METHOD OF INSERTING/DETECTING DIGITAL WATERMARK AND APPARATUS FOR USING THEREOF
Technical Field The present invention relates to a method for embedding and detecting a digital watermark in digital multimedia contents and an apparatus using the same. More particularly, the present invention relates to a digital watermark embedding/detection method and an embedding/detection apparatus using the same which can spatially form a watermark and embed and detect the watermark in an image, to thereby enhance robustness against image variations.
Background Art
Recently, together with the wide spreading of the internet and computers and the rapid distributions of multimedia data, illegal copies(piracy) and distributions are widely prevalent so that an effective protection apparatus for a copyright to multimedia data gets required. Watermarking technologies are ones that embed user information(watermark) in multimedia data to be unrecognizable by a user, to thereby prevent pirated copies and protect a copyright of a copyright owner.
The watermark means a mark developed in a step using a frame for pressing wet fibrous material to get rid of water in a process making paper from papyrus in ancient times. Marks embedded in paper in order for paper manufacturers in the middle ages to prove their own goods are the watermarks in the middle ages, and, nowadays, an image is embedded which can be recognized only with light when, in a process of making banknotes, printing on both sides of a sheet of paper after drying the wet sheet on which
printing has been done, and the image is referred to as a watermark.
In these days, together with the increase of digital media, the concept of a digital watermark has appeared. Like paper in an analog concept is substituted with the concept of digital paper, digitalizing all the analog media in which the past watermarks were embedded has brought about the concept of the digital watermark as a mark hidden in digital image, audio, video, and so on. That is, the watermarking refers to all technical methods hiding and extracting a watermark of a special form in multimedia contents in order to protect a series of multimedia contents. At the beginning, researches have been carried out for methods hiding original multimedia contents themselves, but, at present, it is a trend that strong watermarking technologies using lots of transform methods are developing.
The watermarking is classified into a visible watermarking and an invisible watermarking based on the visibility of a watermark, and the invisible watermarking is again classified into a spatial domain watermarking and a frequency domain watermarking based on the methods embedding a watermark.
The visible watermarking specifies a copyright by embedding in an original image author information which can be recognized with eyes. The visible watermarking can be used with ease but has a drawback in that the originals are damaged.
Accordingly, the invisible watermarking is mainly used in the image watermarking technologies in these days. The invisible watermarking is a technology embedding a watermark not to be visually perceived by using a limit of senses of the human visual system. While the spatial domain watermarking embeds and extracts a watermark with ease, there is a high possibility to lose a watermark by means of signal processing, video processing (non-linear filtering, rotating, cutting, moving, enlarging, and
reducing transforms, and the like), and compressing.
In the meantime, the frequency domain watermarking employs transform techniques such as Fourier transform, discrete cosine transform, or the like for embeddings and extractions, so there exists a drawback in that it has a complicated algorithm and requires lots of arithmetic operations. However, it has an advantage in that it is robust against general attacks such as filtering or compression.
The invisible embedding of a watermark requires an embedding of the same in a low value on a broad area, which is carried out by the spread spectrum technology of Ingemar J. Cox. In the spread spectrum technology, a pseudo-random sequence is used as a watermark, which is a method that can be effectively used since the sequence has a uniform distribution function and is evenly distributed over the entire bandwidth of frequencies.
For methods transforming an original image into a frequency domain, the fast Fourier transform(FFT), discrete cosine transform(DCT), and wavelet transform are generally utilized a lot, which takes a method embedding and restoring a watermark into the original state in a transform plane. However, the method has a high possibility to lose a watermark on attacks such as image rotating, cutting, moving, enlarging, reducing, or the like.
The watermarking methods in the spatial domain or frequency domain have advantages and disadvantages in their own ways, and a watermarking method using the log-polar mapping and Fourier transform has been developed to compensate for the loss of a watermark, which is the weak point of the frequency domain watermarking method, in rotating, enlarging, or reducing an image. The method converts rotations, enlargements, and reductions into a simple movement forms through the log-polar mapping and detects a
watermark by using the characteristics that the amplitudes of the Fourier transform are invariable with movements. However, the method is weak at the video processing such as compressions.
