KR101361108B1 - Dt-cwt watermarking system and method of dibr 3d image - Google Patents

Abstract

The present invention relates to a watermarking system for a DIBR 3D image copyright protection and a method thereof, which includes: a watermark inserting unit (100) which divides an image into partial images and inserts a watermark through a quantization which is caused by a three-level dual tree complex wavelet transform (DT-CWT) about each partial image; and a watermark detecting unit (200) which extracts a watermark which is inserted into an image using a statistical difference which is generated by the quantization of a coefficient between DT-CWT coefficient sub-bands. The present invention like the aforementioned has an effect of preventing an illegal distribution of DIBR 3D content in advance by: dividing a central image into left and right eye images, applying a three-level DT-CWT about each of the partial images, and inserting a watermark by performing the quantization about a coefficient group of DT-CWT; and detecting the inserted watermark by utilizing the statistical difference which is generated by the quantization of the coefficient between the DT-CWT coefficient sub-bands.

Description

Watermarking system and its method for copyright protection of DVD image 3DDT-CWT WATERMARKING SYSTEM AND METHOD OF DIBR 3D IMAGE

The present invention relates to a watermarking system and method for copyright protection of DIBR 3D video, and more particularly, to the central and depth images distributed for copyright protection of DIBR 3D content, as well as the central image and the left eye, which are valuable as 2D content. The present invention relates to a technique for providing a Dual-Tree Complex Wavelet Transform (DT-CWT) domain watermarking technique through quantization suitable for DIBR 3D images in consideration of leakage of images and right eye images.

In line with the revival of the 3D industry, interest in 3D content has naturally increased, and this 3D content can be coded in two ways.

The most intuitive way is to capture and store left and right eye images using two different cameras. This method, called Stereo Image Recording (SIR), has the advantage that two cameras can act as human eyes, providing a high level of stereoscopic experience.

However, it is difficult to set the same brightness, contrast, height, etc. of the two cameras. In addition, since the viewer is watching the initial depth information as it is, it is difficult to adjust the depth sense for each individual, and the storage capacity and bandwidth is increased because the two images of the left and right eyes must be stored.

Another 3D content coding method is Depth-Image-Based Rendering (DIBR). Left and right eye images delivered to the viewer are generated in a DIBR method from a center image and a depth image. Unlike the method of storing left and right eye images, the DIBR method provides several advantages.

First, viewers can adjust the depth as they want. Second, because the depth image is composed of black and white and has many low frequency components, the compression efficiency is high. That is, storage and bandwidth can be lowered. Finally, since the DIBR type video has a center image and a depth image, it can also provide 2D images without any conversion, thus providing backward compatibility with 2D display devices.

Due to many advantages of the DIBR method, researches on the DIBR system and its application technology have been actively conducted. In particular, the rapid expansion of the 3D content market has increased the interest in the copyright protection problem of the contents, and accordingly, copyright protection technology using digital watermarking has been studied.

1 illustrates a scenario in which 3D content may be illegally distributed. First, left and right eye images generated from a center image and a depth image may be simultaneously leaked. In addition, since each of the left and right eye images is in itself a value as a 2D image, this should also be taken into account in copyright protection. Lastly, the central image may be illegally distributed as a 2D image.

Many digital watermarking techniques for 2D and SIR 3D images have been studied (E. Halici and A. Alatan, "Watermarking for depth-image-based rendering," IEEE Int. Conf. Image Process. ), 2009, pp. 4217-4220, 2009.).

However, the above techniques cannot be applied to DIBR 3D images. First, since the pixels of the center image are partially shifted laterally according to the corresponding depth information, left and right eye images are generated, and the conventional method that does not consider such an attack is less robust to the pixel movement.

In addition, the image containing depth information is often subjected to appropriate deformation to make the synthesis of the left and right eye images more natural. Conventional digital watermarking techniques are problematic because they do not take the robustness into this situation.

In addition, several digital watermarking techniques have been studied in consideration of the characteristics of DIBR 3D images, and watermark patterns are added to the image by adding specific weights, and detection is performed through correlation analysis using projection matrices. It has been published in many papers.

However, this method has many limitations in practical use as a method that can be applied only when the original image is always needed and only one object exists in the image.

In another paper, watermarking was inserted in the DCT region of the central image (YH Lin and JL Wu, "A Digital Blind Watermarking for Depth-Image-Based Rendering 3D Images," IEEE Trans.Broadcasting., Vol. 57, no. 2, pp. 602-611, June 2011.).

In this paper, we calculate how much the pixels move in the left eye and right eye image in advance by using the depth image, so that the watermark inserted after the conversion can be successfully detected.

That is, three watermark packets were prepared and inserted for the center image, the left eye image, and the right eye image. Through this method, the watermark pattern was successfully detected from the center image and the left and right eye images. It also showed robustness in JPEG compression and noise insertion.

