- 學位論文
基於最小擴張樹著色之影像資訊隱藏之研究
The Research of Image Data Hiding Based on Coloring in Minimum Spanning Tree
摘要
影像資訊隱藏主要透過附屬秘密通道來傳輸隱藏資訊。兩個互相衝突的特性：影像品質和資訊隱藏量是在影像資訊隱藏領域中主要的考量因素。在本篇研究論文中，我們利用vector quantization(VQ)來實作影像資訊隱藏。在過去的文獻資料中，用來資訊隱藏的編碼書(codebook)在分割的時候並未將分割的編碼書差異最小化。所以，當編碼書之間的差異越大，為了藏入資訊的代價越高。 藉由最小著色擴張樹，我們提出兩個方法以達到較高的影像品質和較多的資訊隱藏量。藉此，我們可以讓分割的編碼書之間的差異最小化。另外，我們也結合discrete cosine transform(DCT)來實作資訊隱藏。為了要降低因為資訊隱藏造成的失真，我們只在頻率較高的DCT係數區域藏入資訊。藉由結合JPEG和VQ兩方法的優點，相對於傳統的VQ方法，我們可以得到較好的影像品質和較大的資訊隱藏量。透過實驗結果得知，我們提出的方法明顯優於之前所提出的資訊隱藏方法。在研究討論中，也藉由數學證明來印證我們提出的方法具有較佳的影像品質。
並列摘要
Image data hiding is mainly served as transmitting secret information via side channel. The two conflicting properties: image quality and data capacity are the two major concerns to image data hiding. We study image data hiding using vector quantization (VQ) in this thesis. In previous literatures, the difference of separated codebooks are not minimized in which will degrade the image quality. Two approaches have been proposed to achieve a high image quality and data capacity by coloring in a minimum spanning tree. This will minimize the difference between separated codebooks. These approaches are also implemented on frequency domain. In order to decrease distortion when embed secret data in images, only the high-band coefficients will be modified. The combined DCT-VQ gets much better image quality and more data capacity than traditional VQ. One of our VQ schemes is shown to be the best theoretic heuristic solution. Our results have significant advantages over the others with experiments support.