As mentioned above, the developed watermarking technologies for video have advantages and disadvantages in general in their own ways. Further, the pseudo-random sequence watermark being widely used at present can confirm what key value a watermark embedded in an image has, but has difficulties in embedding and extracting various copyright information.
Further, indiscriminately embedding a watermark regardless of the characteristics of an inputted image brings about a drawback weak at attacks from external.
Furthermore, in case of firstly casting and then embedding a watermark in an input image, an embedded watermark is changed if an image undergoes rotations, partial cuttings, or the like, causing a problem impairing copyright information.
Detailed Description of the Invention
It is an object of the present invention to provide a digital watermark embedding and detection method and a apparatus using the same which are robust against image variations such as rotation, enlargement/reduction, cutting, and filtering.
It is another object of the present invention to provide a digital watermark embedding and detection method and a apparatus using the same which spatially configure a watermark and embed the spatially configured watermark to thereby be robust against image variations.
It is yet another object of the present invention to provide a watermark detection method and a apparatus using the same which effectively detect a spatially configured
digital watermark embedded in an image signal by fitting to a position at which the watermark is embedded through convolutions.
In order to achieve the above objects, a method for embedding a digital watermark in an image signal according to the present invention comprises steps of: using a user key and an inherent key and generating respective pseudo-noise codes thereof; adding the pseudo-noise code generated based on the user key and the pseudo- noise code generated based on the inherent key; and adding to the image signal a digital watermark formed by a step for arranging in a two-dimensional fashion a watermark formed by the addition.
Further, a method for detecting a digital watermark according to the present invention comprises steps of: strengthening a component of the digital watermark embedded in the image signal; generating a watermark arranged in a two-dimensional format; resizing the generated digital watermark to a image signal size; applying a convolution integral to the resized digital watermark and the image signal and detecting peaks included in the image signal; measuring positions of the detected peaks, capturing respective regions at the peak-measured positions for an addition thereof, and extracting a watermark embedded in the image signal; calculating a correlation of the extracted watermark and the generated digital watermark; and detecting a watermark embedded in the image signal based on the correlation.
As stated above, unlike a method simply embedding a watermark of a certain form in the existing spatial domain, the present invention forms a watermark not linearly but spatially and embeds the watermark so that the watermark robust against external variations such as image rotations, cuttings, or the like, can be embedded. In particular, a watermark embedded according to the present invention is arranged and embedded in a radial fashion or in a form of plural concentric circles from a watermark of a stream fashion.
Furthermore, unlike the existing methods, a high frequency filter is applied to a watermarked image for a watermark detection, all blocks of the watermarked image are added to obtain an effect removing image components, to thereby enhance a watermark detection speed as well as increase a watermark detection rate.
Brief Description of the Drawings
The above objects and other features of the present invention will become more apparent by describing in detail a preferred embodiment thereof with reference to the' attached drawings, in which:
Fig. 1 is a block diagram for schematically showing a structure of a digital watermark embedding and detection apparatus according to an embodiment of the present invention; Fig. 2 is a flow chart for showing operations of an image converter of a watermark embedding apparatus of Fig. 1;
Fig. 3 is a block diagram for schematically showing a structure of a watermark generator of the watermark embedding apparatus of Fig. 1;
Fig. 4 is an exemplary view for showing a two-dimensional watermark
implemented by a watermark configuration of Fig. 3;
Fig. 5A and Fig. 5B are other exemplary views for showing two-dimensional watermarks;
Fig. 6 is a flow chart for showing operations of an image recorder of the watermark embedding apparatus of Fig. 1;
Fig. 7 is an exemplary view for showing filters serving watermark detections, wherein Fig. 7A shows a high boost filter, Fig. 7B shows a Laplacian filter, and Fig. 7C shows a Difference of Gaussian(DoG) filter having 7x7 and 9x9 masks;
Fig. 8A is an exemplary view for showing a watermarked image before filter processing, Fig. 8B to Fig. 8D show exemplary views for showing processing results by a high boost filter, a Laplacian filter, and a DoG filter, respectively;
Fig. 9 is an exemplary view of a mask form employed for effective watermark detections;
Fig. 10 is a block diagram for schematically showing a watermark resizer of Fig. 1;
Fig. 11 is an exemplary view for showing an arithmetic operation result by a watermark convolution operator of Fig. 1;
Fig. 12 is a flow chart of showing operations of a post-processor of a watermark detection apparatus of Fig. 1; Fig. 13 is an exemplary view for showing a peak detection used for a watermark detection; and
Fig. 14 is a flow chart for showing operations of the watermark detector of Fig. 1.
