GISdevelopment.net --> Application --> Miscellaneous
Lilian S.C. Pun-Cheng
The Hong Kong Polytechnic University
lspun@polyu.edu.hk
Zhilin Li
The Hong Kong Polytechnic University
lszlli@polyu.edu.hk
Abstract
Data copyright and security have become very important issues in geo-information technology. Digital watermarking is one of the means to protect the copyright of the data. Current researches are concentrated on the raster image data. That is, no research on geo-referenced spatial data in vector format has been known to the investigators. However, vector is still the most important data format for geo-information. It is therefore of great importance to investigate the methodology for watermarking geospatial data in vector format. Indeed, the objective of this paper is to review the applicability of existing watermarking techniques, which have been developed in computer science for digital images and CAD products, now to fundamental geospatial data, including digital elevation models (DEM), digital line graphs (DLG), digital raster graphs (DRG) and digital ortho-image models (DOM).
Introduction to Watermarking
Digital Watermarking is a digital signal processing for which the process will add some extra information to the source data. It is a kind of information-hiding techniques. Together with fingerprinting, it is classified as a type of robust copyright marking (Figure 1) (Petitcolas, Anderson & Kuhn, 1999). Its information is a hidden object, difficult to damage, destroy or steal. The main reasons for watermarking are to protect the data copyright, to keep information secret and to set a trap to any attack. The watermarking process can be very simple - define the watermarking signal and embed these signals to the data source by the watermarking algorithm (Figure 2).
Figure 1 Classification of information-hiding techniques
Figure 2, The basic work flow of digital watermarking
Any watermarking algorithm ultimately modifies the source data and result in human-made error. In general, watermarking still images will modify the image pixel values whereas watermarking a digital sound will introduce a noise of the sound. By the same principle, watermarking geospatial data will result in location inaccuracy.
A good watermarking algorithm should balance the three elements of robustness, imperceptibility and capacity. That is, the watermarking signal is hard to be detected and cracked by the attacker, imperceptible for human beings (in both vision and hearing) and the watermarked signal is large. However in general, increasing robustness will lead to a loss in imperceptibility. Conventional researches in watermarking have put much emphasis on analyzing the robustness of different watermarking algorithms against different attack mechanisms. This is performed mostly by correlating the watermarked data with the attacked data. Copying is suspected, that is watermarking exist if the coefficient is greater than the threshold value and vice versa. These techniques will be briefly introduced in the following section. It is noted that these are concerned with digital image (raster) data only.
Digital Watermarking
To hide the watermarking signal, the watermark signal has to be concealed within the original data or data sources (asset). Data concealment is the main process for watermarking. In general, it is a reverse approach for data compression. For data compression technology, its aim is to remove unnecessary information from the data sources all those data which are perceptually less important. For the data hiding technology, the goal is to add some information to the data source so that the user will not be perceptually alerted. To keep the robustness, the output of the watermarked data should be difficult to be detected or easily removed by some filtering and digital processing. If there is a big change of the source data, the end-user will be easily alerted by comparing the non-watermarked data and the watermarked data.
There are many researches on watermarking still image on the screen that follows the Human Visual System (HVS). No matter unnecessary data are removed or extra information is added, the user should not be alerted to have a big change from the original image. Human eye is not able to perceive light radiation equally at all frequencies (Figure 3), but is more sensitive to red than blue color. It is more sensitive to middle frequencies and less to low and high frequencies of the visible spectrum.
Figure 3, Human eye visible spectrum response with spectral distribution of the color
For watermarking still images, the image binary data sequence should undergo a transformation (Discrete Fourier Transform DFT or Discrete Cosine Transform DCT). The spectrum in frequency domain is analyzed. The data in the less perceptible frequency band is modified, but not the high contrast pixels region and very high frequency component where human eyes are sensitive to changes and contrast. Finally, an inverse transformation is made to form the watermarked image. The sane principle may apply to watermarking a sound source, by which we should have good understanding of Human Auditory System (HAS). Similar to human eye, human ear exhibits a different sensibility to sounds depending on the frequency. It is almost insensitive to sound having frequency below 10Hz and above 20KHz.
To add details into original image that are not easily perceived, the fingerprint pattern recognition technology may be useful. Most fingerprint image comparison algorithms are using minutiae. Some important minutiae types available in the human finger are shown in Figure 4 and 5. The fingerprint image is first transformed into binary images through an ad-hoc filtering algorithm. The images obtained are then sampled into small sectors and submitted to a thinning process which allows for the ridge line thickness in order to achieve a one pixel width. Finally a simple image scan allows for locating the pixel that corresponds to minutiae (Figure 6). For the extracted information, position and direction of specific feature points like ridges ending, bifurcation are stored and kept as a signature for further comparison.