However, it is not possible to secure the robustness of the geometric transformation because it does not take into account the geometric transformation that occurs during the dissemination of images. In addition, due to the preprocessing of the depth image and the arbitrary control of the depth of the viewer, the robustness to the deformation cannot be secured.

The present invention has been made to solve the above problems, divides the center image into left and right eye images, applies a three-level dual tree complex wavelet transform (DT-CWT) to each partial image, DT-CWT The purpose is to prevent illegal distribution of DIBR 3D content by inserting a watermark by performing quantization on a coefficient group of.

In addition, an object of the present invention is to prevent illegal distribution of DIBR 3D content by detecting a watermark inserted by utilizing statistical characteristic differences generated by quantization of coefficients between coefficient subbands of DT-CWT.

Watermarking system for copyright protection of DIBR 3D image according to the present invention for achieving the technical problem, divides the image into partial images, 3-level Dual Tree Complex Wavelet Transform (DT-CWT) for each partial image A watermark embedding unit 100 for embedding the watermark through quantization by the quantization method; And a watermark detector 200 for extracting a watermark embedded in an image by using a statistical difference generated by quantization of coefficients between DT-CWT coefficient subbands.

개의 계수로 모델링하여 워터마크가 삽입되지 않은 이미지에 대해 false positive 오차를 검출하는 (c) 단계;를 포함한다.In addition, in the watermarking method for copyright protection of the DIBR 3D image based on the system as described above, the watermark inserting unit 100 divides the image into partial images, and 3-level DT for each partial image. (A) inserting a watermark through quantization according to Dual Tree Complex Wavelet Transform (CWT); (B) extracting, by the watermark detection unit, the watermark embedded in the image by using the statistical difference generated by the quantization of coefficients between DT-CWT coefficient subbands; And 10 randomly selected watermark images to measure false positive errors generated by the error analyzer 300 according to the threshold value set by the watermark detector 200.

And (c) detecting false positive errors with respect to the image in which the watermark is not inserted by modeling the coefficients.

According to the present invention as described above, the center image is divided into left and right eye images, 3-level dual tree complex wavelet transform (DT-CWT) is applied to each partial image, and quantization is performed on a coefficient group of DT-CWT. By inserting a watermark, there is an effect of preventing illegal distribution of DIBR 3D content.

In addition, the present invention has an effect of preventing illegal distribution of DIBR 3D content by detecting a watermark inserted by utilizing statistical characteristic differences generated by quantization of coefficients between coefficient subbands of DT-CWT.

1 is an exemplary diagram showing an illegal distribution path scenario of a conventional DIBR 3D content.
2 is a block diagram showing a watermarking system for copyright protection DIBR 3D video according to the present invention.
3 is a block diagram illustrating a watermark insertion unit of a watermarking system for copyright protection of the DIBR 3D image according to the present invention.
Figure 4 is a block diagram showing a watermark detection unit of the watermarking system for copyright protection DIBR 3D video according to the present invention.
Figure 5 is a block diagram showing an error analysis unit of the watermarking system for copyright protection DIBR 3D video according to the present invention.
6 is a diagram illustrating a histogram and an estimated distribution of protons and errors of a watermarking system for copyright protection of a DIBR 3D image according to the present invention.
7 is a diagram illustrating a BER histogram and a probability density function of an image without a watermark of a watermarking system for copyright protection of a DIBR 3D image according to the present invention.
8 is a view showing 12 pairs of center image and depth image used in the experiment of the watermarking system for DIBR 3D image copyright protection according to the present invention.
FIG. 9 is a diagram illustrating an average BER of a watermark bit string for a modified right eye image as a result of an experiment of a watermarking system for copyright protection of a DIBR 3D image according to the present invention. FIG.
10 is a view showing the change in image quality according to the noise level of the watermarking system for copyright protection DIBR 3D image according to the present invention.
FIG. 11 is a diagram illustrating an experimental result of an affine transform of a watermarking system for copyright protection of a DIBR 3D image according to the present invention. FIG.
12 is a view showing a BER value according to the change of each baseline distance value of the watermarking system for copyright protection of DIBR 3D image according to the present invention.
FIG. 13 is a diagram illustrating an original depth image and a depth image after asymmetric smoothing of a watermarking system for copyright protection of a DIBR 3D image according to the present invention; FIG.
14 is a flowchart illustrating a watermarking method for copyright protection of a DIBR 3D image according to the present invention;
15 is a flowchart illustrating a detailed process of step S10 of the watermarking method for copyright protection of DIBR 3D image according to the present invention.
FIG. 16 is a flowchart illustrating a detailed process of step S20 of a watermarking method for copyright protection of a DIBR 3D image according to the present invention; FIG.

Specific features and advantages of the present invention will become more apparent from the following detailed description based on the accompanying drawings. Prior to this, terms and words used in the present specification and claims are to be interpreted in accordance with the technical idea of the present invention based on the principle that the inventor can properly define the concept of the term in order to explain his invention in the best way. It should be interpreted in terms of meaning and concept. It is to be noted that the detailed description of known functions and constructions related to the present invention is omitted when it is determined that the gist of the present invention may be unnecessarily blurred.