Embodiment
Hereinafter, a watermark embedding and detection method and an embedding/detection apparatus using the same will be described in detail with reference to the accompanying drawings. Fig. 1 is a block diagram for schematically showing a structure of a digital watermark embedding and detection apparatus according to an embodiment of the present invention.
The digital watermark embedding and detection apparatus in Fig. 1 comprises a watermark embedding apparatus 100 for embedding a watermark into an inputted image and a watermark detection apparatus 200 for detecting a watermark from a watermark- embedded image. The watermark embedding apparatus 100 includes an image converter 110 for converting an inputted image 10 into a predetermined form based on the characteristics thereof, an image analyzer 120 for analyzing the number of colors, a histogram shape, a ratio of high and low frequencies, or the like, which are indicative of the characteristics of the inputted image 10 and determining the strength of a watermark to be embedded, a watermark generator 130 for generating a watermark spatially arranged, an adder 140 for adding an image signal outputted from the image analyzer 120 to a watermark generated from the watermark generator 130, and an image recorder 150 for recording a watermark-embedded image signal. Further, the watermark detection apparatus 200 includes an image converter 210 for receiving and converting a retrieved image signal into a format of a predetermined form, a pre-processor 220 for strengthening the characteristics of a watermark included in an output signal of the image converter 210, a watermark generator 230 for generating a watermark spatially arranged, a watermark resizer 240 for resizing a magnitude of a
watermark-embedded signal to a size of an image signal, a convolution operator 250 for applying an convolution integral to a watermark characteristics-strengthened signal included in an image signal and a watermark outputted from the watermark resizer 240, a post-processor 260 for playing a role of extracting a watermark from peaks obtained from a result of the convolution integral, a correlation calculator 270 for calculating a correlation between a watermark outputted from the watermark resizer 240 and an image signal processed in the post-processor 260, and a watermark detector 280 for detecting a watermark included in an image signal based on an output value of the correlation calculator 240. The operations of the watermark embedding and detection apparatus having the above structure will be described in respective parts. Firstly, the operations of the watermark embedding apparatus 100 will be described with reference to Fig. 2 to Fig. 6.
The inputted image 10 is inputted to the image converter 110 for embedding a watermark in a digital image signal. Describing an operation flow of the image converter 110 with reference to Fig. 2, the image converter 110 checks if the inputted image 10 is a 24-bit color image(SlOO). At this time, it can be determined by checking the header information of the inputted image signal whether the inputted image is a 24-bit signal. If the inputted image 10 is a 24-bit color, the image converter 110 separates the colors into respective R, G, and B channels(SHO) for outputs to the image analyzer 120. However, if the inputted image is not a 24-bit, an output is made without any separation into channels so that processing is done for one channel.
Further, if an inputted image is a 24-bit color image and is in the NTSC mode processing in a YIQ format rather than in an RGB format, the RGB components of the inputted image are converted into a model of the YIQ(Y: Luminance, I: In-phase, and Q:
Quadrature) format by using Formula 1 as below. [Formula 1]
The I and Q components in a model of the converted format are separately stored, and only the Y component is extracted(S120). The extracted Y component is transferred, for a next process, to the image analyzer 120 for analyzing image characteristics.
As stated above, if an inputted image is not a 24-bit, it is possible without a separate image converter 110.
A processed result of the image converter 110 is outputted to the image analyzer 120. The image analyzer 120 determines the strength of a watermark to be embedded block by block in consideration of the entire characteristics of an inputted image. The determination of the watermark strength can be accomplished in a variety of forms. For example, it is determined based on the number of colors used for respective channels in a block, a histogram shape, an energy ratio of high and low frequencies, and so on. Viewing it in more detail, when an image is divided block by block, the number of colors used for each block and a color value are obtained. As a result, if the number of used colors are large and the color value is high, a real image corresponds to one having severe color changes or colors of brilliant forms. Accordingly, no severe influence is visually undergone even though a watermark to be embedded in a corresponding block is strengthened.
