KR20180031904A - A Robust and Blind Watermarking Method for DIBR Using a Depth Variation Map - Google Patents

Abstract

The present invention relates to a robust blind watermarking method of a DIBR attack using a depth change map, which inserts a watermark into a texture and a depth image composed of a texture image and a depth image. The method includes (a) To generate a depth variation map (DVM); (b) generating a sub-band image by performing a 2-dimensional discrete wavelet transform (2DDWT) on the texture image; (c) sub-sampling the DVM to generate a DVM sub-image having the same resolution as the sub-band image, selecting an insertion region using pixel values in the DVM sub-image, Selecting an area of the image as a watermark embedding area to embed a watermark; (d) encrypting the watermark data; (e) inserting the watermark data into the watermark embedding area of the sub-band image; And (f) generating a watermarked texture image by performing inverse 2DDWT on sub-band images in which the watermarked sub-band image is replaced.
According to the above method, the non-visibility of the inserted watermark information is superior to the existing methods, and the bit error ratio (BER) is close to zero for the considered point attack. It has very good performance.

Description

[0001] The present invention relates to a robust watermarking method for a DIBR attack using a depth change map,

The present invention relates to a digital watermarking method for protecting intellectual property rights of a free-view 2D or 3D image by rendering a texture image received from a viewer side and an arbitrary viewpoint image using the depth image, A depth-change map is generated using the depth-of-view image, a three-level 2D discrete wavelet transform (2DDWT) is performed on the original image, and a position to be watermarked is determined with reference to a depth- , And robustness blind watermarking method of DIBR attack using depth change map.

That is, the present invention is a blind DW method that assumes that a pair of depth + texture is transmitted to a receiver and an arbitrary viewpoint movement is performed from the receiver to the DIBR, and the blind DW method The present invention relates to a robust blind watermarking method of a DIBR attack using a depth change map, which inserts multiple WMs into an LH subband that is most robust against attacks at each level by 2DDWT so as to find regions with depth images and robustness against various attacks.

In addition, the present invention extracts WM information from all the pixels in the LH subband while performing 2DDWT in the same manner as the insertion, extracts the depth map of the embedded WM position using the original WM information, The robustness of a DIBR attack using a blind watermarking method.

Development of digital image media technology not only develops images at high resolution but also develops 3D images (Non-Patent Document 1) and free viewpoint images (Non-Patent Document 2). Recent 3D images have been booming for 3 ~ 5 years due to the spark of the 'Avatar' in 2009, and after that they seemed to be less interested in it, but they have recently emerged again in the field of virtual reality. In addition, studies on multi-viewpoint 3D images [Non-Patent Document 3] and digital holograms [Non-Patent Document 4] have been made steadily. On the other hand, research on free view images has been studied since the early 2000s under the name of freeview TV (FTV), and the notion of free movement of viewpoint was stronger than 3D. However, recently, these concepts have been combined to develop the concept of free-view 3D [Non-Patent Document 5]. The FTV or free-view 3D allows viewers to freely select a viewpoint, but it can also be meaningful in terms of relieving the transmission burden of transmitting a super multi-view image for the viewer.

 FTV or free-view 3D is a concept that enables viewers to change the viewpoint as desired. In order to do this, it is necessary to render a viewpoint image by tracking the viewpoint of a viewer or allowing a viewer to directly select a viewpoint. One of the most widely used methods is depth-image-based rendering (DIBR), which is also called 3D warping [Non-Patent Document 1]. This method is a method of rendering an image at a desired point of time with a depth image having three-dimensional coordinate information. That is, for the FTV or free-view 3D, the RGB image of the current point and the depth image corresponding to the current point of view are sent to the receiver to render the image at the point of DIVR conversion at the receiver.

On the other hand, the development of video means that the video itself is gradually becoming information-implicit media, and gradually becomes high-value content. However, a solution to the problems of piracy and theft, which is a typical problem of digital contents, has become more urgent and up until now digital watermarking (DW) [Non-Patent Document 5] has been continuously studied . They have mainly studied intellectual property protection methods that combine invisibility and robustness of watermark (WM) inserted into various malicious / non-malicious attacks.

However, the above-described viewpoint movement on the receiving side changes the position of the original image pixels, generates occlusion regions and discocclusion regions, and significantly transforms or removes watermark (WM) information inserted in the original image Lays. That is, the viewpoint movement is another powerful attack on the image [Non-Patent Document 6]. Therefore, DW is being studied for a recent point-of-view attack.

The DW for viewpoint movement was first attempted in [Non-Patent Document 8]. Here, it is assumed that the viewpoint movement of the image-based rendering (IBR), that is, the light-ray rendering (LFR) method is assumed. The DW method uses WM for high-pass filtering when inserting WM, and extracts WM by extracting the NC (normalized correlation) value of the embedded WM and data at specific position by high-pass filtering the image. Therefore, this method is a non-blind method in which both the original image and the inserted WM information are used for extraction. On the other hand, Zhu et al. Proposed a method of inserting WM into the intermediate frequency band of 8 × 8 DCT by assuming the viewpoint movement of the DIBR and finding the farthest object with the smallest change in viewpoint movement as a flood fill method [ Non-Patent Document 9]. This method is also a non-blind method because it is necessary to find the insertion area by the flood fill method in WM extraction.

In addition, Lee et al. Proposed a method of preliminarily calculating the occlusion region after DIBR and inserting 64-bit WM in the avoidance of the region [Non-Patent Document 10]. The WM extraction method of this method performs Weiner noise elimination filtering on the attacked image and obtains the difference image with the pre-filtering image. If the NC value with the original WM in the DW area is larger than the predetermined value, WM is considered to be extracted Method. This method uses blind watermarking (blind DW) because it uses only the original WM information in WM extraction. The team then developed a method to extract the noise from the original image, and to divide the average value into 0 and divide it in the vertical direction, and insert 128-bit watermark information into each noise to extract the watermark with noise information Patent Document 11].

