Nearly Reversible Data Hiding for Encrypted Audios by Modified Histogram Shifting

這篇論文針對加密的音訊提出一個幾乎可逆式且高容量的資訊隱藏方法。有鑑於傳統方法: 從經過加密的音訊中挪出空間來存放額外的資訊，可能對還原後訊號的品質造成一定程度的影響，由於原始音訊具有連續的特性，我們以一半數量的取樣去預測另外一半的取樣，進而得到預測誤差，再從這些誤差中騰出空間，以利於音訊加密後將資訊藏入。內容擁有者需採取一套特定的加密技術去加密這些誤差，而其餘一半的取樣則可以使用任一種加密機制，如: AES 或位元exclusive OR 運算子。資訊隱藏者使用資訊隱藏金鑰加密資訊，並將其藏入經過加密的音訊中。在接收端，擁有資訊隱藏金鑰者可以正確無誤地取出藏入的資訊；並且，本方法支援兩種機制: 先解密音訊再取出藏入資訊或是先取出藏入資訊再解密音訊，使用者可以根據不同的應用採取不同的順序；最後，實驗結果指出我們提出的方法可以讓藏有大量資訊的音訊達到較好的品質，並且還原回來的音訊幾乎能被完全還原，其分段式訊號雜訊比的值可以超過或接近60 分貝。A nearly reversible and high capacity data hiding mechanism on encrypted audios is proposed in this thesis. In the first phase, instead of vacating room for embedding additional data from the encrypted audios, which may lead to inferior qualities of recovered signals, we exploit the continuity of original audio to estimate half of the samples and embed bits into their estimating errors. A content owner can encrypt the estimating errors with a specific encryption scheme. Note that any one of the existing encryption schemes, such as AES or bitwise exclusive OR, is allowed to be applied to the rest half of samples. With the data hiding key, a data hider can embed data in the encryption version of the audio. At the receiver side, those who have no knowledge about the encryption key are able to extract concealing bits without errors. Moreover, the order of signal decryption and data extraction can be exchanged depending on the requirements of applications. Furthermore, experimental results demonstrate that the proposed work achieved a superior imperceptibility of marked audios with large amount of embedding data and, more importantly, the recovered audios are also nearly totally reversible with segSNR higher than or equal to 60 dB