Figure 4 Minutiae Types for Human Figure
Figure5 Minutiae Types for Human Figure
Figure 6 Minutiae Extraction
Attack Methods to Watermarking
There are numerous ways that watermarks in digital images may be destroyed electronically, especially for low robustness and easily detectable watermarks. Common examples are JPEG Compression to a lower quality image, adding image with Gaussian Blur, Gaussian Noise or Median Noise, Image Rotation to distort location and orientation, Image Sharpening, image processing by High Pass Filter and Image Chopping. Past researches often compare these resulting images (post attack to watermarked images) with the original or watermarked images to evaluate the effectiveness of different attack methods. There is often a trade off between robustness and attack. In general, watermarking low frequency components of the image is robust to attack, but can easily be changed and disappeared by compression. By contrast, the high frequency components are difficult to be compressed or filtered, but are easily perceived if the attacker is knowledgeable of the original data.
Research Directions to Watermarking Vector Data
Geospatial data are mostly vector graphics, especially for large-scale map information. Although the end output is image on the screen, it comes with a graphic engine by which the vector information such as coordinate location, magnitude and direction can be interpreted. Unlike raster images, the vector data do not have color information. Watermarking can only be in the thoughts of modifying and documenting the coordinates, magnitude and directions. Yet, these are also used the same for attack. Some geo-spatial data owners tend to insert extra but not easily noticeable information (lines, symbols etc.) on an ad-hoc basis to protect the copyright claims. By this approach, increasing robustness will be at the expense of data capacity, else a slight change from data generalization or compression can corrupt the watermarks. Hence, there is a need to investigate an appropriate systematic method to inserting or removing information to the data such as pattern recognition and data encryption. Undoubtedly, this will lead to some errors or distortions in location and shape, but the general pattern can still be preserved. Another approach is to uncover the attack types and extents but necessitates a detailed and time-consuming comparison of datasets. More difficulty will be posed on comparing data of two different formats, e.g. image and vector, analog and digital. For whatever approach, the algorithm on watermarking or on attack detection should balance the issues of robustness, imperceptibility and capacity or complexity.
Acknowledgement
The work described in this paper was supported by a grant from the Research Grant of the Hong Kong Polytechnic University (Project No. A-PG48).
References
Watermarking GEO-Spatial Data - A Review
Lilian S.C. Pun-Cheng
The Hong Kong Polytechnic University
lspun@polyu.edu.hk
Zhilin Li
The Hong Kong Polytechnic University
lszlli@polyu.edu.hk
Abstract
Data copyright and security have become very important issues in geo-information technology. Digital watermarking is one of the means to protect the copyright of the data. Current researches are concentrated on the raster image data. That is, no research on geo-referenced spatial data in vector format has been known to the investigators. However, vector is still the most important data format for geo-information. It is therefore of great importance to investigate the methodology for watermarking geospatial data in vector format. Indeed, the objective of this paper is to review the applicability of existing watermarking techniques, which have been developed in computer science for digital images and CAD products, now to fundamental geospatial data, including digital elevation models (DEM), digital line graphs (DLG), digital raster graphs (DRG) and digital ortho-image models (DOM).
Introduction to Watermarking
Digital Watermarking is a digital signal processing for which the process will add some extra information to the source data. It is a kind of information-hiding techniques. Together with fingerprinting, it is classified as a type of robust copyright marking (Figure 1) (Petitcolas, Anderson & Kuhn, 1999). Its information is a hidden object, difficult to damage, destroy or steal. The main reasons for watermarking are to protect the data copyright, to keep information secret and to set a trap to any attack. The watermarking process can be very simple - define the watermarking signal and embed these signals to the data source by the watermarking algorithm (Figure 2).
Figure 1 Classification of information-hiding techniques
Figure 2, The basic work flow of digital watermarking
Any watermarking algorithm ultimately modifies the source data and result in human-made error. In general, watermarking still images will modify the image pixel values whereas watermarking a digital sound will introduce a noise of the sound. By the same principle, watermarking geospatial data will result in location inaccuracy.
A good watermarking algorithm should balance the three elements of robustness, imperceptibility and capacity. That is, the watermarking signal is hard to be detected and cracked by the attacker, imperceptible for human beings (in both vision and hearing) and the watermarked signal is large. However in general, increasing robustness will lead to a loss in imperceptibility. Conventional researches in watermarking have put much emphasis on analyzing the robustness of different watermarking algorithms against different attack mechanisms. This is performed mostly by correlating the watermarked data with the attacked data. Copying is suspected, that is watermarking exist if the coefficient is greater than the threshold value and vice versa. These techniques will be briefly introduced in the following section. It is noted that these are concerned with digital image (raster) data only.
