- 學位論文
一個以小波轉換與人體視覺系統為基礎的強健影像浮水印法
A Robust Image Watermarking Method Based on Wavelet Transform and Human Visual System
摘要
網際網路的快速成長對多媒體資料的安全(如數位視訊、音樂、影像等資料的未授權傳遞)形成一個極大的挑戰。為了能在無所不能的網路上對多媒體資料有效的保護，數位浮水印(digital watermarks)技術逐漸受到重視。將一個稱為浮水印的註冊商標藏在多媒體資料裡，可用來解決資料合法擁有者的版權爭議。 本論文將針對多媒體中的數位影像資訊，探討其數位浮水印技術。我們利用小波轉換(wavelet transform)提供的空間局部區域(spatial local)及空間全域(spatial global)的特性，並加入人體視覺系統(human visual system, HVS)的考量與錯誤控制碼(error control coding, ECC)，使數位浮水印具有透明度(imperceptibility)與堅強度(robustness)的特性，進一步達到 網際網路多媒體智慧財產權保護的目的。另外不同於傳統上在影像中所嵌入的浮水印，我們將一幅做為辨識用的二元標誌影像或一個數位簽章的浮水印隱藏至另一幅影像內，並由浮水印數列和原始影像中用單方向雜湊函式產生浮水印私鑰，避免發生多重所有權死結的問題及大幅降低萃取時需要的原始影像資訊大小。 我們提出數個實驗的結果以展示所提演算法的效能。我們主要測試的攻擊模式有線性及非線性濾波器、裁切、雜訊、對比度增強、轉換式影像壓縮等攻擊。此外我們計算各攻擊時對各小波頻帶所造成的錯誤率(error rate)，並比較有無錯誤控制編碼時其各自強韌度的表現，以證明本演算法的強韌度。最後，從受過許多個別不同的攻擊下所得影像所萃取出的浮水印中可看出，無論從主觀還是客觀的標準來衡量，其品質都相當好。結果驗證了本論文所提方法的有效性。
並列摘要
The rapid growth of the Internet brings a new challenge in multimedia data security, say the unauthorized delivery of video, sound, image, etc. To protect the copyright effectively in a ubiquitous network environment, one promising solution to this problem is digital watermarking, where a registered trademark called watermark is hidden in multimedia data. It can be used to settle a copyright dispute by identifying the legal owner of the data. This thesis will focus on the study of watermarking techniques for digital images. We employ the spatial local and global properties of wavelet transform, the human visual system and the error control coding to achieve the imperceptibility and the robustness of a digital watermark. As a result, the goal of protecting multimedia data copyright in the network environment can be achieved. Note that unlike the watermark embedded in an image using traditional methods, we embed a recognizable mark or digital seal called watermark into another image, and use the so called private key watermark produced by applying a one-way operation function on a watermark sequence and the original image to avoid the occurrence of a multiple ownership problem and substantially reduce the size of original image information during the watermark extraction process. Several experimental results are given to demonstrate the performance of the proposed scheme. The watermark attacks used in the experiments include linear and nonlinear filtering, cropping, noise addition, contrast enhancement, and transform-based image compression. In addition, we calculate the error rate for each wavelet subband after various attacks and evaluate the enhancement of robustness performance when the error control coding is applied. The experimential results show that the proposed scheme is robust in the sense that the watermarks extracted from the images under many different and separate attacks have good quality evaluated in both subjective and objective criteria. The results demonstrate the effectiveness of method proposed in this thesis.