- 學位論文
量子類神經元網路於視覺密碼學之應用研究
Visual Cryptography Using Q'tron Neural Networks
摘要
視覺密碼學是一種達成視覺秘密分享的加解密方法，這個方法不需要任何計算便可以解碼得到被隱藏的影像。例如：給定某一目標影像，經加密後形成n張印製於透明片(transparencies)上的子圖(shares)，參與分享的n個使用者各會得到一張影像，各透明片所顯示的是一些無法辨識或與主題無關影像，唯有重疊了至少k張的原始子圖，才能以肉眼辨識出目標影像的內容，這樣的視覺加解密法則稱為(n, k)法則。視覺密碼學最原始的概念在1995年由Naor和Shamir提出，它存在了若干缺點。使用我們所提出的類神經網路，這些缺點都可以解決，以下摘要式列出本方法的特性：(1) 灰階影像的分享，並不侷限在黑白影像的處理 (2) 加密時不需依賴編碼冊來編成不同加解密結構所要求的子圖 (3) 編碼後的子圖和原始目標影像一樣大。 我們提出了一個利用量子類神經網路模型來分享灰階影像的新方法，這是一個以能量驅動執行(energy-driven)的類神經網路，量子類神經網路模型最大的特點在於其特殊的“恆久性雜訊注入機制”，使得解品質可以控制在理想的範圍內，這樣的機制可以解決網路停滯於局部最低能量狀態(local minimum)。給定一個加解密結構，我們依據影像半色調轉換規則(image-halftoning rule)和子圖重疊規則(share-stacking rule)，將問題轉換成為量子類神經網路模型的能量函式，進而建造出一對應於該加解密結構的量子類神經網路。編碼時，輸入量子類神經網路的是灰階目標影像，待網路穩定時，我們可從對應的類神經元得到與目標影像同樣大小的各個子圖。此方法適用於各種複雜的加解密結構(access structure)，我們將在本論文中描述如何利用Q’tron類神經網路完成視覺密碼學的實例應用，包括資訊隱藏(message concealment)、視覺授權(visual authorization)及半公開加密法 (semipublic encryption)，實驗結果將在論文裡呈現。
並列摘要
Visual cryptography is a cryptographic scheme to achieve secret sharing. For example, it decomposes a secret image into n shares which are distributed to the participants, such that only qualified subsets of participants can "visually" recover the secret image. The "visual" recovery consists of xeroxing the shares onto transparencies, and then stacking them. The secret image will reveal without any cryptographic computation. Originally, the cryptographic paradigm introduced by Naor and Shamir has some drawbacks. This dissertation proposes a novel technique using neural networks (NNs) to fulfill visual cryptography schemes with some extended capabilities: i) the access schemes are described using a set of graytone images, and ii) the codebooks to fulfill them are not required; and iii) the size of share images is the same as the size of target images. The neural network model to conduct this research is called quantum neural-network (Q'tron NN; for short) model. It is an energy-driven NN model. A Q'tron NN is able to achieve local-minima free if it is constructed as a known-energy system and noise-injected, to be detailed in the dissertation. To fulfill an access scheme of visual cryptography, two energy sub-terms, which describe the image-halftoning rule and share-stacking rule, are considered to build the Q'tron NN. The proposed Q'tron NN structures are quite general and, hence, can be applied to fulfill any access schemes of visual cryptography. Some applications of visual cryptography based on the Q'tron NN approach are also discussed, including message concealment, visual authorization, and semipublic encryption.