抗傳輸通道損失與幾何失真之強韌性浮水印

Abstract

隨著網際網路、多媒體與電子商務的盛行，大量數位化資訊的傳送與儲存變的既快速又方便，數位化資訊已逐漸融入我們日常生活中，因此在資訊安全課題上對於個人私密性資料的保護與智慧財產權的保障越來越受到重視。浮水印(資訊隱藏)技術是將某些重要訊息隱藏於文字、聲音、影像或視訊等多媒体資料中，以達到所有權保護、防止盜拷、認證、內容連結(隱藏性標題)與秘密通訊等多種應用。由於浮水印用途廣泛且具潛在商機，其牽涉到的技術包括密碼學、數位信號(影像與聲音)處理、資訊理論與數位通訊各個研究領域，兼具理論與實用價值，因此近十多年來吸引了學術界與產業界眾多人士投入相關的研發。本論文主要目標則在於研發抗傳輸通道損失與幾何失真之強韌性浮水印嵌入與偵測演算法，適於所有權保護、內容連結與秘密通訊之多種應用需求。 在論文的第一部份，我們探討適用於在不可靠之IP傳輸網路所需之容錯架構及其浮水印嵌入與抽取演算法，並提出兩種可行之方法。採用的容錯架構為多重描述編碼(Multiple Description Coding, MDC)，首先在浮水印嵌入與抽取演算法提出了索引值餘數量化法(Quantization Index Modulus Modulation, QIMM)，此法經理論分析與大量實驗結果顯示其與目前最尖端的索引值量化法(Quantization Index Modulation, QIM)效能相當。接著我們整合了QIMM與QIM於MDC，得到多重描述浮水印(Multiple Description Watermarking, MDW)架構。方法一將浮水印嵌入在任一個子描述 (side description)，其可由接收到之任一個子描述抽取出浮水印。方法二則另提出一更佳之浮水印嵌入與抽取演算法，稱之為多速率格子索引值量化法(Multi-Rate Lattice Quantization Index Modulation, MRL-QIM)，並將浮水印嵌入在中央描述(central description)，也可由接收到之任一個子描述抽取出浮水印。相較於方法一，方法二之優點在於其所利用之MRL-QIM浮水印嵌入編碼效益較高且浮水印嵌入在中央描述較一般化，彈性較好。 在論文的第二部份，我們探討了幾何不變性數位浮水印之可行方式。首先我們提出結合根據情報先行編碼(informed coding)與Foruier-Mellin轉換之抗旋轉、縮放與平移以及其它多種攻擊之浮水印方法。主要是根據情報先行編碼前處理(informed coding pre-processing)以獲得一最佳浮水印，再嵌入於影像特定區域之Foruier-Mellin幾何不變域。此機制相較於原先未採用根據情報先行編碼前處理之方法，在相同的誤判率(false alarm rate)下大幅提高了偵測率(detection rate)。不同於利用幾何不變域之方法，另一類為利用重新同步(re-synchronization)或自我同步(self-synchronization)之方法，我們也嘗試了多種解決方式並提出一種利用二維條碼之自我同步數位浮水印嵌入與偵測方法。此方法為採用兩階段式(two-stage)之方式，主要為在第一階段採用QR Code (Quick Response Code)二維條碼之編碼方法來編碼數位浮水印訊息(payload)，第二階段利用索引值餘數量化(QIMM)來執行QR Code數位浮水印訊息之嵌入與抽取。由於QR Code 二維條碼具有高容量(capacity)、高密度(compact size)與高容錯之特性，有利於數位浮水印訊息之編碼與解碼。另外透過QIMM在高容量訊息嵌入與偵測能力可有效的完成幾何不變性數位浮水印。本方法可有效的將數位浮水印訊息隱藏於影像中並且能抵抗包含旋轉、縮放與平移之幾何攻擊，在可隱藏訊息量、影像保真度(transparency)與強韌性三方面之綜合功效優於現有之方法。

Watermarking is a technique to hide data or information imperceptibly within image, audio or video so that valuable contents can be protected. Since the application of watermark is extensive and its market potential is quite promising, and the design of watermarking algorithms implies the integration of many concepts coming from cryptography, digital communication and signal processing. The development of efficient watermarking algorithms has been a very active topic for researchers in this area. We focus on two categories of problems. One is the problem of watermarking for error-prone transmission over unreliable network and the other is the problem of achieving watermark robustness against geometric attacks. In the first part of the dissertation, we study the problem of watermarking for error-prone transmission over unreliable network and two approaches are proposed. The first approach is to integrate oblivious watermarking techniques (quantization index modulus modulation (QIMM) and QIM) with the multiple description coding (MDC) to get a multiple description watermarking (MDW) framework. In this framework, the watermark embedding is computed in either one description and could be extracted with the reception of either one or two descriptions. The main drawback of previous mentioned approach is that both the watermark embedding and detection are performed on side description rather than on central description. The other problem is that both QIMM and QIM are quite limited under value-metric attack. Stimulated by the above-mentioned issues, in the second approach we attempt to find an improvement by studying the problem of watermarking under multiple description diversity transmission from a different perspective; namely, watermark embedding is done in the central description while watermark detection can be done in either central or side description. The merit of watermark embedding done in the central description is that the embedding and detection do not interfere with the MD mechanism. Therefore, this approach is more flexible than the one done in the side description. Furthermore, we propose a blind multi-rate lattice quantization index modulation (MRL-QIM) watermarking technique to boost the effectiveness. As the proposed MRL-QIM encodes two watermark bits into each of the four co-set points of a lattice (multi-rate), the payload (capacity) and robustness of watermark detection will be significantly upgraded. In the mean time, the fidelity of the watermarked image is also preserved. In the second part of the dissertation, we study the problem of achieving watermark robustness against geometric attacks. This problem has always been a challenging research topic. We firstly propose an RST (rotation, scaling and translation) resilient image watermarking technique using Fourier-Mellin transform and informed coding of watermark message. The watermark is embedded in the geometric invariant Fourier-Mellin domain, and no additional features need to be extracted to form a geometric invariant embedding space. Moreover, by informed watermark coding, our scheme could embed a weak watermark signal (i.e. one that needs only small perturbations with the host signal) and detect a slightly weaker watermark under the ILPM and the inverse Fourier transform. Secondly, we employ the famous QR Code (Quick Response Code) by first encoding the watermark payload, and then embedding the QR coded watermark into the image spatial domain. Thanks to the characteristic of position detection pattern of QR Code, the self-synchronized QR coded watermark payload can be recovered against geometric distortions once the QR Code is extracted during detection. Experimental results demonstrate that by adopting our approach, the resulting watermark is robust to a variety of combinations of RST attacks while preserving the visual quality of the watermarked image, thereby resolve the unavoidable dilemma faced by the other schemes.