Digital Watermarking
To hide the watermarking signal, the watermark signal has to be concealed within the original data or data sources (asset). Data concealment is the main process for watermarking. In general, it is a reverse approach for data compression. For data compression technology, its aim is to remove unnecessary information from the data sources all those data which are perceptually less important. For the data hiding technology, the goal is to add some information to the data source so that the user will not be perceptually alerted. To keep the robustness, the output of the watermarked data should be difficult to be detected or easily removed by some filtering and digital processing. If there is a big change of the source data, the end-user will be easily alerted by comparing the non-watermarked data and the watermarked data.
There are many researches on watermarking still image on the screen that follows the Human Visual System (HVS). No matter unnecessary data are removed or extra information is added, the user should not be alerted to have a big change from the original image. Human eye is not able to perceive light radiation equally at all frequencies (Figure 3), but is more sensitive to red than blue color. It is more sensitive to middle frequencies and less to low and high frequencies of the visible spectrum.
Figure 3, Human eye visible spectrum response with spectral distribution of the color
For watermarking still images, the image binary data sequence should undergo a transformation (Discrete Fourier Transform DFT or Discrete Cosine Transform DCT). The spectrum in frequency domain is analyzed. The data in the less perceptible frequency band is modified, but not the high contrast pixels region and very high frequency component where human eyes are sensitive to changes and contrast. Finally, an inverse transformation is made to form the watermarked image. The sane principle may apply to watermarking a sound source, by which we should have good understanding of Human Auditory System (HAS). Similar to human eye, human ear exhibits a different sensibility to sounds depending on the frequency. It is almost insensitive to sound having frequency below 10Hz and above 20KHz.
To add details into original image that are not easily perceived, the fingerprint pattern recognition technology may be useful. Most fingerprint image comparison algorithms are using minutiae. Some important minutiae types available in the human finger are shown in Figure 4 and 5. The fingerprint image is first transformed into binary images through an ad-hoc filtering algorithm. The images obtained are then sampled into small sectors and submitted to a thinning process which allows for the ridge line thickness in order to achieve a one pixel width. Finally a simple image scan allows for locating the pixel that corresponds to minutiae (Figure 6). For the extracted information, position and direction of specific feature points like ridges ending, bifurcation are stored and kept as a signature for further comparison.
Figure 4 Minutiae Types for Human Figure
Figure5 Minutiae Types for Human Figure
Figure 6 Minutiae Extraction
Attack Methods to Watermarking
There are numerous ways that watermarks in digital images may be destroyed electronically, especially for low robustness and easily detectable watermarks. Common examples are JPEG Compression to a lower quality image, adding image with Gaussian Blur, Gaussian Noise or Median Noise, Image Rotation to distort location and orientation, Image Sharpening, image processing by High Pass Filter and Image Chopping. Past researches often compare these resulting images (post attack to watermarked images) with the original or watermarked images to evaluate the effectiveness of different attack methods. There is often a trade off between robustness and attack. In general, watermarking low frequency components of the image is robust to attack, but can easily be changed and disappeared by compression. By contrast, the high frequency components are difficult to be compressed or filtered, but are easily perceived if the attacker is knowledgeable of the original data.
Research Directions to Watermarking Vector Data
Geospatial data are mostly vector graphics, especially for large-scale map information. Although the end output is image on the screen, it comes with a graphic engine by which the vector information such as coordinate location, magnitude and direction can be interpreted. Unlike raster images, the vector data do not have color information. Watermarking can only be in the thoughts of modifying and documenting the coordinates, magnitude and directions. Yet, these are also used the same for attack. Some geo-spatial data owners tend to insert extra but not easily noticeable information (lines, symbols etc.) on an ad-hoc basis to protect the copyright claims. By this approach, increasing robustness will be at the expense of data capacity, else a slight change from data generalization or compression can corrupt the watermarks. Hence, there is a need to investigate an appropriate systematic method to inserting or removing information to the data such as pattern recognition and data encryption. Undoubtedly, this will lead to some errors or distortions in location and shape, but the general pattern can still be preserved. Another approach is to uncover the attack types and extents but necessitates a detailed and time-consuming comparison of datasets. More difficulty will be posed on comparing data of two different formats, e.g. image and vector, analog and digital. For whatever approach, the algorithm on watermarking or on attack detection should balance the issues of robustness, imperceptibility and capacity or complexity.
Acknowledgement
The work described in this paper was supported by a grant from the Research Grant of the Hong Kong Polytechnic University (Project No. A-PG48).
References
- Ho, K.P.S. (2001) Watermark techniques for digital images, Multi-disciplinary Studies, The Hong Kong Polytechnic University.
- Petitcolas, F.A.P., Anderson, R.J. & Kuhn, M.G. (1999) Information hiding – a survey, IEEE Proceedings, Vol.87, No.7, pp.1062-1078.
- Rabani, M., Jones, P.W. & Wash, B. (1991) Digital image compression techniques, Spie Optical Engineering Press.