| 摘要: | 真正資訊隱藏技術是將機密資訊從負載影像嵌入後，在將影像傳送給對方後，在由對方利用相同演算法將機密訊息取出。真正可逆資訊隱藏是可以完完全全取出機密訊息並且讓影像恢復原始樣貌。對於一些敏感的影像的應用，例如：醫學影像、軍事影像或是衛星影像等方面是很重要影像。因為這類的敏感影像是不容許出現些微的失真，這種情況造成應用上誤判。本論文將設計具影像可逆式資訊隱藏技術在這類的敏感影像。故而發展出真正可逆式資訊隱藏技術，它目的是將機密取出後可以徹底地還原出原始影像的容貌，可以避免發生任何因藏入後的影像失真並且不讓秘密訊息的任何型態被第三者所偵測到。
在此篇論文中，我們將提出四種資訊隱藏技術植基於直方圖處理機制。在我們提出的方法中不但能夠處理自然影像外還能夠處理非自然影像。在先前處理上我們對影像分析以及溢位處理。在影像分析方面能夠分辨影像是否為自然影像或是非自然影像。在溢位處理能夠在嵌入機密訊息後能夠避免像素溢位。接下來我們將介紹本論文中提出各式各樣的預測方式並且這一些預測方法能夠增加嵌入量以外還可以提昇影像品質。就像是在自然影像中的邊吻合方式預測則是使用被預測像素的四周來做預測或是九宮格預測方式則是利用被預測像素四周像素分成兩群、四群、八群等。接下來，使用一個權重的策略來做預測，利用影像漸層關係，像素的左右兩邊像素值差異為最小之關係；則對相同列像素加一個兩倍或是三倍之權重。使用一個類似動態規劃預測方法，在每次預測中找出影像品質較高的預測方法並將它當成下一次的預測。在預測策略之後將產生一張差值影像並且利用直方圖方法找出差值影像中高點與零點。最後產生一張偽裝影響。我們所提出方法能夠保證在第一次嵌入機密訊息中讓PSNR保持48dB並且在不會影響人類視覺品質PSNR大約30dB並且在多次嵌入會比其他方法會有較高的嵌入量與影像品質。;The real reversible information concealment technique embeds secret information from the image carrier. After the information is embedded in the image carrier, the image is transferred from the Internet. Secret information is extracted with the same algorithm that is used to recover the stego image, which hides the information image. This concealment technique can completely extract the image and return it to its original appearance. Image quality is very important for sensitive imaging applications, such as medical, military, and satellite imaging. These sensitive images must not appear even slightly distorted; the recovered images can be misinterpreted if any distortion is present. In this dissertation, we propose a reversible information concealment technique for sensitive images. Using the proposed method, real reversible information concealment techniques can avoid image distortion while being undetected by application software.
In this dissertation, we first discuss four types of information concealment based on a histogram mechanism. In our proposed scheme, both natural and non-natural images can be processed. In the pre-processing phase, images are processed by analysis and overflow (or underflow). In the analysis of these images, natural and non-natural images can be distinguished. The overflow/underflow process can prevent overflowing/underflowing of pixels after the secret information is embedded. Furthermore, various prediction strategies can be used to increase the image capacity and quality. For example, a side-match scheme employs the surrounding pixels or a 3 × 3 block to predict the values of the pixels that are divided into two, four, eight, or more classes in the natural images. From that point, a weight strategy uses the predicted pixel values of each of the proximate sides and doubles or triples these pixels for the respective sides. A threshold strategy is then used to prevent image pixel underflow/overflow. A similar dynamic programming approach is then used after each prediction to select the highest image quality for the next predicted image. Then, a difference image is generated to find the peak and zero points using the histogram method. The selected pixels are situated between the peak and zero points by shifting and embedding. Accordingly, a stego image is obtained. The peak signal-to-noise ratio (PSNR) can be maintained at approximately 48 dB, while human perception is not affected under image visual quality of approximately 30 dB. Based on our experimental results, the capacity of the proposed method is higher than those of existing methods. |
