WATERMARKING GIS DATA FOR DIGITAL MAP COPYRIGHT PROTECTION

Transcription

1 WATERMARKING GIS DATA FOR DIGITAL MAP COPYRIGHT PROTECTION Shen Tao Chinese Academy of Surveying and Mapping, Beijing , China Xu Dehe Institute of resources and environment, North China University of Water Resources and Eectric Power, Zhengzhou , China Li Chengming Chinese Academy of Surveying and Mapping, Beijing , China Sun Jianguo Coege of Computer Science and Technoogy, Harbin Engineering University, Harbin , China Abstract: In recent years, with the rapid deveopment of computer communication and the Internet, it makes easy to dupicate and distribute the GIS data because of its digita forms. So the appication of GIS data faces some issues such as digita copyright protecting, data authenticating, and data source tracing, etc, and these issues coud be the main obstaces restricting the further usage of digita maps. In order to sove these probems, we shoud construct the GIS production copyright protection system. The technique of watermarking provides potentia soutions for copyright protection. Considering the characteristics of digita watermarking for digita maps, the DFT (Discrete Fourier Transform) bind watermarking mode is put forward in this paper. First, the vertex sequence that extracted from vector map is carried on DFT transform. Second, divide the phase of DFT according to the quantization step size. Finay, we introduce a new watermarking scheme to embed the watermark into the phase of DFT through quantification, which provides robustness for watermark attacking. Extensive experiments are conducted to vaidate the avaiabiity, invisibiity and the robustness of our scheme. Keywords: digita map; digita watermarking; DFT; 1 Introduction Digita map is a kind of important and strategic information resource and is widey used in economic and socia appications, it aso cosey reated to nationa defense affairs. The acquisition and maintenance of the data is a high costs and efforts process, there are needed arge amount of manua abor and anaysis in order to acquire origina geospatia data such as and survey, aeria photography, sateite image, etc. Furthermore, it is a hard work for peope obtain the information through digitaization and vectorization from the origina data [1][2]. So as a resut, geospatia data especiay the high precision GIS data normay cannot be used for free. In addition, some high precision data are ony permitted to be used in miitary or government units and extra security requirements are aso needed. But in recent years, with 1

2 the rapid deveopment of computer communication and the Internet, it makes easy to dupicate and distribute the GIS data because of its digita forms. So there is an emergent need for copyright protection and authentication of vector maps. Watermarking is a promising soution to address this issue. In addition to the genera characteristics of digita products, vector data has the foowing features [3] : There is no fixed sequence of data; there is no cear concept of samping rates; data contains not ony the geometric information incudes topoogy information; and vector geospatia data is required high precision, sma redundancy, hierarchica organizations, which are an increase of watermark embedding and extraction more difficut. Vector geospatia data with digita watermark in addition to the digita watermark must be non-perception, and stabiity, deectabiity, but aso to ensure its avaiabiity, the data embedded watermark shoud not affect its avaiabiity. Over the past few years, the research on the digita watermark of raster data is more, but that of vector map in the iterature is very few [4-9], among which the origina vector data is needed in the watermark extraction phase, the detection method is reativey easy to achieve, and the extracted resuts are highy dependent on the quaity of the raw data. In practica appications, raw data can t be obtained easiy on many occasions, so the bind watermarking mode [10-15] which the raw data is not required when the watermark is extracted shoud be paid more attention. In this paper, Considering both the characteristics of digita watermarking for vector maps and the high-accuracy data requirement for vector data, a new DFT-based bind watermarking method for the vector maps is proposed, by transforming the vertex sequence using DFT transformation, dividing the phase of DFT according to the quantization step size, the watermark sequence can be embedded into the phase of DFT, which ensures the avaiabiity, invisibiity and robustness. 2 System mode The main purpose of watermark mode in this paper is to achieve bind detection [16]. And the main idea is that the coordinate axis woud be divided into the A set of intervas and the B set of intervas by the equa-space, interva size is Δ. Here, A set represents 0, B set on behaf of 1. Adjust the coefficient vaue of the watermark to be embedded to causes it to be equa to one recent correspondence set in the intermediate quantity. When the watermark is detected, just to determine the range of the coefficient fas on the A or B set, you can get the corresponding watermark information 0 or 1. Because the transform domain, the moduation of the object data is the range of transform domain or phase, may aso be a rea part or imaginary part. Transform domain coefficient of assumptions, embedded digita watermark for the principe as shown: Figure 1. The phase quantification 2