However, if the color changes are small, a feeling may be given like lots of noise
is included in an original image even though a less strengthened watermark is embedded. Therefore, under considerations of the number of colors used in a block and a color value, it may be determined to strengthen a watermark to be embedded when the value is high and to less strengthen a watermark when the value is low. Further, when applying a DCT transform to an image and indicating the result in a block, it is characterized that part corresponding to a low frequency region is clustered on the upper left of the block, part corresponding to an intermediate frequency region on the middle, and part corresponding to a high frequency region on the lower right. That is, the characteristics of an inputted image may be grasped based on a ratio of a low frequency energy and a high frequency energy as a result of the DCT.
Moreover, if an inputted image is analyzed channel by channel, for example, in case of a 24-bit image, respective R, G, and B channels have 8 bits(28=256) and 0 ~ 255 values, a histogram is prepared based on the above values for image regions, and it is possible to grasp the changes and used colors in the image based on the shape and variation of the histogram. That is, if the number of used colors is small, the distribution of a histogram becomes narrow, and, to the contrary, if the number of used colors is large, the distribution of a histogram becomes wide. The large number of used colors means that an image has severe variations, and, to the contrary, the small number of colors means that an image is dull without particular variations. Therefore, through this, it may be determined whether the energy of an image is concentrated on a high frequency region or on a low frequency region.
As stated above, the image analyzer 120 analyzes the characteristics of an inputted image and determines the strength of a watermark to be embedded block by block. Based on such an analysis result, inputted images separated channel by channel are added
with a watermark generated from the watermark generator 130 for embeddings into the inputted images.
The watermark generator 130 generates a watermark based on a structure shown in Fig. 3. First, if a user key is inputted to a pseudo-noise code generator 122 for a seed value, a pseudo-noise code is generated by using the seed value. In the meantime, an inherent key, separately from the user key, generated to facilitate a watermark detection is inputted to a pseudo-noise code. generator 124 to generate a pseudo-noise code in the same manner as well.
The two pseudo-noise codes so generated are transferred to an adder 126 to be added each other. An added noise code is inputted to a watermark configurer 128. The watermark configurer 128 newly configures a watermark in a spatial arrangement form instead of one formed in a one-dimensional stream fashion. Viewing an example of a shape shown in Fig. 4, a watermark is configured in a two-dimensional radial fashion while rotating 360 degrees about a first value of a watermark of a predetermined length. In case of configuring a watermark in the two-dimensional radial fashion as above, no influence is given since the form of a watermark does not vary even though a watermark-embedded image varies by rotations and the like.
Further, other than the radial fashion, such a watermark shape may be configured in a circular form as shown in Fig. 5A and Fig. 5B. In case that a watermark shape is configured in a circular form as shown in Fig. 5B, first, a one-dimensional watermark is formed((a) of Fig. 5A), and a watermark is arranged in a concentric form about the center of a block. In case of arranged in the concentric form, there exists a radius difference between a watermark formed inside and a watermark formed outside, so a difference between the lengths of the bits configuring the watermarks takes place. Accordingly, the
watermarks are sampled at a predetermined rate based on radius sizes to be enlarged and reduced for preparations((b) of Fig. 5A), and the sampled watermarks are arranged in a circular fashion.
When arranged in a circular fashion, after forming the upper side in a watermark stream, the stream is duplicated and then arranged in the same rotation direction on the lower side, so the two-dimensional watermark can be formed in the shape shown in Fig. 5B.
The watermark so formed is outputted to be added to an image signal of the image analyzer 120. The adder 140 first divides an image into blocks of a predetermined size, that is, watermark sizes, in order to add a watermark formed from the watermark generator 130 and an image signal outputted from the image analyzer 120. A divided image signal is added to a watermark signal with an "application of a value adjusting a strength of a watermark to be embedded which is determined based on the entire characteristics(the entire characteristics for one inputted image scene) of an inputted image in the watermark analyzer 120. That is, the strength of a watermark is adjusted by multiplying a watermark of Fig. 4 by a watermark strength determined in the image analyzer 120, and then added to the image signal.
