Data Hiding Algorithms for Vector Maps

• Main Authors: Kai-Wei Chen, 陳凱威
• Other Authors: 王宗銘
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/10776035661119794016

碩士 === 國立中興大學 === 資訊科學系所 === 95 === With the development of the internet and the spread and advancement of computers, people can conveniently send digital information through the internet. Therefore, how to protect the information being sent and sustain information security has always been an important topic. As opposed to traditional encoded protection, data hiding provides another mechanism for insuring information security. Data hiding uses a camouflage concept to embed secret messages into less noticeable cover media to confuse the extractor’s attention and transmit messages through sending cover media. Currently there is much research regarding data hiding, focusing on Image, audio, video and 3D models. Vector Maps have become a recently very popular form of graph. For instance, web-based maps, Global Position System and Car Navigations are all examples of its use. Due to its general application and the fact that usual vector map files are small and convenient to transmit over the internet, we believe that vector maps can be the cover media for delivering secret message. In this thesis, we focus on using vector maps as cover media and propose data hiding algorithms. In chapters 3 and 4 respectively, we propose two data hiding algorithms that individually belong to Spatial Domain and Transform Domain. In chapter 3 of the thesis, we first propose a reversible data hiding algorithm in the spatial domain. We first employ principle component analysis on the vertices of the vector map to generate two principal axes. Then, we use the two principal axes to sort all vertices. Based on the resulting two sorting lists, we can later use them to embed messages. During embedding, we next use layering concepts to divide the sorting lists into the odd or even list. We then modulate the vertex coordinates and individually embed the messages into the odd or even lists. Meanwhile, we can also record the modulation information among them. During extracting, we deduct the modulation that the vertex has created through embedding, and recover the cover vector map. The experiment results prove: our algorithm can embed 2(n-2) bits of messages into vector maps containing with n vertices, the capacity being almost twice the amount of vertex numbers. Furthermore, during embedding, our methods maintain good mapping exteriors, and the RMSE ratio is always under 0.02%. Moreover, our algorithm belongs to the blind detection technique and doesn’t need the original cover map for message extraction. Due to the fact that vector maps vertex coordinates are represented as the floating point, the recovered and the original maps differ in very small areas. The RMSE ratio shows a range between 0.0000001~0.000000001%, indicating a very successful in model recovery. Our algorithm is robust against the affine transformation such as rotation, translation, uniform scaling and vertex reordering. With regard to security, our method uses principle component analysis on the vertices to produce sorting lists for all vertex coordinates before using the secret key to decide the order of embedding. This mechanism deters illegal extractors from using brute-forced methods to extract hidden messages. In chapter 4, we propose a data hiding algorithm based on the transform domain. We first use the Delaunay triangulation to connect the vertices of the vector map into a triangle mesh. Next, we use the contagious method to decide the order of the messages to be embedded. Accordingly, we transform each vertex into a corresponding vertex vector, and embed secret messages onto its length and angle of vertex vector with respect to the X-axis. Finally, we use the inverse vertex vector transformation to transform the vertex vectors that is embedded with message into stego vector maps. Experiment results show that for a vector map with n vertices, our algorithm can embed 2(n-1) bits of messages, the capacity being nearly twice that of vertex numbers. In addition, the RMSE ratio is small, ranging between 0.0001~0.000001% . This RMSE ratio is not likely to be visually detected by the human eyes. Our method doesn’t require the original map for message extraction so long as little amounts of data and the secret key are available. Our algorithm is also robust against affine transformation. Regarding to security, our method uses contagious method to determine the order of embedding, preventing from from brute forced methods for message extraction. In conclusion, we propose two data hiding algorithms for vector maps. They belong to the spatial domain and transform domain. Experiment results show that these two algorithms have high embedding capacity yet with insignificant distortion. In addition, they are with blind detection, requiring no original map for secret message extraction. Furthermore, two methods are robust against affine transformation including rotation, translation, uniform scaling and vertex reordering. We believe that our algorithms provide consolidate contribution for data hiding on vector maps.