KR100339860B1 - Method for embedding image watermark into audio signal or extracting watermark from watermarked audio signal and recording medium therefor - Google Patents

Abstract

The present invention relates to a method of extracting a watermark by using an approximate original after inserting a video signal as a watermark into an audio signal. The method of inserting a video watermark into an audio signal according to the present invention comprises: (a) original Sampling the audio signal at a predetermined sampling period to generate a sampling audio signal; (b) generating a watermark corresponding to the seal image according to a seal pattern generated from a predetermined seal image and a secret key determined for the seal pattern; (c) setting a period at which the watermark is inserted into the sampling audio signal; And (d) inserting the watermark into the sampling audio signal in units of the insertion period, and inserting a seal image as a watermark in the original audio signal, wherein the watermark signal of the seal image is audio. Minimize damage to the original audio signal due to the insertion of watermark by inserting it into the signal, and when extracting the watermark contained in the audio signal of the inspection object, generate an approximate original and then extract the watermark more simply. It can be checked whether the audio signal is stolen.

Description

Method for embedding image watermark into audio signal or extracting watermark from audio signal and extracting watermark from watermarked audio signal and recording medium therefor}

The present invention relates to a method of inserting and extracting a watermark in an audio signal, and more particularly, to a method of inserting a video signal as a watermark into an audio signal and then extracting the watermark using an approximate original.

Recently, copyright protection of digital multimedia data has become a social problem, starting with MP3 music files. Watermarking technology allows copyrights to be identified by placing a watermark on certain data that can represent the original owner, especially in images, audio, text, video, and multimedia data, for sale or distribution. As a technique for copyright protection, it is a legal means by which the owner can claim the copyright by extracting the watermark from the piracy data by notifying the distributor that it is watermarked and extracting the watermark from the piracy data. Can also be utilized.

The watermarking technique consists of two algorithms for inserting a mark and an extraction algorithm that extracts the inserted mark and compares the similarity with the mark inserted by the owner. According to the prior art, an image watermark signal is used when the original material is an image signal, and a watermark signal that is an audio signal is used when the original material is an audio signal. In particular, when a watermark is embedded in an audio signal, a certain range of amplitude displacement is necessary to be a meaningful watermark audio signal. When such a watermark signal is added to the original audio signal, the original audio signal is distorted and white noise is generated. There is a problem that the original sound is significantly damaged by the back.

Also, in order to prove that the third party illegally uses the original, the original without the watermark was required to prove that the original included the watermark. That is, a method of confirming whether a watermark is extracted by comparing an original with a watermark and a third party is used. However, since the original used by the user already includes a watermark, when compared with a third party's image, the watermark is also included in the third party's image, and thus the watermark is not extracted. Therefore, there is a problem that an original without the original watermark is always required to check the watermark.

SUMMARY OF THE INVENTION The present invention has been made in an effort to provide a method for inserting a video signal converted into a bit string into a watermark and a recording medium for the original audio signal.

Another object of the present invention is to provide a method for extracting an approximate original from an audio data to be examined and extracting a watermark included in the data without an original data, and a recording medium therefor.

1 is a flowchart illustrating a method of inserting a seal image into an original audio signal according to the present invention.

2A illustrates an example of a seal image to be used as a watermark, and FIG. 2B illustrates a watermark in which binary data of the seal image and a random number by a secret key are combined.

3 is a diagram illustrating a process of generating a sampling signal from an analog audio signal and inserting a watermark signal into the sampled audio signal.

4 is a flowchart illustrating a method of extracting an original seal image from an audio signal into which a watermark signal obtained from a seal image is inserted according to the present invention.

FIG. 5A shows an audio signal with a watermark embedded therein, and FIG. 5B shows an approximated original audio signal obtained using a peripheral value.

FIG. 6 illustrates an An * Bm-sized seal image in each pixel unit.

FIG. 7A illustrates a seal image extracted from an audio signal, and FIG. 7B illustrates a seal image after post-processing the seal image of FIG. 7A.

In order to achieve the above object, a method of embedding a video watermark in an audio signal according to the present invention,

(a) sampling the original audio signal at a predetermined sampling period to generate a sampling audio signal; (b) generating a watermark corresponding to the seal image according to a seal pattern generated from a predetermined seal image and a secret key determined for the seal pattern; (c) setting a period at which the watermark is inserted into the sampling audio signal; And (d) inserting the watermark into the sampling audio signal in the embedding period, and inserting a seal image as a watermark in the original audio signal.

Step (b) further includes inserting a frame sync signal indicating a watermark at a position where the seal pattern starts to generate a watermark, or further inserting a line sync signal indicating a line feed at a line feed position of the seal pattern. It is desirable to generate a watermark.

