影像之視覺式分享，權重式分享，修復與高品質資訊隱藏

Abstract

本論文提出數種保護數位影像的安全性與正確性的方法。為了分散影像的機密，視覺密碼學與影像分享是本論文首先討論的兩大主題。首先，我們提出可翻轉的視覺式分享法。兩張機密影像會被編碼為兩張投影片。疊合兩張投影片可以顯示出第一張機密影像，而翻轉第一張投影片並與第二張投影片疊合，則可顯示第二張機密影像。除此之外，我們亦證明本方法在安全條件下能產生最佳對比值。 在第二主題，我們提出一種權重式機密影像分享法。影像的每個分存會被賦予一個權重值，此數值反映該分存的重要性。收到數份分存後，只要分存權重總和不小於設定的門檻值，即可解碼回機密影像。此權重式分享方法是基於多項式除法而得，且傳統的多項式機密影像分享法只是本方法的特例。另外，藉由觀察GF(2k)的特性，我們提出本方法的一套快速編碼演算法。 在第三主題，為了藉由影像來傳輸隱藏的機密，我們提出一種基於權重加總的資訊隱藏方法。比起其他已被提出的方法，本方法的優點為：i) 可調整的隱藏率，且範圍極廣(例如從0.5到4.0位元/像素)。ii) 在各種不同的隱藏濾下，其結果像影像的PSNR值有相當的競爭性。iii) 只要給予隱藏率，其結果影像的PSNR值可不須經由真正執行隱藏過程來獲得，而是光藉由查表即可得預測值。 在第四主題，為了保護影像的正確性，我們提出一種具有回復能力的半脆弱型浮水印。其回復資料會編碼成許多分存，並各自隱藏至影像區塊的DCT域中。當浮水印影像的某些部份被破壞，這些分存會從影像的未破壞區塊中抽出，然後回復資料會從這些抽出的分存中解碼，並用於修復被破壞的區域。實驗顯示在合理程度下，本浮水印法可以抵抗一些能保存內容的影像處理，像是JPEG壓縮，高斯模糊，與亮度調整。

In the dissertation, several technologies to protect digital images are proposed. To diffuse the content of a secret image, Visual Cryptography and Sharing are both studied in the dissertation. First, a Flip Visual Cryptography system is proposed. Two secret images are encoded as two transparencies. Stacking the two transparencies can reveal one secret, and the second secret is revealed by stacking the second transparency with the flipped version of the first transparency. The proposed scheme is proved to have perfect security and conditionally-optimal contrast. In the second topic of the dissertation, a weighted secret image sharing method is proposed. Each shadow has a weight which indicates the shadow’s relative importance. The secret image can be decoded as long as the total sum of received weights reaches a specified threshold. The weighted sharing method is based on polynomial division, and the traditional polynomial-style secret image sharing is a sub-case of ours. Moreover, by observing the characteristics of GF(2k), a fast sharing algorithm is also proposed. In the third topic of the dissertation, in order to have confidential transmission, an image-hiding method based on weighted-sum function is proposed. Compared with previous works, the proposed method has following advantages: i) wide range of embedding rate (such as 0.5 to 4.0 bits per pixel); ii) competitive PSNR values over the whole wide range; iii) once an embedding rate is given, our look-up table can predict the PSNR value, even before the actual embedding. In the fourth topic of the dissertation, in order to protect the accuracy of an image, a semi-fragile watermarking method with recovery ability is proposed. The recovery data are shared among many shadows; then each shadow is embedded in the DCT domain of an image block. When the watermarked image is tampered with in some area, the non-damaged shadows are extracted from image blocks; then the recovery data are decoded from those non-damaged shadows. The recovery data are then used to repair the damaged area. Experiments show that the proposed method can resist some content-preserving operations within certain degrees, such as JPEG compression, Gaussian noise, and brightness adjustment.