A method adding a watermark is to directly embed a watermark into each channel without passing through the image converter 110, that is, to independently embed a watermark into a gray channel for a gray image and into respective R, G, and B channels for an RGB image(24 bits). If a watermark-embedded signal is outputted, the signal is inputted to the image recorder 150 for storage as a watermarked image.
The recording operations in the image recorder 150 is described with reference to
Fig. 6.
The image recorder 150 determines whether a watermarked signal is a 24-bit image or not(S200). If the watermarked signal is a 24-bit image, the image recorder 150 adds signals which are separated channel by channel(S210), and, if not a 24-bit image, the signals are ones inputted without a separate conversion process in the preceding image converter 110, so that a step S230 directly proceeds for storage since the above conversion is not necessary.
Further, in case that a signal processing is applied to an inputted image in the YIQ format of the NTSC mode rather than the RGB format, a watermark-embedded image signal is added to the Y component and the IQ components remaining beforehand after extracting the Y component from the components in the image converter 110. After the addition, the signal of the YIQ format is converted again to the RGB signal by using Formula 2 as below(S220). [Formula 2]
The signal converted as above is stored in a storage medium as a watermarked image.
However, in case that a watermarked signal is not a 24-bit image in step S200, the signal is one inputted without a separate conversion process in the preceding image converter 110 and the above conversion is not necessary, to thereby proceed directly to a step S230 for storage.
The watermark embedding apparatus 100 as above arranges a watermark in a two-
dimensional space for embeddings, so that, even though a watermark is directly embedded in a spatial domain without a conversion process into a frequency domain, there exists an effect robustly keeping a watermark alive against image variations such as rotations, cuttings, and the like with respect to a watermark-embedded image. Further, upon embedding a watermark, the strength of the watermark to be embedded is determined for embedding in consideration of image characteristics, to thereby enable a watermark to be more effectively embedded.
Further, a watermark detection apparatus for detecting a watermark from an image signal in which a watermark is embedded as above is described with reference to Fig. 1, and Fig. 7 to Fig. 14.
A watermarked image can flow into illegal users through various ways, and its piracy, variations, and so on, the image can be carried out. However, spatial arrangements and embeddings of a watermark by the watermark embedding apparatus 100 according to the present invention as above makes the watermark robust enough to be able to maintain a shape of the watermark embedded in an image despite image variations such as image rotations, cuttings, and so on, and an apparatus and method for detecting a watermark embedded by such a method is described.
If a watermark-embedded image is inputted to the watermark detection apparatus 200, the image is firstly converted into a signal of a predetermined format through the image converter 210. The structure and operations of the image converter 210 in the watermark detection apparatus 200 are the same as those of the image converter 110 in the watermark embedding apparatus 100. That is, if an inputted image is a 24-bit image, the inputted image is separated into channels(R, G, and B) for outputs, and, if not a 24-bit image, the image is outputted without a separate process.
Further, if the inputted image is a 24-bit and processed in the YIQ format based on the NTSC mode other than the RGB format, an RGB-format signal is converted into the YIQ format, and only the Y component is extracted and outputted to detect a watermark.
An image signal outputted from the image converter 210 is inputted to the pre- processor 220. The pre-processor 220 is for boosting the characteristics of a watermark embedded in the image signal, which performs high-pass filtering, sharpen-filtering, and high-boost filtering processing. Examples of filters employed in such a pre-processor are shown in Fig. 7 to Fig. 9.
Fig. 7 shows examples of various spatial filters playing a role of boosting the high frequency components of an image signal, Fig. 7A, Fig. 7B, and Fig. 7C show mask forms for a high boost filter, a Laplacian filter, and a Difference of Gaussian(DoG) filter, respectively. The high boost filter of Fig. 7A serves watermark detections and plays a role of boosting a watermark signal. That is, The filter reduces an image component energy and increases a watermark signal energy, to thereby play a role capable of effectively detecting a watermark. Further, the DoG filter of Fig. 7C follows Formula 3 as below. [Formula 3]
DoG(x,y) =
2π σ 1 2π σ 2
One example of a processing result by the filter as above is shown in Fig. 8. Fig. 8A is an exemplary view for showing a watermarked image before filter processing, Fig.
