Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/19159
標題: 有效的光譜分析三維模型資訊隱藏演算法之研究
A Study of Effective Information Hiding Algorithms for 3D Polygonal Models
作者: 何宇庭 
He, Yu-Ting 
關鍵字: 光譜分析;Spectral Analysis;強韌性;小波轉換;基爾霍夫矩陣;特徵值分解;漢明碼編碼法;反向小波轉換;資訊隱藏;Robustness;Wavelet Transformation;Kirchhoff Matrix;Eigenvalue Decomposition;Hamming Code Encoding Method;Inverse Wavelet Transformation;Information Hiding
出版社: 資訊科學研究所
摘要: 
摘 要
隨著網際網路的普及與發展,使得人們容易取得或傳送數位化之資訊,但卻也產生了智慧財產權與資訊安全(Information Security)等問題。由於目前三維模型被廣泛地使用與普及,因此藉由浮水印(Watermark)來保護三維模型創作者的智慧財產權,或利用其祕密傳送訊息(Message)來維護資訊安全,是一個相當重要以及值得探討的研究領域。
Ohbuchi學者以光譜分析(Spectral Analysis)對三維模型做浮水印處理獲致不錯的成效。然而,他的演算法需花費大量的計算時間,且其強韌性(Robustness)也有待加強。本篇論文據此提出一個有效的光譜分析三維模型浮水印演算法,來降低計算時間及增高浮水印的強韌性。我們的演算法以四個步驟嵌入(Embed)浮水印。第一、我們利用小波轉換(Wavelet Transformation)將原始模型轉換成低複雜度模型,藉此降低計算維度並分離出模型的高、低頻資訊。第二、我們求得模型之基爾霍夫矩陣(Kirchhoff Matrix),計算特徵值分解(Eigenvalue Decomposition)並求出模型之光譜係數值(Spectral Coefficients)。第三、我們將浮水印先以漢明碼編碼法(Hamming Code Encoding Method)予以編碼,並與光譜係數做加權處理後,即可將浮水印嵌入於模型之光譜係數值內。第四、我們以嵌入浮水印的光譜係數,配合反向小波轉換(Inverse Wavelet Transformation),即可構建出具有浮水印的三維複雜模型。浮水印的擷取(Extract)則依照嵌入處理的逆向步驟處理。實驗結果顯示:我們的演算法可節省之時間為28秒~2163秒,效率31%~94%優於Ohbuchi學者的演算法。此外,我們的演算法可抵抗隨機雜訊(Random Noise)、相似轉換(Similarity Transformation)、網面平滑化(Mesh Smoothing)以及網面切割(Mesh Resection)攻擊。我們演算法的強韌性優於Ohbuchi學者之演算法,可降低14%~52%之位元錯誤率。整體而言,我們的演算法在計算效率與強韌性上都優於Ohbuchi學者所提出的演算法。
另外,我們延伸Ohbuchi學者所提出的演算法,發展出光譜分析之三維模型資料隱藏演算法。首先,我們依序求得模型之基爾霍夫矩陣、特徵向量(Eigenvectors)以及光譜係數值。接著,我們將光譜係數值依照其距離原點之大小分區。最後依照各區間之嵌入方式,便可將訊息嵌入於光譜係數值內。而訊息的擷取,則依照逆向嵌入步驟處理之。實驗數據顯示,我們的演算法所嵌入的訊息量能達三倍模型頂點數之多,且經過旋轉(Rotation)、平移(Translation)以及縮放(Scaling)攻擊後,所擷取出的訊息能完整無誤的呈現出來。另外,掩護模型(Cover Model)與偽裝模型(Stego Model)間之Laplacian Difference平均值介於0.006943~7.379368以及兩個模型間之Geometric Laplacian平均值介於0.001513~4.052466,再加上人類視覺的判斷,其結果顯示我們的演算法依然能維持住掩護模型的外觀(Appearance),使偽裝模型與掩護模型之間的差異不大。
總結本篇論文,我們提出一個有效的光譜分析三維模型浮水印演算法,藉由小波轉換以及漢明碼編碼法可以有效降低模型複雜度與提高強韌性。實驗結果顯示,我們的演算法在時間效率以及強韌性都優於Ohbuchi學者之演算法。另外,我們更首創提出頻率域之光譜分析三維模型資料隱藏演算法,所嵌入的訊息量可達到三倍模型頂點數之多。經由量化以及視覺判斷,我們的演算法對於嵌入前與嵌入後之模型外觀差異不大。我們認為本文所提出的兩個演算法,提升了光譜分析在資訊隱藏(Information Hiding)之應用範疇。