However, low WM extraction rate was still observed in the viewpoint shifted image. In order to increase the WM extraction rate after moving the viewpoint, there has been proposed a method of inserting WM in consideration of the distortion of the WM information in the viewpoint moving image [Non-Patent Document 12]. However, the WM extraction rate at the predicted point was high, but it was low at any other point. A dual-tree complex wavelet transform (DT-CWT) is proposed as a robust special transform method in DIBR, and a method of inserting 64-bit WM information in the form of quantization for each block of 3-level conversion result has been proposed 13]. Although this method shows good performance with almost constant error rate at 3 ~ 7% viewpoint change, it has a disadvantage that the possibility of loss of WM in left and right blocks increases according to the size of block and the amount of viewpoint movement.

Then, feature point-based watermarking method using SIFT has been proposed to find a watermark position in a blind manner at a virtual point of view [Non-Patent Document 14]. The non-visibility of this method is superior to other methods, but it is impossible to compare it with attacks targeting images that are moved to a point-in-time image that does not move from point-to-point JPEG attacks or average filtering attacks. In other words, it is supposed that an attack to remove WM from a viewpoint moving image on the receiving side should be targeted. However, in this method, it is assumed that an image that has not been moved is attacked. Also, a DW method has been proposed for depth images because of the importance of depth images in DIBR [Non-Patent Document 15]. This method has a good WM extraction rate without affecting the invisibility, but it has a fatal disadvantage that it can not protect the copyright of the midpoint or virtual view image, not the depth image. Since the depth image is not used after the virtual point of view is created, it can be illegally circulated to the virtual point of view itself.

Unseen visible watermarking (UVW) has been proposed to embed WM into a depth image and to extract WM from a viewpoint moved by DIBR [Non-Patent Document 16]. In this method, the WM is inserted using the magnitude of the disparity which does not affect the viewpoint movement in the DIBR, and when the WM is extracted, the rendering condition is changed so that the WM of the inserted depth image is highlighted. Therefore, this method is a non-blind method that requires depth image extraction. Rana et al. Proposed a method using DT-DWT, which is a real part of DT-CWT, and motion-compensated temporal DCT filtering (DCT-MCTF) [Non-Patent Document 17]. This method assumes that left and right view images are given and that depth and texture images are provided at each view point. As a result, it showed superior performance in nonvisuality and robustness compared to existing methods. Although it does not target the other viewpoint movement, it has also proposed a method of inserting a watermark by using 3D DWT and Jacket matrix in an anaglyph stereo image [Non-Patent Document 18].

Therefore, it is necessary to develop a blind (non-patent document 7) DW method considering a point-of-view attack as a main attack and an attack targeting DW of an existing 2D image.

 Diniel Minoli, 3DTV, Content capture, encoding and transmission, IEEE-Wiley, 2010.  Y. Mori, N. Fukushima, T. Yendo, T. Fujii, and M. Tanimato, "Image Generation for Image Processing", Image Processing 24 (2009) 65-72.  T-Y. Ho, D-N. Yang, and W. Liao, "Efficient resource allocation of mobile multi-view 3D videos with depth-image-based rendering ", IEEE Trans. on Mobile Computing, Vol. 14, No. 12, pp. 344-357, Feb. 2015.  Y-H. Seo, Y-H. Lee, J-M. Koo, J-S. Yoo, and D-W. Kim, "Digital holographic video service system for natural color scene ", Optical Engineering, 52 (11), 113106 (Nov. 2013).  Aljoscha Smoloc, "3D video and freeview video-From capture to display," ELsevier Pattern Recognition, 44 (2011) 1958-1968.  A. Smolic, et al., "Coding Algorithm for 3DTV-A sunrvey," IEEE Trans. On Circuits and Systems for Video Technology, Vol. 17, No. 11, pp. 1606-1621, Nov. 2007.  J. Cox, M. Miller, J. Bloom, T. Kalker, Digital Watermarking and Steganography, 2nd Ed., Elsevier, 2008.  A. Koz, C. Cigla, and A. Alatan, "Free-view watermarking for free-view television," ICIP 2006, pp. 1405-1408. 2006.  N. Zhu, G. Ding, and J. Wing, "A novel digital watermarking method for new viewpoint video based on depth map," Intl. Conf. on Intelligent Systems Design and Applications, pp. 3-7, 2008.  M-J. Lee, J-W. Lee, and H-K. Lee, "Perceptual watermarking for 3D stereoscopic video using depth information," IEEE Intl. Conf. Intelligent Information Hiding and Multimedia Signal Processing, pp. 81-84, 2001.  J-W. Lee, H-D Kim, H-Y. Choi, S-H. Choi, and H-K. Lee, "Stereoscopic watermarking by horizontal noise mean shift," Proc. of SPIE Vol. 8303, 830307-1, 2012.  Y-H. Lin and J-L Wu, "A digital blind watermarking for depth-image-based rendering 3D images," IEEE Trans. on Broadcasting, Vol. 57, No. 2, pp. 602-611, June 2011.  H-D Kim, and J-W Lee, "Robust DT-CWT Watermarking for DIBR 3D Images." IEEE Transactions on Broadcasting, Vol. 58, No. 4, pp. 533-543, Oct. 2012.  S. Wang, C. Cui, and X. Niu, "Watermarking for DIBR 3D images based on SIFT feature points," Elsevier Measurement 48 (2014) 54-62.  Y. