影像隱藏、驗證與修復之研究

Abstract

本論文提出了數種在網路或儲存系統上保護重要或隱私的影像之方法。在保護機密影像或資料方面，我們提出兩個資料隱藏的方法來隱匿隱藏之資料。在傳輸影像前，人們常將影像先進行壓縮以減少接收端的等待時間。有鑑於此，我們所提出的第一種方法是利用搜尋順序編碼法，將機密資料藏於利用向量量化壓縮法所產生的索引檔中，不僅不會對索引檔產生任何的失真，而且接收端在解壓縮的同時亦可以獲得所藏匿的機密資料。為了要隱匿更大的機密影像，第二種方法則是利用改良式搜尋順序編碼法(一種利用周圍像素值常有高度相似性的關係的編碼法)對機密影像進行編碼，並提供一個可調整的門檻值來控制抽取出的影像品質。在嵌入步驟中，利用周圍像素的變異數來決定掩護影像上每個像素的隱藏量，並搭配模運算將上述的MSOC碼嵌入。從實驗結果得知，所產生的偽裝影像與抽取出的影像的視覺品質，與一些已發表的方法相比具有競爭性。 在保護重要影像方面，我們設計兩種具有不同修復能力的影像驗證系統。第一種是應用於保護單張重要影像。我們利用某些相關於影像區塊的性質產生驗證資料，並嵌入影像區塊中，可用來檢測影像的完整性。同時，我們利用向量量化編碼產生關於影像的修復資料，並搭配(r, n)門檻式分享方法，分散地藏到影像本身中。使得影像本身除了能偵測是否遭到惡意的竄改之外，還具有自我修復被破壞區域的能力。第二種是應用於多張影像傳輸或分散式儲存時，可能會有因網路或儲存系統不穩定而導致某些影像遺失的情形。為了解決這個問題，我們提出了一個具有交互修復能力的影像驗證系統，亦即設計一套雙層式分享機制來保存所有影像的修復資訊，使得影像本身除了原有的驗證功能之外，還能藉由其它倖存的影像互相合作，來修復那些遺失影像。

In this dissertation, we propose several techniques to protect important data and private images in a transmission or storage system. For a confidential image or secret data, two hiding methods are proposed to conceal the existence of the hidden data in the cover images. Because people often compress digital images to reduce the waiting time of the receiver before transmitting the images, the first method uses the search-order coding to embed secret data in the index file of the vector quantization compression result. The proposed method causes no distortion to the VQ version of the image, and the receiver end can obtain both the hidden data and the VQ image. To embed a bigger confidential image, the MSOC scheme in the second method utilizes the feature of high correlation among adjacent pixels (i.e. neighboring pixels are often with similar gray-values) to encode the important image. An adjustable threshold T is used in the MSOC; and this T directly controls the quality of the extracted image. In the embedding part, we use a variance-based criterion to estimate the hiding capacity of each pixel in the cover image. Then the MSOC code is embedded in the cover image using two sets of modulus function. Experimental results show that the quality of both the stego-images and extracted important images are competitive to those obtained in many existing steganography methods reported recently. To protect important images in public environment, we develop two image authentication methods along with different recovery abilities. The first method is used to protect a single image. In the method, the authentication data for each block is generated using some related information within the block, and then embedding it into the block to serve as the attestation for the integrity of the image. Meanwhile, the recovery data obtained by vector quantization technique are shared by using an (r, n)-threshold sharing method, and then scattered all over the image. The proposed method can not only detect whether malicious manipulations have occurred, but also self-recover the tampered parts. In the transmission of multiple images, it is possible that the network connection is unstable; and hence, some images at the receiver end are lost. To solve this problem, we develop an image authentication method with cross-recovery ability to protect a group of images. In the method, a two-layer sharing scheme is designed to preserve the recovery data of all images. The proposed method can not only verify the integrity of each member of the image group, but also reconstruct those lost images by the mutual support of the surviving members.