_{小波及離散餘弦域上使用子區塊分類之}

_{金匙數位影像浮水印系統}

摘要

在本文中我們提出一套不僅具有良好影像品質且經影像攻擊後仍具有高強韌性的數位浮水印技術，另外在萃取端並無對原始影像之需求。這套方法有三個重要的特性來提升整體效果並可同時使用在小波轉換域及數位餘弦轉換域中。

第一個特點為依據數位浮水印與原始影像特性產生一組私密鑰匙，在萃取端中利用轉換後係數與私密鑰匙將數位浮水印完整的萃取出來，因而並無傳統做法中對原始影像之需求。

第二個特點為使用子區塊重組以及三態互斥或(TXOR)操作來過濾不適合嵌入浮水印的子區塊，並將數位浮水印嵌入較不容易改變極性之子區塊以確保數位浮水印的強韌性。

第三個特點為係數禁區淨空，依據子區塊極性及人眼視覺特性兩極化地調整係數值大小，確保影像品質及降低極性變化機率以增加數位浮水印之強韌性。

整合以上三個特點及經程式驗證後可證明我們所提的此套數位浮水印技術在經影像壓縮，如JPEG、SPIHT、JPEG2000，以及裁切後所萃取出的數位浮水印具有極高的強韌性。

　

A Key-based image Watermarking system using

subblock composition in DCT domain and DWT domain

Abstract

In this thesis, we propose a novel watermarking system that is highly robust against attacks without perceptible image degradation. In addition, the original image is not needed in watermark extraction. It operates either in the DWT domain or the DCT domain. This system integrates three important ideas to provide extellent watermarking performance.

The first idea is to use a secret key that is generated based on the watermark logo and the host image. Based on the polarity of the transform coefficients and the key, the original watermark can accurately be extracted.

The second idea is to use subblock composition to filter out frequency transform blocks that are not suitable for watermark embedding. A Tri-state Exclusive-Or (TXOR) operator is then applied in qualified subblock to embed the watermark. This technique further improves the robustness.

The third idea is the deadzone shifting with human visual system. For watermark embedded transform blocks, if the coefficients are near the threshold, they will be shifted toward two ends so that the embedded watermark is not noticeable but more robust.

With the integration of the above three state-of-art techniques, the resulting watermarking system has excellent performance. Simulation results confirm that the proposed system is extremely robust against cropping, JPEG, JPEG2000, and SPIHT compression.