3 Demarcation interva set: seect the quantization step size wi be divided into axes as shown in Figure1. A and B set interva. Those who beong to set the coordinates of A band represents bit "1", where a range of set B represents the position coordinates of "0". The mathematica operation that seeks the quotient and remainder: seect the quantization step size to quantify the parameters which is phase of DFT coefficients, integer vaues and remainder is operated. Assumptions a for the quotient, b for remainder, there are: 360 a = b= 360 aδ Δ Quantified embedded: The quantification of parameters θ is cosey reated to the watermark bit w i : When w i =1, enabe the quantification resuts θ equivaents to the intermediate coordinates which on a certain range of A set the nearest θ ',When w i =0, enabe the quantification resuts θ equivaents to the intermediate coordinates which on a certain range of B set the nearest θ.the expression are as foows(for k=0,1,2, ): (1) 当 w i =1 时 : θ a = 2k θ ' = θ +Δ a = 2k + 1 (2) 当 w i =0 时 : θ +Δ a = 2k θ ' = θ a = 2k + 1 After quantification ofθ, the contained information of the watermark bit w i is ony determined by the range set in the quantization resuts θ ' :If θ is in A set, θ ' on behaf of the watermark bit information w i =1; the other hand, θ is in the range of B set, θ ' wi represent the digita watermark information w i =0, the above (1) beonging to A range set, (2) beonging to B range set. From where we can see that θ ' θ Δ, the maximum error caused by the quantization operation is quantify step. As can be seen from the above description, the process of embedding the watermark based on the quantification is different from the previous embedding rues of addition or mutipication, it can be seen as appropriate modifications of parameters θ and process which be given the watermark information. In order to reduce interference resuted from the random error in the process of quantify, improve the robustness of embedded watermark, θ wi be quantified to the nearest A or B areas. When the watermark is detected, test data is carried on DFT, the corresponding watermark information w i =0 or w i =1 is get from the ocation of the embedded watermarks, the vaue of θ ' beonging to A or B set: 1 θ ' A wi = 0 θ ' B The extracted watermark sequence is carried an anti-scrambing, if raw watermark is twodimensiona images, one dimensiona sequence is transformed to two dimensiona sequence. 3

4 In this artice, the binary image is get from the mapped text, so the extracted watermark sequence is done with to get the two dimensiona image. 3 The bind watermarking scheme Based on the above system mode, DFT as the transform toos, the watermark is embedded into the phase of DFT coefficients, the phase vaue of DFT coefficients is quantified by Appropriate step size,as we as the watermark is embedded in quantified vaue. The embedded watermark vector geospatia data is obtained after that the embedded phase vaue and the unmodified magnitude carries inverse DFT; when the watermark embedded in the vector data is extracted, first, the embedded vector data is carried on DFT, then the coefficient of DFT is cacuated, the phase is get, the watermark is extracted according to the quantify program. 3.1 The embedded phase The embedded mode of digita watermark of the vector geospatia data is shown in Figure 2. Figure 2. The embedded mode Concrete steps are as foows: Step 1: The origina watermark ength is transformed to meet 2 N (FFT agorithm conditions, a quick agorithm of DFT). In order to increase the security of the watermark, the watermark can be encrypted or scrambed. Step2: Geospatia vector data V is read according to the storage order of coordinates, and then the coordinates of vertex sequence constitute a compex sequence in accordance with formua (1). a k = xk + iyk k [0, N 1] (1) Here N is the Vertices number in V. a is carried on DFT, the DFT coefficients is get by formua (2). Step3: Sequences { } k N / k = 0 ( πin) k A = a e [0, N 1] (2) N k The DFT coefficients of a sequence contain the ampitude A and phaseθ. Step4: The phase of the coefficient is quantified by the seected appropriate step size, the quantified vaue equas to the median of the interva in accordance with the "0" or "1". Step5: The embedded magnitude combined with the unmodified magnitude is carried inverse DFT, the embedded watermark vector geospatia data V is obtained. 3.2 The extracted phase The extracted phase of digita watermark is shown in Figure.3. 4