8B to Fig. 8D show exemplary views for showing processing results by a high boost filter,
a Laplacian filter, and a DoG filter, respectively.
A filter shown in Fig. 9 may be used, besides the filters in Fig. 7, to reduce an image component energy and to strengthen a watermark component energy.
The pre-processor 220 as stated above is for strengthening a watermark component of an image signal, which can be processed by using any one of the filters shown in Fig. 7 or Fig. 9.
In the meantime, the watermark generator 230 of the watermark detection apparatus 200 does not include the adder 126 compared to the watermark generator 130 of the watermark embedding apparatus 100 in Fig. 3, uses pseudo-noises generated from the pseudo-noise code generators 122 and 124 based on an inherent key and a user key, respectively, and generates a two-dimensional watermark spatially arranged for the respective pseudo-noises from the watermark configurer 128. Further, it is possible to generate a two-dimensional watermark spatially arranged after adding pseudo-noise codes obtained with respect to the respective pseudo-noise codes, with the same process as that of the watermark generator 130 of the watermark embedding apparatus 100. However, respective watermarks inputted to the correlation calculator 270 to be later described are ones not added.
A generated watermark is inputted to the watermark resizer 240. The watermark resizer 240, as shown in Fig. 10, includes a zero padding part 242 for resizing the generated watermark to an image signal size, and a watermark enlargement/reduction part 244 for enlarging/reducing a watermark size.
If a watermark generated in the watermark generator 230 is inputted to the zero padding part 242, the watermark is fit to an image size and '0' values are filled, for an output, in remaining portions except for portions in which a watermark actually exists.
The reason for filling with '0' in all the portions except for the watermark after the watermark is fit to an image size, as stated above, is because processing can be simply done in a multiplication form if converting into the frequency domain a convolution integral in the spatial domain which is a next operation step, so the same size as an image signal size is formed and a zero padding is performed for the sake of calculations.
The zero padding-processed watermark is inputted to the convolution operator 250. The convolution operator 250 carries out a convolution integral with respect to the watermark component-strengthened signal of an image signal by the pre-processor 220 and the zero padding-processed watermark. Since the convolution in the spatial domain performs lots of repeated multiplications, for the sake of operation efficiency, transforming is applied with a Fourier transform into the frequency domain, a multiplication is carried out with respect to the transformed watermark-embedded image and a watermark, and a result of which is inverse-Fourier-transformed into the spatial domain again, to thereby enable a convolution result to be obtained. Such calculations are carried out by using Formula 4 as below. [Formula 4]
Conv = IFFT2( vFFT2(W ι.rag ) XFFT2( vW m-zeropad .di .ng ))
In here, IFFF2 denotes a two-dimensional inverse Fourier transform, FFT2 a two- dimensional Fourier transform, Wimg a watermark-embedded image, and Wm-zeropadding is a zero-padding watermark.
As a calculation result by the above Formula 4, it is checked whether there is a peak having a value over a predetermined magnitude from the result values. A position at which a watermark is embedded in an image can be grasped from a position at which such
a peak exists. If any peak does not appear as a result of the convolution calculations, the calculations continue while resizing a watermark until a peak appears. That is, if a peak does not appear, it is notified to the watermark enlargement/reduction part 244 in the watermark resizer 240, the watermark enlargement/reduction part 244 resizes a watermark generated from the watermark generator 230 for transferring to the zero padding part 242. The zero padding part 242 fills with 'O's in an image size remaining except for a resized watermark or an output.
A watermark can be resized in various ways, and, in the present invention, peaks appear every repeatedly embedded period if an autocorrelation is obtained with respect to a watermarked image, and intervals among the peaks are measured for resizing a watermark. However, since a peak interval may occur at a different position due to noise, intervals among peaks are defined as candidate scale elements, and a watermark is resized with respect to the candidate scale elements.
The convolution operator 250, as stated above, carries out the convolution with respect to a watermark component-strengthened signal of the image signal by the preprocessor 220 and a zero padding-processed watermark. Such a calculation finds out a peak, as shown in Fig. 11, while resizing a watermark until the peak is detected within a range of the predetermined number of times.
In the watermark embedding, a corresponding image with respect to one image is divided into plural regions corresponding to a watermark size and a watermark is embedded for the respective regions, so peaks appearing in the plural small regions as a result detected through the above process is that the watermark embedded in corresponding regions is detected.