In order to achieve the above another object, a method of extracting a watermark from the audio signal to be inspected according to the present invention,

(a) randomly generating bit string data according to a secret key determined for the original audio signal corresponding to the inspection target audio signal; (b) generating an approximate audio signal by obtaining a value of each sample data in the inspection target audio signal according to a predetermined number of sample data in its vicinity; (c) comparing the approximate audio signal with the inspection target audio signal and extracting a watermark included in the inspection target audio signal; And (e) restoring a seal pattern included in the inspection target audio signal by combining the extracted watermark and the bit string data obtained in the step (a).

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the present invention.

1 is a flowchart illustrating a method of inserting a seal image into an original audio signal according to the present invention.

First, the user selects a seal image to be used as a watermark (11). The watermark used in the example of the present invention is an additional image embedded in the original audio signal to symbolize the owner, which means the owner's name or logo, and is also called a seal image. Inserting a seal image as a watermark can increase the visual effect on ownership by extracting the owner's unique mark at the time of extraction. Watermarks should use invisible watermarks that are robust to intentional and unintentional attacks.

An example of a seal image that can be used as a watermark is shown in FIG. 2A and generally uses recognizable letters or patterns. In this example, the size of the seal image is 128x128 (pixels), and the black and white image is used to easily convert the random bit stream into -1 and +1. Of course, if the seal image is black and white, it is converted into a bit string of two digital levels (ie, -1, +1). If the seal image is a four color level (ie -2, -1, +1, +2), it can be converted into a bit string. Accordingly, the method described below may be applied to the seal image formed of a plurality of colors by pluralizing the digital level of the bit string.

Next, the analog audio signal as an original material is converted into a digital audio signal (12). At the time of reproduction, an analog signal is sampled at a sampling period such that the original audio signal is not damaged, and a discrete signal is generated and converted into a digital signal having a certain number of bits. If the original material is a digital signal, this step is of course not necessary.

A watermark signal is generated from the seal image and the secret key (13-15). Here, the secret key may be a key given to each user, or may be a key uniquely assigned to each original material. When each pixel of the seal image shown in FIG. 2A is replaced with a combination of +1 and -1, and the bit stream is S, it may be expressed as Equation 1 (13). The bit string S is a one-dimensional bit string consisting of -1 and +1. The bit string S is defined to have a value of -1 when each pixel is black and 1 when it is white in the black and white seal image to be inserted.

FIG. 6 is a diagram illustrating an An * Bm-sized seal image for each pixel unit. The S bit string includes a sequential bit string starting from A ₁ B ₁ pixel to AnBm pixel. That is, the S bit string is composed of {A ₁ B ₁ , A ₂ B ₁ , ..., AnB ₁ , A ₂ B ₁ , ..., AnBm}, and -1 if the pixel is black, +1 if the pixel is black. Has a value.

The random bit string p _i generated by the secret key is defined as in Equation 2 below (14).

p _i is a one-dimensional bit string consisting of -1 and +1 generated by the random number generator, and corresponds to the size of the bit string S. The random number generator generates a random number using the secret key as 'seed number'. The watermark W _i to be inserted is defined as follows by combining s _i and p _i (15).

Here, α _i is an element (intensity element) that determines the intensity of the watermark. When the value is large, the robustness is increased, but the degree of damage of the original audio data is increased. Since the audio signal is very sensitive to its deformation, determining the appropriate intensity factor is an important factor in determining the quality of the watermarked audio data. In this embodiment, the intensity factor α _i for the i th audio sample data is determined in proportion to the magnitude value of the discrete signal obtained by sampling the original audio signal to be inserted. In other words,

here, Is the size (amplitude) of the i-th audio sample data to be inserted, and K is a constant (K> 0) indicating the insertion strength, and about 0.01 is appropriate. 2b is a binary image of the seal image and a random number combined by a secret key ( Indicates a watermark of the state.

When the watermark signal is obtained, the period (or embedding frequency) to be inserted into the audio signal is set (16), and then the watermark signal is inserted into the digital audio signal (17). If a watermark signal is added in units of an audio sampling signal, the audio sample data having the watermark embedded therein is expressed as follows.

here, Refers to the i th watermark signal, Denotes the i th sample data of the audio signal into which the i th watermark is inserted.

If the period in which the watermark signal is inserted is larger than the sampling period of the audio signal, that is, when one watermark signal is inserted per r audio sample data, the i th watermark signal is input to the ri th audio sample data. Is added.

Referring to FIG. 3, the analog audio signal 31 is sampled at a constant sampling period T _A to generate audio sample data 32, and the watermark signal 35 is audio sample data at a constant period T _W. Is inserted into Accordingly, the watermark signals W ₀ , W ₁ , W ₂ ,,, and the like are inserted into the audio sample data such as X ₀ , X ₁ , X ₂ , ..., and the like. Audio sample data between ₁ and X ₂ ) remains unchanged.