As shown in FIG. 2, the watermarking system S for copyright protection of the DIBR 3D image according to the present invention divides the center image into partial images and 3-level DT-CWT (Dual Tree) for each partial image. Watermark inserting unit 100 for inserting the watermark through quantization according to the Complex Wavelet Transform, and water for extracting the watermark embedded in the image using the statistical difference generated by the quantization of coefficients between DT-CWT coefficient subbands The mark detection unit 200 is configured.

Hereinafter, the watermark insertion unit 100 of the watermarking system S for copyright protection of the DIBR 3D image according to the present invention will be described with reference to FIG. 3.

개의 부분 블록

로 분할한다.First, the image block dividing module 110 of the watermark inserting unit 100 detects the center of the image to be inserted into the watermark.

Part blocks

.

와

는 중앙 영상의 넓이와 높이이다. 따라서

와

는 각 부분 블록의 넓이와 높이이다.here,

Wow

Is the width and height of the center image. therefore

Wow

Is the width and height of each partial block.

In addition, the three-level DT-CWT module 120 inserts a feature point for each partial block through a three-level DT-CWT (Dual Tree Complex Wavelet Transform), but it is a two-level or three-level that gives robustness to the compression attack. A low frequency filter and noise are added using the coefficients of the level.

)을 수평적 성분, 수직적 성분 및 대각 성분을 갖는 (

), (

) 및 (

)로 쌍을 이루도록 구성한다.In addition, the pairing module 130 includes six subbands (

) With horizontal, vertical and diagonal components

), (

) And (

) To be paired.

로 명명한다.Here, the reason for grouping by each component is that the watermark embedding method through the quantization method according to the present invention to make a statistical difference between the two subbands in each pair, hereinafter, in the present invention

.

로 변환한다.In addition, the message coding module 140 uses the secret key value K to insert the M × N length message to be inserted into the watermark sequence.

.

In addition, the quantization module 150 compares each line of the DT-CWT in the DT-CWT region between pairs and inserts a watermark through quantization, where each pair is robust to DIBR and various geometric transformation attacks, and is invisible. Select in consideration of characteristics to increase.

First, a row of DT-CWT is selected as a quantization object. Since each pixel moves horizontally in the DIBR process, when the column is selected, a lot of changes are made to the inserted watermark, so the change is minimized by selecting a row of coefficients as quantization targets.

과

만을 사용하는데, 이는 수직성분에는 DIBR 과정에서 많은 변화가 가해지므로 수평성분을 다수 포함하고 있는 두 개의 쌍을 양자화 대상 쌍으로 선정한다. Next, out of the three pairs that were configured at each level

and

Only the vertical component is used. Since many changes are made in the DIBR process, two pairs containing a large number of horizontal components are selected as quantization target pairs.

와 phase

는 다음의 [수학식 1] 및 [수학식 2]와 같이 계산된다.At this time, the magnitude of the DT-CWT coefficient for the partial block of the selected center image

With phase

Is calculated as in Equations 1 and 2 below.

[Equation 1]

&Quot; (2) "

의 DT-CWT 계수의 magnitude j번째 행을

로 도출하여

의

와

에 대해 삽입하고자 하는 워터마크 비트가 0이면

를 양자화 하고, 비트가 1이면

를 양자화 한다.Followed by each sub-station

The magnitude jth row of the DT-CWT coefficients of

Derived by

of

Wow

If the watermark bit you want to insert for is 0

Quantize, if bit is 1

Lt; / RTI >

의 와

에 대해 삽입 비트가 0이면

를 양자화 하고, 1이면

를 양자화 하며, 주어진 워터마크 비트

와 해당하는 부분 블록

, 그리고

,

에 대한 양자화 과정은 아래의 Algorithm 1과 같다.Likewise

of Wow

If the insert bit is 0 for

Quantize, if 1

Quantizes a given watermark bit

And corresponding partial blocks

, And

,

The quantization process for is given by Algorithm 1 below.

은 양자화 에러이며,

는 정밀한 양자화를 위한 magnitude의 가중치이고,

는 좌표 j, k위치에 양자화된 magnitude값을 나타내며,

은 너무 큰 화질 손상을 방지하기 위한 최대 양자화 레벨이고,

은 부대역간 충분한 통계적 차이를 위한 에러값의 임계치이며,

,

,

는 성능과 비가시성을 고려하여 실험적으로 선정된 값이다.At this time,

Is the quantization error,

Is the weight of magnitude for precise quantization,

Represents magnitude values quantized at coordinates j and k,

Is the maximum quantization level to prevent too much image quality damage,

Is the threshold of error value for sufficient statistical difference between subbands,

,

,

Is an experimentally selected value considering performance and invisibility.

The partial block synthesizing module 160 applies the inverse DT-CWT to the quantized DT-CWT coefficients for watermark insertion, reconstructs the watermark-embedded image into a partial block, and combines the partial blocks into a block image. Restores a single image.

Hereinafter, the watermark detection unit 200 of the watermarking system S for copyright protection of the DIBR 3D image according to the present invention will be described with reference to FIG. 4.

개의 부분 블록

로 분할한다.First, the image block dividing module 210 of the watermark detection unit 200 detects the center of the image that is the watermark detection target.

Part blocks

.

와

는 중앙 영상의 넓이와 높이이다. 따라서

와

는 각 부분 블록의 넓이와 높이이다.here,

Wow

Is the width and height of the center image. therefore

Wow

Is the width and height of each partial block.

In addition, the three-level DT-CWT module 220 extracts feature points for each partial block through a three-level dual tree complex wavelet transform (DT-CWT), but uses low-frequency using two- or three-level coefficients. Detect filters and noise.

)을 수평적 성분, 수직적 성분 및 대각 성분을 갖는 (

), (

) 및 (

)로 쌍을 이루도록 구성한다.In addition, the pairing module 230 includes six subbands (

) With horizontal, vertical and diagonal components

), (

) And (

) To be paired.

로 변환한다.In addition, the message coding module 240 uses the secret key value K to extract the M × N length message to be watermarked.

.

과

을 양자화 대상의 쌍으로 선정한다.In addition, the quantization module 250 performs quantization by comparing pairs of DT-CWTs in a DT-CWT region between pairs, selecting a row of DT-CWT as a quantization target, and selecting from three pairs configured at each level. Two pairs containing multiple horizontal components

and

Is selected as a pair of quantization targets.

와 phase

는 전술한 바와 같이 [수학식 1] 및 [수학식 2]와 같이 계산된다.At this time, the magnitude of the DT-CWT coefficient for the partial block of the selected center image

With phase

Is calculated as shown in [Equation 1] and [Equation 2] as described above.

의 DT-CWT 계수의 magnitude j번째 행을

로 도출하여

의

와

에 대해 삽입하고자 하는 워터마크 비트가 0이면

를 양자화 하고, 비트가 1이면

를 양자화 한다.In addition, each sub-station

The magnitude jth row of the DT-CWT coefficients of

Derived by

of

Wow

If the watermark bit you want to insert for is 0

Quantize, if bit is 1

Lt; / RTI >

의

와

에 대해 삽입 비트가 0이면

를 양자화 하고, 1이면

를 양자화 하며, 주어진 워터마크 비트

와 해당하는 부분 블록

, 그리고

,

에 대해 양자화 과정은 전술한 Algorithm 1과 동일하다.Likewise

of

Wow

If the insert bit is 0 for

Quantize, if 1

Quantizes a given watermark bit

And corresponding partial blocks

, And

,

The quantization process for is the same as Algorithm 1 described above.

에 주어진 양자화 오차

,

를 검출하고, 각 행의 양자화 오차를 합한 두 계수 부대역의 양자화 오차를 비교하여 부분 블록에 삽입되었던 워터마크 비트를 검출한다.The watermark extraction module 260 divides the divided partial block.

Quantization error given in

,

And detect the watermark bits inserted in the partial block by comparing the quantization errors of the two coefficient subbands, the sum of the quantization errors of each row.

At this time, the detection of the watermark bit is detected through Algorithm 2 below.

은 레벨

에서의 총 행의 개수이다. 워터마크 삽입 과정 중 하나의 쌍에서 둘 중 하나의 계수 행은 양자화 되기 때문에, 두 계수 부대역간에 통계적 차이가 존재한다. 따라서 각 행의 양자화 오차를 합한 두 계수 부대역의 양자화 오차를 비교하여 어느 쪽 계수 부대역이 양자화 되었는지 판단하며, 이때 기준은 false positive 오차를 고려하여 설정된

과

를 임계값으로 한다.Where j is the index of row,

Silver level

The total number of rows in. Since one of the coefficient rows in one pair of watermark embedding processes is quantized, there is a statistical difference between the two coefficient subbands. Therefore, by comparing the quantization error of the two coefficient subbands, the sum of the quantization errors of each row, to determine which coefficient subband is quantized, and the criterion is set in consideration of the false positive error.

and

Is the threshold value.

가 오른쪽 부대역에서의 양자화 되었다고 판단되어진 행의 개수

보다 많은 경우, 해당 워터마크 비트

는 0이라고 추정하고, 반대의 경우에는 1로 추정한다.The number of rows in the coefficient subband that are considered quantized in the left subband

The number of rows that are considered to have been quantized in the right subband

If more, the corresponding watermark bit

Is assumed to be 0 and vice versa to 1.