If peaks are detected, the post-processor 260 is performed in accordance with the
operation flows shown in Fig. 12. First, as shown in Fig. 11, if peaks are detected as a convolution result, the positions of all the spots at which corresponding peaks appear are detected(S300). A peak detection is to extract the certain number of peaks after arrangement in the descending order. Such positions are the above convolution result and the peaks are detected at the corners of a region a watermark occupies, so the regions in which peaks appear are captured by watermark sizes(S310), and the regions are added(S320).
Since the peak detection in the above process is done when two identical signals are matched and a signal correlation is obtained while sliding the two signals, the biggest peak appears when the two signals are matched. Accordingly, since a part at which the signals are matched is one coincident with a region of a set watermark block, a peak is detected at a corner.
As stated above, with the addition of the regions captured in the post-processor 260, an image component energy comes closer to an average, but a watermark component gradually increases, to thereby play a role of strengthen the watermark component as well.
If a result is obtained by the addition, a result signal and a watermark signal generated by the watermark generator 230 are inputted to the correlation operator 270 for obtaining a correlation Corr by using Formula 5 as below. The correlation is respectively carried out with respect to a watermark generated by an inherent key and a watermark generated by a user key. [Formula 5]
Corr = IFFT2(FFT2( lmaesum) Xconj(FFT2(W J))
In here, Wimgsum denotes a watermark-strengthened signal processed by the post-
processor and Wm denotes a watermark generated by the watermark generator. IFFT2 denotes a two-dimensional inverse fast Fourier transform, FFT2 a two-dimensional fast Fourier transform, and conj a complex conjugate.
In the correlation calculation using the above Formula 5, a multiplication is performed with respect to data obtained through the two-dimensional fast Fourier transform of a watermarked image Wimgsum and data obtained through the two-dimensional fast Fourier transform of a watermark Wm generated from the watermark generator 230 based on a user key or an inherent key, and the inverse fast Fourier transform is applied to the multiplication again for a conversion into the spatial domain. The conversion into the frequency domain and the multiplication calculation as stated above can reduce the number of calculations compared to the case of the convolution of an image watermarked in the spatial domain and a watermark, so that data processing can be performed at a higher speed.
Fig. 13 is a floating view of a correlation calculated by Formula 5 under assumption of a case that a watermark is embedded. Since the correlation obtained by Formula 5 is a two-dimensional sequence form rather than a certain value, the maximum peak value and its position can be obtained through a process as below in comparison of such plural values.
Just like the watermark generator 230 generates a watermark through an inherent key and a user key, as shown in Fig. 14, the watermark detector 280, if peaks occur, checks whether all the two key values exist and the peaks appear at the same position(S400). Sharpness is calculated if the two peaks occur at the same position(S410). The sharpness calculation determines, through the fourth moment(Kurtosis; K) in Formula 6 as below, whether a K value has a value above a predetermined threshold value(S420), and it is determined that a watermark is detected when satisfying both of the two conditions.
[Formula 6]
Here,
denotes a result value of a correlation between two watermarks
X X
Wm and Wimg, an average of 1'" "' N , and a standard deviation. The determination as to whether a value of K is more than a certain threshold value in the above step is to determine whether a watermark is embedded through a comparison between a peak value and a set threshold value since a peak appears high at an calculated value in case that the watermark is embedded. In here, the threshold value is shown as a value constantly allocated by experiments. However, when the condition is not satisfied in the step S420, it is determined that a watermark is not detected.
Industrial Applicability
As described above, the present invention relates to a method and apparatus for embedding and detecting a watermark, which uses a watermark formed in a two- dimensional radial fashion for its configuration changes upon embeddings thereof, enhancing the robustness of a watermark against signal variations. Further, upon the detection of the watermark, a convolution integral is used to detect a watermark embedded in the two-dimensional radial fashion as above for effective detections, enhancing the accuracy and promptness for watermark detections. Although the preferred embodiment of the present invention has been described in particular, it will be understood by those skilled in the art that the present invention should
not be limited to the described preferred embodiment, but various changes and modifications can be made within the spirit and scope of the present invention as defined by the appended claims.