隱藏資料於圖像

Abstract

本論文提出數種在不同應用上的隱藏資料於圖像之方法；這些應用包括重要資料傳輸儲存、影像驗證、及浮水印製作。在應用於重要資料傳輸儲存的方法上，我們所提出的第一種方法是以最不重要位元(LSB)置換為基礎，並採用基因遺傳演算法(Genetic Algorithm)尋找出較佳置換法則，以求取較佳品質的隱藏結果。第二種方法則是將重要資料隱藏到影像的中段位元層中，並配合較佳之置換法則與個別像素調整程序，來減低資料隱藏過程對影像品質的改變。在應用於影像驗證的方法中，我們在影像的每個區塊中嵌入某一特定的區塊性質，用來檢測影像的完整性。同時，我們也將一影像的縮小版本自藏到影像本身中，使它在受到局部破壞時能擁有自我修復的能力。在浮水印的製作方法上，我們所提出的第一種方法是以視覺圖樣(Visual Pattern)為基礎，利用人類視覺對相似影像區塊的不敏感性，經由些微調整部份影像像素值，將浮水印嵌藏到影像中。第二種方法之設計是基於我們對影像中之各個跳動區塊(Hop-block)的像素平均值與整體影像平均值十分相似這個特性之觀察，所設計出不用原圖即可偵測浮水印存在與否的方法。此方法在對抗一些浮水印攻擊如濾波、JPEG壓縮、剪裁、列印後再掃瞄等，都有不錯的效果。

In this dissertation, we proposed several techniques of hiding data in images for different kinds of applications: including important data transmission and storage, image verification, and watermark design. In the application of important data transmission and storage, the first method was based on the least-significant-bit (LSB) replacement. We designed a genetic algorithm in the method to search for a nearly optimal replacement way to get a resulting image of high quality. The second method embedded the important data in the moderate bit of an image. A nearly optimal replacement stage and a pixel adjustment procedure were designed to reduce the distortion to the image. In the application of image verification and recovery, we embedded a preset property in each block of an image to serve as the attestation for the integrity of the image. Together, a shrunk version of the image was self-embedded in the image to enable the image to restore the corrupted part without the help of the backup copy when the image was partially corrupted. Finally, in the design of digital watermarks, the first method was based on the concept of visual patterns. By utilizing the property of the insensitivity of human eyes to similar image blocks, we embedded the watermark by slightly adjust some gray values of an image. The second method was based on the observation to the constancy property of the hop-block means of an image. It is an oblivious method that detects watermark without resorting to the original image. Note that the proposed method was found to be robust against numerous image manipulations such as filtering, JPEG compression, cropping, and printing-and-rescanning.