(

)를 추출한다. 그리고 마지막으로 Bit Error Rate(BER)를 계산하여 최종적으로 검출 여부를 판별하며, BER은 [수학식 3]을 통해 계산된다.This process is applied to all partial blocks so that the final watermark information

(

). Finally, the bit error rate (BER) is calculated to finally determine whether the detection is performed, and the BER is calculated through Equation 3.

&Quot; (3) "

Meanwhile, the watermarking system S for copyright protection of the DIBR 3D image according to the present invention further includes an error analyzer 300 as shown in FIG. 5.

과

를 임계값에 따라 발생하는 false positive 오류를 계측하기 위해 10개의 무작위로 선택된 워터마크가 삽입 안 된 이미지로부터

개의 계수로 모델링하여 워터마크가 삽입되지 않은 이미지에 대해

를 기준으로 false positive 오차를 검출한다.The error analyzer 300 is set by the watermark extraction module 260.

and

From 10 randomly selected watermarked non-inserted images to measure false positive errors that occur according to thresholds

For images without watermarks by modeling with

Detect false positive errors based on

In this case, the difference between two unquantized coefficients in each pair was calculated by the same method as the proposed watermark detection process.

The histogram and the estimated distribution for quantum and error are shown in FIG. 6, and the quantization error is modeled as the sum of the Gaussian distribution of Equation 4.

&Quot; (4) "

이고,here,

ego,

이며,

Lt;

이다.

to be.

(양자화 되지 않은 행이 양자화 되었다고 판단되는 경우)는 다음의 [수학식 5]와 같이 도출할 수 있다.In addition, false positive error from [Equation 4]

(If it is determined that the non-quantized row is quantized) can be derived as shown in Equation 5 below.

&Quot; (5) "

는 j번째 행의 양자화 오차이고,

와

는 양자화 되지 않은

과

에 소속된 행이 양자화 되었다고 잘못 판단되는 false positive 오류이다.here,

Is the quantization error of the jth row,

Wow

Is not quantized

and

A false positive error that is incorrectly determined to have been quantized.

라고 가정할 때, 워터마크가 삽입되지 않은 부분 블록에 대해 특정한 비트가 잘못 검출될 false positive 오류

는 [수학식 6]과 같이 도출할 수 있다.At this time, false positive error

, A false positive error that causes certain bits to be detected incorrectly for partial blocks that do not contain a watermark.

Can be derived as shown in [Equation 6].

&Quot; (6) "

은 부분블록의 행의 개수이고, i는 본 발명의 알고리즘에 의해 양자화 되지 않았다고 판별되는 행의 개수이며, k는

이다.here,

Is the number of rows of the partial block, i is the number of rows determined not to be quantized by the algorithm of the present invention, and k is

to be.

는 [수학식 7]과 같이 도출된다.Finally, the probability that a bit string is detected for an image without watermark embedding

Is derived as shown in [Equation 7].

[Equation 7]

은 워터마크 비트열의 길이이고,

은 의미 있는 BER를 판별하는 임계값이다.here,

Is the length of the watermark bit string,

Is a threshold for determining meaningful BER.

를 기준으로 false positive 오차를 계산할 수 있다.For an image without a watermark inserted by Equation 7

The false positive error can be calculated based on.

In order to verify the theoretical error calculated above, we measured the BER of 1305 non-watermarked images.

이다. 도 7에 도시된 바와 같이 본 발명을 통해 도출한 수식과 실험적으로 구한 오류가 거의 일치함을 알 수 있다.FIG. 7 is a diagram illustrating a probability density function of Equation 8 derived based on the BER histogram and Equation 7 of an image without a watermark inserted therein. here

to be. As shown in FIG. 7, it can be seen that the equations derived through the present invention are almost identical to the errors obtained experimentally.

[Equation 8]

Hereinafter, referring to FIGS. 8 to 13, the experimental results of the watermarking system for copyright protection of the DIBR 3D image according to the present invention are as follows.

In Lin's method, we experiment with a block size setting value of 16 × 16 (notation Lin's method *), and also compare the block size with 64 × 64 or 128 × 128 to compare the information with similar capacity of the present invention. (Lin's method **) was used for comparative experiments.

1. Experimental image and experiment setup values

FIG. 8 is a diagram illustrating 12 pairs of a center image and a depth image used in an experiment. All images were obtained from Middlebury Stereo Datasets and Microsoft Research 3D Video Datasets, with resolutions ranging from 450 × 375 to 1390 × 1110. The depth image is an 8 bit monochrome image.

,

,

를 각각 450, 8, 2로 설정하였다. 각 부분 블록의 크기는

로 하였다. 여기서,

와

는 각각 중앙 영상의 넓이와 높이이다. 시청자의 시청 편의를 고려하여 baseline distance

는 이미지 넓이의 5%로 설정 하였다.