Abstract
Along with the popularity and development of the Internet, people can easily obtain or transmit digital information. This, in return, raises to the problems of the intellectual property rights and information security. Since three-dimensional models are more popular at present, it is important either to employ the watermarking techniques to protect the intellectual properties, or to secretly deliver messages by means of 3D model to maintain the information security.
Ohbuchi et al. proposed a watermarking algorithm for 3D polygonal models using the spectral analysis. However, their algorithm has two drawbacks: long computing time for the eigenvalue decomposition and unsatisfactory watermark robustness. In this thesis, we present an efficient spectral-based watermarking algorithm for 3D polygonal models. Our algorithm consists of four steps for watermark embedding. In the fist step, we transform the complex model into the simplified one using a wavelet transformation. This results in decreasing the degree of eigenvalue decomposition as well as separating the high and low frequency information in the cover model. Second, we proceed on spectral analysis, producing their spectral coefficients. In the third step, we first encode the watermarks using the Hamming code before we embed them by modifying the magnitude of the weighted spectral coefficients. In the final step, we apply the inverse wavelet transformation with the spectral coefficients to reconstruct the stego complex model. Experimental results show that, compared to Ohbuchi et al.'s method, our algorithm requires less computing time (around 28~2163 seconds). It achieves 31%~94% more efficiency than Ohbuchi et al.'s algorithm. Our algorithm can resist against common attacks, including the noise addition, similarity transformation, mesh smoothing and mesh resection. In addition, our algorithm has less error bit rates (14%~52%) during watermark extraction, achieving higher watermark robustness. In conclusion, our algorithm is more computationally efficient and has higher watermark robustness than our counterpart.
Moreover, we extend the algorithm we proposed for watermarking and develop a novel spectral-based data hiding algorithm for 3D polygonal models. First, we proceed on spectral analysis, producing their spectral coefficients. Second, we categorize spectral coefficients into four ranges, according to the distance with respect to the origin of the coordinates. Finally, we embed messages by modifying the spectral coefficients according to the embedding method at each range. Experimental results show that, the message capacity of our algorithm can be three times the numbers of the vertexes in the 3D model. After some common attacks, such as model rotation, translation, and scaling, we can extract identical messages from the stego model. Besides, the mean of Laplacian difference between the cover model and the stego model is between 0.006943 and 7.379368, while the mean of geometric Laplacian is between 0.001513 and 4.052466. In addition, inspecting from the human vision, the stego model shows almost indistinguishable difference in comparison with the appearance of the over model. This achieves the goal of the imperceptibility for the proposed information hiding algorithm.
In conclusion, we proposed an efficient spectral-based watermarking algorithm for 3D polygonal models. Experimental results show that our algorithm performs much efficient and robust than Ohbuchi et al.'s algorithm. Besides, we present a novel spectral-based data hiding algorithm for 3D polygonal models. The capacity of this algorithm can be three times the numbers of the vertexes in the 3D model with indistinguishable model distortion. These two algorithms make feasible the applications of the spectral analysis in the information hiding technology.
URI: http://hdl.handle.net/11455/19159
Appears in Collections:資訊科學與工程學系所

Show full item record
 
TAIR Related Article

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.