利用反覆解碼回授技術改進展頻影像藏密法

Abstract

資訊隱藏是一門綜合科學其融合了密碼學、通訊原理、數位影像處理等學科，按其應用可概分為數位浮水印技術及藏密學兩大類，此兩者主要藉由載體嵌入訊息達成不同的目的，數位浮水印技術的目的係智財權保護主要強調強固性，藏密學目的在於秘密通訊強調不可見性。 然而傳統密碼學已無法完全滿足秘密訊息保護的要求，尤其在嚴密的監控環境下，倘若冒然傳送密文亂碼往往更引人注意，此時，須藉由藏密技術將秘密訊息隱藏在無害（innocuous）的媒體中才不致使人懷疑，傳送者與接收者才能不著痕跡地完成訊息交換。 本文主題係改進由Marvel等人提出的展頻影像藏密法（SSIS）高誤碼率的缺點，我們針對SSIS的藏密方式進行探討，提出合理的質疑並藉由相關的實驗來驗證我們的推論，設法找出SSIS誤碼率高的根源，並針對問題來源提出兩種改善方案。第一種方案利用適應性濾波器將載體影像於藏密前作前置處理以降低載體影像的變異數，第二種方案則利用反覆解碼回授的技術，對系統的通道進行估測，運用估測結果將藏密訊號等化，經實驗證實，我們提出的兩種改善方案可大幅減少SSIS的誤碼率並維持影像品質。同時，我們也探討了空間域藏密技術的強固性，並應用展頻及容錯技術保護秘密訊息以抵抗JPEG壓縮。

Information hiding is a comprehensive science which merges the disciplines of the cryptography, communication principle and digital signal processing etc. It can be categorized into Digital Watermark and Steganography base on its applications. The both scenarios embed data into carrier to achieve different task. The goal of digital watermark is protection of intellectual properties and mainly emphasizes robustness. Another purpose of steganography is covert communication and highlights invisibility. However, traditional cryptographic technique can’t fully satisfy the requirements for secret information protection. If transmit encrypted code may bring more attention especially under strict surveillance. At the mean wile, someone must hide secret messages into innocuous media to avoid suspicions by means of steangraphic techniques, therefore the sender and receiver could accomplish information exchange without even the spirits knowing about it. Our thesis is to improve the higher bit error rate of “Spread Spectrum Image Steganography-SSIS” proposed by L. M. Marvel et al. We discussed the hiding scheme of SSIS and brought up reasonable queries and testified our inferences by related experiments which help us to find the source of higher bit error rate of SSIS. We proposed two improvements that aim at this drawback, first approach preprocess the cover image before being embedded by adaptive wiener filter to decrease the variance of cover image and the second utilized iterative decoding feedback technique to estimate the channel then equalized the modulated signal in accordance with estimating results. After experiments, our techniques can decrease the bit error rate of SSIS significantly and maintain the fidelity of stego image. We also discussed the robustness of steganagraphic technique which was applied in spatial domain and took advantage of spread spectrum and error correcting techniques to protect hidden messages against the JPEG compression.