In this example, α, which determines the magnitude of the watermark signal, is proportional to the amplitude of the audio sample data to be inserted, so that when the amplitude of the sample data is large, as shown in Fig. 3, the magnitude of the watermark signal is added to the sample data. If the amplitude of the sample data is large in proportion to each other and the amplitude of the sample data is small, it can be seen that the magnitude of the watermark signal added to the sample data is also relatively small. Therefore, it is possible to minimize the damage of the original audio signal by adaptively inserting a watermark according to the size of the original audio signal.

When the size of the seal image is 128x128 pixels, the number of data of the watermark signal is 16K, and the number of sampling data of the audio signal is usually much larger than that. Therefore, the watermark signal of 16K size may be inserted into the audio signal several times. In this case, it should be possible to check the start and end of each unit watermark signal. In this method, the absolute position of the audio sample data into which the watermark signal is inserted (for example, the start of the watermark signal starts at the first sample data and ends at the Nth sample data, and immediately after the N + 1th sample data. It is possible for the owner to know in advance whether to start the mark signal or to skip over M sample data at the end of the previous watermark signal) and use it when extracting the watermark signal. If a part of the is damaged, the watermark signal cannot be extracted or the desired watermark signal is not extracted.

Thus, a frame synchronization signal can be inserted to indicate the start point and / or end point of the watermark signal. For example, a random format of data (e.g., 15 bits of data such as 0100 0001 1111 111) is randomized by the owner's secret key and inserted into the audio sampling data just before the watermark signal of the seal image starts. . In order to know whether the watermark signal is the end, a synchronization signal indicating an end point may be inserted and used to extract the watermark signal. If the owner knows the magnitude of the watermark signal, the number of data of the extracted watermark signal may be counted.

The seal image is composed of horizontal (n pixel) x vertical (m pixel) data, and is displayed as a normal image only after line feed in n pixel data unit. Therefore, the owner may know the number of horizontal and vertical pixels of the seal image in advance and give a line feed to the watermark signal extracted based on this, but some of the watermarked audio signals are lost (either intentionally or unintentionally). If the watermark signal extracted thereafter is line fed with a known data size, an image of a desired shape is not extracted. Therefore, inserting data of a certain format into the line synchronization signal in each line feed of the seal image helps to extract the outline of the entire seal image even if a part of the audio signal is lost.

4 is a flowchart illustrating a method of extracting an original seal image from an audio signal into which a watermark signal obtained from a seal image is inserted according to the present invention.

In order to extract the seal image embedded in the audio signal, a secret key known only to the owner is required, which must be the same as the secret key used when inserting the seal image into the original audio signal. According to the present invention, an approximate original is used from a watermarked audio signal so as not to require the owner's original when extracting the seal image.

First, a random bit string p _i 'is obtained using the owner's secret key (41). This is the same as the bit string (see Equation 2) p _i generated during the watermark embedding process.

An approximate original is obtained by averaging the neighboring sample values from the watermarked digital audio signal to their own values (42). That is, the i th audio sample data ( ) Is obtained by averaging (c-1) data of itself and its surroundings.

Where c (> 1, odd) represents the average number of samples to be averaged. For example, when c = 5, the i th sample data X _i 'of the approximate original audio signal is the {i-2, i-1, i, i + 1, i + 2} th of the audio signal to be inspected. The average value of sample data is obtained. FIG. 5A shows an audio signal with a watermark embedded therein, and FIG. 5B shows an approximated original audio signal obtained using a peripheral value.

Next, the extracted watermark signal W _i 'compared to the' factor of X in the water and the audio signal mark _"i approximation X _i of the coefficient of the original 34.

The seal image s _i 'is obtained by multiplying the extracted watermark signal W _i ' by the bit string p _i 'obtained by the owner's secret key (44). As a result, the extracted seal image is represented as shown in FIG. 7A.

After extracting the seal image s _i 'through the extraction step described above, the degree of similarity between the extracted seal image s _i ′ and the original inserted seal image s _i for the claims should be measured (46). In other words, the greater the degree of similarity between the extracted watermark (seal image) and the original inserted watermark, the more obvious the ownership. In the present invention, the similarity between the two seal images can be calculated by calculating the correlation. That is, when the correlation coefficient (Correlation Value: CV) is defined as an equation, the following equation (15) is obtained.

In formula, N _i 'represents the number of bits coincident between the extracted seal image and the original seal image, and N _i represents the number of bits of the original seal image inserted in the audio signal. As shown in the equation, the correlation coefficient for similarity comparison is calculated by comparing the number of bits of the inserted seal image with the extracted seal image.