5 Figure 3. The extracted mode Step1: Vertex coordinates in ' a ' k constructed in accordance with formua (1). a is carried on DFT by formua (2), the coefficient of DFT is Step2: The sequence { ' k} V is read, then a compex sequence is { } cacuated. Step3: The phase of the DFT coefficient is quantified to mean vaues of the ocation of embedded watermark. The phase vaue is cacuated to judge in which interva quantified, the watermark is extracted according to the quantization range where the phase vaues is in. Step4: The extracted watermark bit sequence is carried on anti-scrambing, then and the sefcorreation c, in order to enhance the objectivity of the resut of extracted watermark, the sefcorreation method is used, the sef-correation coefficient obtained sc θ is used to hep determine the effect of watermark extraction, if the raw watermark is two-dimensiona images, the extracted watermark is processed to get the fina two-dimension watermark. 4 Experiments The test data contains 11,080 vertex of river ines of 1:50 miion. The watermark data is bit as Figure 4(a), according to FFT, the data is extend to the 4096 (2 12 ), and the quantization step size is 5,10,15 respectivey. ' (a) the origina watermark (b) =5 (c) =10 (d) =15 The quantization step size is variabe parameters when the range is fixed, the different step size are used in experiment to contrast, the used quantization step size are the =5, =10 and =15. The extracted watermark are shown in Figure 4(b), Figure 4(c) and Figure 4(d). Figure 5, Figure 6 and Figure 7 respectivey show the compared resut of the origina vector geospatia data and the embedded data when step size =5, =10, =15, the dotted ine for the embedded watermark data and the soid ine for the raw data. We can find the embedded mode is we invisibiity, and the embedded data does not affect the dispay quaity of 5

6 graphics, compared to three-step size, the invisibiity of graphics dispayed in figure 5 is best, Figure 6 foows, the Figure 7 is the worst. Figure 5. Comparison of before and after watermark embedded when =5 Figure 6. Comparison of before and after watermark embedded when = Avaiabiity Figure 7. Comparison of before and after watermark embedded when =15 The resuts of the a bit error ratio(ber) anaysis corresponding to =5, =10, =15 are shown in figure8. As can be seen from the statistica resuts, the BER increases with the step ength becomes arge, but the argest BER is smaer that 2.5 percent, which vaidates the avaiabiity of the proposed scheme. 6

7 4.2 Robustness Figure 8. BER anaysis corresponding to =5, =10 and =15 The robustness of the watermark is evauated for different types: format conversion, deeting the vertices, rotate, transation, scae in experiment, and find that the watermark can not be extracted effectivey after that the vertices are deeted randomy and the embedded data is scaed, so the foowing anaysis of ony a few attacks on the remaining cases and the resuts are shown in Tab.1, which exhibit the abiity of resisting attacks, and the vaues of NC (sefcorreation coefficient) cacuated by formua 3. ww. ' NC = (3) ww. w'. w' Tabe 1 the robustness experiments attacks Format conversation Transation (units) Rotation (degree) NC ( Δ =5 0 ) NC ( Δ =10 0 ) NC ( Δ =15 0 ) Experimenta resuts anaysis Through the experiment, we coud anayze and find: (1) The step size (that is ) is greater, the difference of the extracted watermark and origina watermark is smaer, so the robustness is better. (2)If the step size seected is too arge, the graphic quaity dispayed and visibiity are affected. If the step size seected is too sma, avaiabiity and visibiity can be ensured, but the effect of watermark extracted is bad. Therefore step size used to quantify is determined in accordance with the robustness of the watermark and the used condition of vector geospatia data. (3)The error in the extracted watermark of integer data embedded vector geospatia data after transation, rotation, format conversion comes mainy from its own error and rounding error when the FFT computation, the effects of the extracted watermark are the same after the format conversion and transation of the embedded vector geospatia data. After the rotation, the watermark extraction error in the data mainy from the data is cut-off. The mode is fragie after deeted vertices, scaded. 5 Concusions In this paper, a bind watermarking scheme based on DFT is proposed. By transforming the vertex sequence using DFT transformation, dividing the phase of DFT according to the quantization step size, the watermark sequence can be embedded into the phase of DFT. The experimenta resuts show that the embedded vector geospatia data has good avaiabiity, invisibiity, the watermark is robust after transation, rotation and format conversion of the 7