과

는 전술한 [수학식 4] 내지 [수학식 7]을 통하여 false positive 오류를 계산하여 정하였다. 워터마크가 삽입되지 않은 이미지에 대해 BER의 값이 0.75보다 클 확률을

으로 만드는 임계값

,

로 정하였다.
When inserting watermark considering the invisibility of watermark

,

,

Were set to 450, 8, and 2, respectively. The size of each partial block

It was set as. here,

Wow

Are the width and height of the center image respectively. Baseline distance for viewers' convenience

Was set to 5% of the image width.

and

Is calculated by calculating a false positive error through the above Equations [4] to [Equation 7]. For images without watermarks, the probability that the value of BER is greater than 0.75

Thresholds

,

Was set.

2. Invisibility Experiment

Subjective and objective invisibility experiments were performed based on the set values as described above. For subjective experiments, a 120-hertz 23-inch LG Platron full HD 3D monitor, NVIDIA GeForce GTX 460, and 3D Vision active shutter glass were used. Subjective experiments were assigned a score from 1 point (different from the original) to 5 points (difficult to distinguish from the original) using the Double Stimulus Continuous Quality Scale (DSCQS) method. Subjective experiments were carried out by ten people, a mix of experts and non-experts. Table 1 shows the subjective visibility test results for the present invention and Lin's method. Both methods showed good results.

[Table 1]

Objective invisibility experiments used mean PSNR and mean structure similarity (SSIM). Table 2 shows the results. Although the present invention showed a slightly lower result than Lin's method, the result of the present invention also shows a level that is generally indistinguishable from the original.

[Table 2]

3. Toughness Experiments on Signal and Geometry Deformation

For 12 pairs of experimental images, the watermark was inserted into the center image, and left and right eye images were generated from the center image and the corresponding depth image with the watermark inserted by the DIBR method.

As shown in [Table 3], the result of comparing the BER values of the center image, the generated left eye image, and the generated right eye image, it can be confirmed that Lin's result value is better. However, it can be seen that the results of the method proposed by the present invention are more prevalent for the left eye and right eye images.

[Table 3]

In addition to the center image, one of the generated left eye and right eye images, or left eye and right eye images may be leaked to generate copyright infringement. Therefore, robustness must be ensured even when the signal is generated from left or right eye images and undergoes various signal and geometric transformations during distribution.

According to the present invention, experiments were conducted on JPEG compression, noise addition, median filtering, reduction and enlargement, and rotation, and all deformations were performed by the Stirmark benchmark tool.

From Figure 9 it can be seen that the experimental results of the present invention shows a lower BER than the experimental results of Lin. However, in the case of noise addition, Lin's result was better when the noise level exceeded 2. However, if the noise level is 3 or more, the PSNR is 20.72dB compared to the original, and the deterioration in image quality is very large as shown in FIG. 10. Since such large image degradation rarely occurs in the Internet environment, a high noise level is not significant compared to other result values.

In order to consider more various geometric attacks, we have tested the robustness of the affine transform as shown in [Equation 9].

&Quot; (9) "

The matrix applied to [Equation 9] is as shown in [Equation 10] below.

&Quot; (10) "

9 is a diagram illustrating an average BER of a watermark bit string for a modified right eye image, (a) JPEG compression, (b) noise addition, (c) median filtering, and (d) zooming in and out , (e) is a rotation, and (f) is a diagram according to cutting after rotation.

10 is a view showing the change in image quality according to the noise level, (a) is noise level 3, PSNR is 20.72dB. (B) the noise level 7, PSNR is 13.62dB.

FIG. 11 is a diagram illustrating an experimental result for an affine transform. The present invention records low BER in all cases.

4. Robustness Experiment for Baseline Distance Adjustment

를 적당한 범위 안에서 조절 할 수 있다. 본 실험에서는 baseline distance를 영상 넓이의 3%~7%로 변화시켜 가며 실험을 하였다. 도 12는 각 baseline distance 값의 변화에 따른 BER값을 나타낸다. 본 발명의 방법은 다른 방법들에 비해 확연히 낮은 BER을 유지함을 알 수 있다.
One of the advantages of the DIBR system is that the viewer can freely adjust the depth as desired. That is, baseline distance according to individual taste and viewing distance

Can be adjusted within the proper range. In this experiment, we changed the baseline distance from 3% to 7% of the image width. 12 shows the BER value according to the change of each baseline distance value. It can be seen that the method of the present invention maintains significantly lower BER than other methods.

5. Robustness Experiment for Preprocessing Depth Image

Depth images can be pre-processed in the viewer's DIBR system to produce natural left and right eyes. For the experiment, asymmetric smoothing was applied to the experimental depth image as shown in FIG. 13.

As a result of the watermark detection experiment from the left and right eye images generated by the asymmetric smoothed depth image, the algorithm proposed by the present invention confirmed low BER.

Hereinafter, referring to FIG. 14, a watermarking method for copyright protection of a DIBR 3D image according to the present invention will be described below.

First, the watermark inserting unit 100 divides an image into partial images and inserts a watermark for each partial image through quantization according to a 3-level dual tree complex wavelet transform (DT-CWT) (S10). ).

