盲目的偵測方式下， 離散弦波轉換域的浮水印的效能極限。 由於目前的影像資料多經由JPEG壓縮以達成有效的傳送與儲存， 所以JPEG壓縮被引用為設計中攻擊方式的第一個例子。 但事實上， 由於我們所提出的程序具一般性且系統化的特性， 所以此程序亦可運用來開發抗其他種類攻擊的離散弦波轉換域的浮水印有效係數。 因此， 我們亦引用JPEG2000壓縮為另一種的攻擊方式。 根據由以上理論最佳化程序運用於自然影像而獲得的資料， 我們推衍出一組簡化的有效係數選擇的規則， 以快速地選擇有效係數， 降低計算複雜度。 我們藉由參數化的分類法來達成此設計目標。 這些統計歸納出的簡單規則雖然大量降低了原來最佳化程序所需的計算量， 但是仍能同時增進浮水印強軔度與降低錯誤偵測率。 在論文的第二個部份， 我們建構一個盲目偵測的強軔影像數位浮水印。 這個方案結合了抽取影像特徵與影像正規化的技術。 設計的目標則是抗幾何形變與一般數位訊號處理的攻擊。 藉由嵌入浮水印時所抽取出的特徵點仍能在影像經過攻擊後存在的特性， 而使得浮水印嵌入器與偵測器可藉由參考這些特徵點以減少複雜的同步問題。 另一方面， 根據影像中正規化的物件具有在影像旋轉後不變的特性， 浮水印偵測前的同步問題亦可獲得妥善的改善。 實驗結果顯示這個數位浮水印方案能對抗低品質的JPEG壓縮， 色彩降級， 高斯濾波， 中位濾波， 欄和列的移除， 修剪， 旋轉， 局部形變， 切割， 以及線性幾何轉換的攻擊。

In the past few years, digital watermarking has received much attention mainly due to the urgent demand for the copyright protection on the widely distributed digital data. In this thesis, we study the relationship among watermark robustness, detection reliability, data payload, and imperceptibility for digital images. We propose two different types of image watermarking schemes. They both include theoretical analysis and realization and testing. In the first part of this thesis, we design an optimization procedure for selecting the most effective DCT coefficients for watermark embedding. Using this set of coefficients improves the watermark robustness and reliability against attacks and in the meanwhile it maintains the visual transparency of the embedded watermark. To a certain extent, we try to find the “performance limit” of the DCT-domain invisible watermarking technique under the assumptions of known attack and non-blind detection. Since digital images are often compressed for efficient storage and transmission, the popular JPEG compression is used as one attacking example in our design. However, what we propose is a generic and systematic approach of finding the most effective watermarking coefficients in DCT-domain watermarking. A second example using JPEG2000 as the attacking source is also presented. Based on the theoretically optimized data set obtained using the preceding scheme, a set of coefficient selection rules is derived with the help of parametric classification technique for determining the effective DCT watermarking coefficients without going through a costly iterative process. These rules are simple in computation. They improve the watermark robustness (correctly decoding) and, in the mean time, decrease the error detection probability (correct detection). In the second part of this thesis, a robust and blind digital image watermarking scheme combining image feature extraction and image normalization is developed. Its goal is to resist both geometric distortion and signal processing attacks. The extracted feature points can survive a variety of attacks and can be used as reference points for both watermark embedding and detection. The normalized image of an image (object) is nearly invariant with respect to rotations. As a result, the watermark detection task can be much simplified when it is applied to the normalized image. Simulation results show that our scheme can survive low quality JPEG compression, color reduction, sharpening, Gaussian filtering, median filtering, row or column removal, shearing, rotation, local warping, cropping, and linear geometric transformation.