8 embedded watermark vector map, the computationa compexity of the mode is ow and easy to impement. It is confirmed that the proposed mode is effective in protecting the copyright of the vector GIS data. Future work wi incude the robustness against deeting vertices, cipping, scaing to the achievement of the copyright protection of the vector map and appying this mode to the other types of vector data. Acknowedgement This work was supported by Research and Deveopment Program of CASM and Funded by Key Laboratory of State Bureau of Surveying and Mapping (No ) References [1] Anbo Li, Bing-xian Lin, Ying Chen, Study on copyright authentication of GIS vector data based on Zero-watermarking. The Internationa Archives of the Photogrammetry, Remote Sensing and Spatia Information Sciences. Vo. XXXVII. Part B4. Beijing 2008.pp [2] Ohbuchi, Robust watermarking of vector digita maps, Proc. of the IEEE Internationa Conference on Mutimedia and Expo., Lausanne, Switzerand, 2002 vo.1, pp [3] XU De-he, ZHU Chang-qing, WANG Qi-sheng. A Survey of the Research on digita watermarking for the Vector Digita Map, Geometrics Word. Editoria Dept. of Geomatics Word, Harbin, China, 2007, pp [4] Voigt M, Busch C. Watermarking 2D-Vector data for geographica information systems. Proceedings of the SPIE, Security and Watermarking of Mutimedia Content, San Jose, USA, 2002, pp [5] Schuz G, Voigt M. A high capacity watermarking system for digita maps. Proceedings of the 2004 Mutimedia and Security Workshop on Mutimedia and Security, Magdeburg, Germany, 2004, pp [6] Wang Xun, Lin Hai, Bao Hu-Jun. A robust watermarking agorithm for vector digita mapping, Journa of computer-aided design & computer graphics, Editoria Dept.of Journa of computer-aided design & computer graphics, Beijing, China, 2004, pp [7] Nikoaidis N, Pitas I, Soachidis V, Fourier descriptors watermarking of vector graphics images, Proceedings of the Internationa Conference of Image Processing, Vancouver, BC, Canada, 2000, pp [8] Soachidis V, Nikoaidis N, Pitas I. Watermarking poygona ines using Fourier descriptors, Proceedings of the IEEE Internationa Conference on Acoustics, Speech and Signa Processing, Istanbu, Turkey, 2000, pp [9] Nikoaidis N, Pitas I, Giannoua A, Watermarking of sets of poygona ines using fusion techniques, Proceedings of the 2002 IEEE Internationa Conference on Mutimedia and Expo, Lausanne,Switzerand, 2002, pp [10] LI Yan-ping, TANG Zhen-min, QIAN Bo, Audio Watermark Agorithm Robust to Desynchronization Based on Speech Parameter Mode, Computer Engineering, Editoria Dept. of Computer Engineering, Shanghai,China, 2008,pp [11] Suivan K, Bi Z, Madhow U, et a, Steganaysis of Quantization Index Moduation Data hiding, IEEE Internationa Conference on Image Processing, Suntec City, Singapore, Oct.2004, pp [12] Benedens, O. and C. Busch, Towards bind detection of robust watermarks in poygona modes, Eurographics 2000 Proceedings, vo.19,.no.3,2000. pp. C199-C208. 8

9 [13] Brian C, Gregory W, Pre processed and post processed quantization index moduation methods for digita watermarking, Processing of SPIE: Security and Watermarking of Mutimedia Contents II, San Jose, USA,2000, pp [14] Chen B, Worne G, Quantization index moduation: A cass of provaby good methods for digita watermarking and information embedding, IEEE Transactions on Information Theory, IEEE Inc, USA, 2001, pp [15] Ronad N, Bracewe. The Fourier Transform and Its Appications. Xi an Jiaotong University Press, Xi an, China, 2005, pp [16] YANG Yi-xian, NIU Xin-xin. Theory and Appications of Digita Watermarking. Higher Education Press, Beijing, China,2006 9