Next, the watermark detection unit 200 extracts the watermark embedded in the image by using the statistical difference generated by the quantization of the DT-CWT coefficients (S20).

개의 계수로 모델링하여 워터마크가 삽입되지 않은 이미지에 대해

를 기준으로 false positive 오차를 검출한다(S30).
Then, the error analysis unit 300 from the image that is not inserted 10 randomly selected watermark to measure the false positive error that occurs according to the threshold value set in the watermark detection unit 200

For images without watermarks by modeling with

Detecting a false positive error based on (S30).

Hereinafter, referring to FIG. 15, the detailed process of step S10 of the watermarking method for copyright protection of the DIBR 3D image according to the present invention is as follows.

개의 부분 블록

로 분할한다(S11).First, the image block dividing module 110 of the watermark inserting unit 100 selects the center of the image to be inserted into the watermark.

Part blocks

It is divided into (S11).

Subsequently, the three-level DT-CWT module 120 inserts a feature point for each partial block through a three-level dual tree complex wavelet transform (DT-CWT), which is a two-level or three-level method that provides robustness to the compression attack. The low frequency filter and the noise are added using the coefficient of the level (S12).

)을 수평적 성분, 수직적 성분 및 대각 성분을 갖는 (

), (

) 및 (

)로 쌍을 이루도록 구성한다(S13).Subsequently, the pairing module 130 has six subbands in each level (

) With horizontal, vertical and diagonal components

), (

) And (

It is configured to be paired with (S13).

로 변환한다(S14).Subsequently, the message coding module 140 uses the secret key value K to insert the M × N length message to be inserted into the watermark sequence.

(S14).

Subsequently, the quantization module 150 compares each line of the DT-CWT in pairs in the DT-CWT region and inserts a watermark through quantization (S15).

The partial block synthesizing module 160 applies the inverse DT-CWT to the quantized DT-CWT coefficients to insert the watermark, reconstructs the watermark-embedded image into the partial block, and combines the partial blocks into the block image. Restore to a single image (S16).

Hereinafter, referring to FIG. 16, a detailed process of step S20 of a watermarking method for copyright protection of a DIBR 3D image according to the present invention will be described.

개의 부분 블록

로 분할한다(S21).After operation S10, the image block dividing module 210 of the watermark detection unit 200 detects the center of the image that is the watermark detection target.

Part blocks

It is divided into (S21).

Subsequently, the 3-level DT-CWT module 220 extracts the feature points through the 3-level dual tree complex wavelet transform (DT-CWT) for each partial block, but uses low-frequency using 2 or 3 levels of coefficients. The filter and the noise are detected (S22).

)을 수평적 성분, 수직적 성분 및 대각 성분을 갖는 (

), (

) 및 (

)로 쌍을 이루도록 구성한다(S23).Subsequently, the pairing module 230 includes six subbands (

) With horizontal, vertical and diagonal components

), (

) And (

It is configured to be paired with (S23).

로 변환한다.(S24).Subsequently, the message coding module 240 uses the secret key value K to extract the M × N length message to be extracted.

(S24).

Subsequently, the quantization module 250 performs quantization by comparing pairs of DT-CWTs in a DT-CWT region between pairs (S25).

에 주어진 양자화 오차

,

를 검출하고, 각 행의 양자화 오차를 합한 두 계수 부대역의 양자화 오차를 비교하여 부분 블록에 삽입되었던 워터마크 비트를 검출한다(S26).
The watermark extraction module 260 divides the divided partial block.

Quantization error given in

,

Next, the quantization error of the two coefficient subbands including the quantization error of each row is compared to detect the watermark bit inserted in the partial block (S26).

While the present invention has been particularly shown and described with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. It will be appreciated by those skilled in the art that numerous changes and modifications may be made without departing from the invention. And all such modifications and changes as fall within the scope of the present invention are therefore to be regarded as being within the scope of the present invention.

S: Watermarking System for DIBR 3D Video Copyright Protection
100: watermark insertion unit 110: image block segmentation module
120: 3-level DT-CWT module 130: pairing module
140: message coding module 150: quantization module
160: partial blow synthesis module 200: watermark detection unit
210: video block segmentation module 220: 3-level DT-CWT module
230: pairing module 240: message coding module
250: quantization module 260: watermark extraction module
300: error analysis unit

Claims (7)

In the watermarking system for DIBR 3D video copyright protection,
A watermark inserting unit 100 for dividing an image into partial images and inserting a watermark for each partial image through quantization using a 3-level dual tree complex wavelet transform (DT-CWT); And
Watermarking system for copyright protection DIBR 3D image comprising a; watermark detection unit 200 for extracting the watermark embedded in the image by using the statistical difference generated by the quantization of the coefficients between the DT-CWT coefficient subband . The method of claim 1,
The watermark inserting unit 100,
The center of the image to be watermarked