Meanwhile, in order to increase the sharpness of the extracted image after measuring the similarity after extracting the seal image, post-processing may be performed on the extracted seal image (45). Since the extracted seal image is a pattern of a black and white image, a black and white edge is detected more clearly through post-processing to increase the recognition of the pattern. Post-processing uses a kind of filtering method that masks the extracted seal image, calculates the number of black and white pixels in the mask window, and replaces them with relatively large pixel values.

In Fig. 6, when the size of the mask window is 5x5, the value of the pixel A5B5 65 is determined depending on whether there are many black pixels or back cell pixels in the pixels in the window 63 formed by {A3B3-A7B7}. do. FIG. 7B shows the seal image after the seal image of FIG. 7A is post-processed, and it can be seen that the sharpness of the pattern is much improved.

According to the present invention, original data information is not required by inserting a watermark of a video signal into an audio signal and generating an approximate original from the watermarked signal during extraction. In order to obtain an approximate source in the extraction step, a method of averaging the peripheral values of the watermarked audio signal is used.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

As described above, according to the method of the present invention, the watermark signal of the seal image is inserted into the audio signal, thereby minimizing the damage of the original audio signal due to the watermark insertion. When extracting the watermark, an approximate original can be generated, and then the watermark can be extracted to confirm whether or not the original audio signal is stolen.

Claims (11)

(a) sampling the original audio signal at a predetermined sampling period to generate a sampling audio signal; (b) generating a watermark corresponding to the seal image according to a seal pattern generated from a predetermined seal image and a secret key determined for the seal pattern; (c) setting a period at which the watermark is inserted into the sampling audio signal; And (d) embedding the watermark in the embedding period in the sampling audio signal; A method of embedding an image watermark in an audio signal, characterized by embedding a seal image in the original audio signal as a watermark. The method of claim 1, wherein step (b) And inserting a frame synchronization signal indicating that the watermark is at a position where the seal pattern starts, thereby generating a watermark. The method of claim 1, wherein step (b) And inserting a line sync signal indicating a line feed at a line feed position of the seal pattern to generate a watermark. The method of claim 1, wherein step (b) (b1) converting bit string data having a predetermined value according to the seal pattern; (b2) randomly generating bit string data according to a secret key determined for the seal pattern; And and (b3) generating a watermark for the seal pattern by combining the bit string data according to the seal pattern and the bit string data according to the secret key. Way. The method of claim 1, wherein step (b) The seal pattern is converted into bit string data having a value of {+1, -1} according to each pixel value, and randomly generates bit string data having a value of {+1, -1} according to the secret key. And multiplying the bit string data of the seal pattern with the bit string data according to the secret key and assigning a predetermined weight to generate a watermark for the seal pattern. . The method of claim 5, wherein the weight is A method of embedding an image watermark in an audio signal, characterized in that the watermark is set in proportion to the size of audio sampling data to which the watermark is added. (a) sampling the original audio signal at a predetermined sampling period to generate a sampling audio signal; (b) generating a watermark corresponding to the seal image according to a seal pattern generated from a predetermined seal image and a secret key determined for the seal pattern; (c) setting a period at which the watermark is inserted into the sampling audio signal; And (d) a computer-readable recording medium having recorded thereon a program for realizing a method of embedding an image watermark in an audio signal comprising the step of embedding said watermark in said embedding period unit in said sampling audio signal. In the method for extracting a watermark from the audio signal of the inspection object, (a) randomly generating bit string data according to a secret key determined for the original audio signal corresponding to the inspection target audio signal; (b) generating an approximate audio signal by obtaining a value of each sample data in the inspection target audio signal according to a predetermined number of sample data in its vicinity; (c) comparing the approximate audio signal with the inspection target audio signal and extracting a watermark included in the inspection target audio signal; And (e) restoring a seal pattern included in the inspection target audio signal by combining the extracted watermark and the bit string data obtained in the operation (a). How to extract it. The method of claim 8, and (f) a filtering step of setting a mask window having a predetermined size in the restored seal pattern and resetting the seal pattern according to the pixel value included therein to improve the sharpness of the image. A method of extracting a watermark from a target audio signal. The method of claim 8, and (g) determining a correlation coefficient based on the number of bits coincident between the seal pattern inserted into the watermark and the seal pattern restored in step (e). How to extract it. In the method for extracting a watermark from the audio signal of the inspection object, (a) randomly generating bit string data according to a secret key determined for the original audio signal corresponding to the inspection target audio signal; (b) generating an approximate audio signal by obtaining a value of each sample data in the inspection target audio signal according to a predetermined number of sample data in its vicinity; (c) comparing the approximate audio signal with the inspection target audio signal and extracting a watermark included in the inspection target audio signal; And (e) restoring a seal pattern included in the inspection target audio signal by combining the extracted watermark and the bit string data obtained in the operation (a), and extracting a watermark from the inspection target audio signal. A computer-readable recording medium having recorded thereon a program for realizing this.