分享還原法─多媒體檔案保護與應用

Abstract

本論文研究分享還原法在多媒體檔案保護的應用。一般通用型演算法即使同時適用於多種多媒體格式，表現仍遠遜於針對特定格式做處理的專用演算法。因此我們檢視下列三種多媒體格式特性，分別是Portable Document Format (PDF)、三維模型、與網頁影像。首先，PDF格式已是主流且常見的傳遞文件方式，如何安全地保護PDF檔案是一個重要課題。我們提出的 (t, n) PDF分享機制，能將機密PDF檔案拆解成n個偽裝PDF檔案，每個偽裝PDF檔案外觀上不會出現絲毫異樣。一旦使用者收集到t份偽裝PDF檔案，即可將機密PDF檔案完整無失真還原，但如果收集到的偽裝PDF檔案小於t份，使用者只能得到雜訊。所提保護架構能容忍遺失並避免惡意人士取得機密。而在三維模型的應用與保護裡，我們提出了交叉還原機制，允許被保護的三維模型以群組為單位做保護。考慮過往研究，研究重點常放在探討獨立存在的三維模型，然而這意味著當某個模型完全遺失，該模型即無法還原。我們提出的三維模型交叉還原分享法，讓自身的還原資訊交由其它同組的三維模型交叉攜帶。一旦組內有部分模型遺失或遭受竄改，能藉由同組其他模型的互相支援，復原受損資料，達到交叉還原目的。除了針對安全保護，此架構亦能應用在慢速網路環境裡，提供加速觀看到粗略模型外觀的能力。最後，我們提出了一個新的應用領域Web-VC。將傳統視覺密碼技術實現於網頁上。在解密時，使用者能利用CSS色彩混合模式(blend mode)作解密動作，此行為無需撰寫任何程式碼，其解密複雜度與傳統視覺密碼相當，僅需直接將分存擺放在相同網頁位置並配上所需的CSS模式，即能看到所藏機密影像。我們使用CSS色彩模式模擬OR與XOR兩種運算子，並發現獨有的Web-VC運算子，以此為基礎設計出翻轉型與友善使用者之分存的視覺密碼。Web-VC具有良好抵抗壓縮能力，即使在有失真壓縮情況，解密時仍能使機密影像能被肉眼觀看。

This thesis is on the sharing and recovery for multimedia protection and application. We consider three kinds of media formats: Portable Document Format (PDF), three-dimensional (3D) models, and web images. PDF is a popular file format to store documents; the protection of PDF files is thus important. We present a (t, n) PDF-sharing scheme which shares the given secret PDF file among n steganographic (stego) PDF files. Each stego-PDF file looks completely identical to its original PDF version. When people collect sufficient number t of stego-PDF files among the n created stego-PDF files, the secret PDF file can be revealed error-freely, while less than t stego-PDF files gets nothing but noise. Therefore, our protection not only tolerates missing, but also resists thief’s peeping. As for 3D models, we proposed a cross-recovery scheme to protect a group of 3D models. Let us consider how to transmit 3D models in an unreliable environment or channel. Most of the reported studies usually pay attention to a single model. This implies that there is no way to recover if the single model is lost. Hence we consider the joint support of a group of n models. The lost or damaged models can be reconstructed using the mutual support of the survived authenticated models, as long as the number of lost or damaged models is not larger than a given value n−t. We call this benefit as “cross-recovery” ability. The proposed scheme can also be utilized to accelerate the viewing of a group of models in a transmission mission. Finally, we propose a new Web-VC scheme. The conventional encryption techniques of visual cryptography (VC) are extended to webpage. The key point is that CSS blend modes can simulate OR and XOR operators. As a result, the secrets can be revealed in a web browser by stacking shares together without any additional computation. The decoding complexity of Web-VC is as simple as VC. Moreover, Web-VC has strong ability of resisting compression.