Part blocks

A video block dividing module (110) for dividing into;
For each partial block, feature points are inserted through a 3-level dual tree complex wavelet transform (DT-CWT), but low-frequency filters and noise are added using two or three levels of coefficients that provide robustness to compression attacks. Three-level DT-CWT module 120;
6 substations in each level (

) With horizontal, vertical and diagonal components

), (

) And (

Pairing module 130 configured to be paired with;
Watermark sequence using the secret key value K for M × N length message to be inserted

Message coding module 140 to convert to;
A quantization module 150 for comparing pairs of DT-CWTs in a DT-CWT region between pairs and inserting a watermark through quantization; And
Partial block synthesis module for restoring a watermark-embedded image into a partial block by applying an inverse DT-CWT to quantized DT-CWT coefficients for embedding a watermark, and reconstructing the block image into a single image by combining the partial blocks Watermarking system for copyright protection of DIBR 3D video, comprising: (160).
here,

Wow

Is the width and height of the center image,

Wow

Is the width and height of each partial block. The method of claim 1,
The watermark detection unit 200,
The image block dividing module 210 divides the center of the image to be watermark detected.

Part blocks

A video block dividing module 210 divided into;
3-level DT-CWT module extracts feature points through 3-level dual tree complex wavelet transform (DT-CWT) for each sub-block, and detects low frequency filters and noise using 2-level or 3-level coefficients 220;
6 substations in each level (

) With horizontal, vertical and diagonal components

), (

) And (

Pairing module 230 configured to be paired with;
Watermark sequence of M × N length message to be extracted by using secret key value K

Message coding module 240 to convert to;
A quantization module 250 for comparing and quantizing each row of the DT-CWT in a DT-CWT region between pairs; And
Split partial block

Quantization error given in

,

And a watermark extraction module 260 for comparing the quantization errors of the two coefficient subbands summing the quantization errors of each row to detect watermark bits inserted in the partial block. Watermarking system for copyright protection.
here,

Wow

Is the width and height of the center image,

Wow

Is the width and height of each partial block. The method of claim 1,
From 10 randomly selected watermarked non-inserted images to measure false positive errors that occur according to thresholds

The watermarking system for DIBR 3D image copyright protection, further comprising: an error analyzer 300 for detecting false positive errors with respect to an image without a watermark by modeling the coefficients. In the watermarking method for protecting DIBR 3D video copyright,
(a) the watermark inserting unit 100 dividing an image into partial images and inserting a watermark for each partial image through quantization according to a 3-level dual tree complex wavelet transform (DT-CWT) ;
(b) the watermark detection unit 200 extracting the watermark embedded in the image by using a statistical difference generated by quantization of coefficients between DT-CWT coefficient subbands; And
(c) From the image in which ten randomly selected watermarks are not inserted in order for the error analyzer 300 to measure false positive errors generated according to the threshold set by the watermark detector 200,

And detecting false positive errors with respect to an image in which the watermark is not inserted by modeling the number coefficients. 3. The method of claim 5, wherein
The step (a)
(a-1) The image block dividing module 110 of the watermark inserting unit 100 adjusts the center of the image to which the watermark is to be inserted.

Part blocks

Dividing into;
(a-2) inserting feature points by the three-level DT-CWT module 120 through the three-level dual tree complex wavelet transform (DT-CWT) for each partial block;
(a-3) The pairing module 130 has six subbands in each level (

) With horizontal, vertical and diagonal components

), (

) And (

Configuring in pairs;
(a-4) Watermark sequence of M × N length message to be inserted by message coding module 140 using secret key value K

;
(a-5) the quantization module 150 comparing pairs of DT-CWTs in a DT-CWT region between pairs and inserting a watermark through quantization; And
(a-6) The partial block synthesizing module 160 applies the inverse DT-CWT to the quantized DT-CWT coefficients for watermark embedding to restore the watermarked image into the partial block and combine the partial blocks. And restoring the block image to a single image by using a watermarking method for copyright protection of the DIBR 3D image. The method of claim 5, wherein
The step (b)
(b-1) The image block dividing module 210 of the watermark detection module 200 moves the center of the image targeted for watermark detection.

Part blocks

Dividing into;
(b-2) extracting, by the three-level DT-CWT module 220, a feature point through a three-level dual tree complex wavelet transform (DT-CWT) for each partial block;
(b-3) The pairing module 230 has six subbands in each level (

) With horizontal, vertical and diagonal components

), (

) And (

Configuring in pairs;
(b-4) Watermark sequence M × N length message to be extracted by message coding module 240 using secret key value K

;
(b-5) the quantization module 250 comparing and quantizing each row of the DT-CWT in pairs within the DT-CWT region; And
(b-6) The watermark extraction module 260 divides the divided partial block

Quantization error given in

,

And detecting the watermark bits inserted in the partial blocks by comparing the quantization errors of the two coefficient subbands, the sum of the quantization errors of the respective rows, and detecting watermark bits inserted into the partial blocks. Way.

Images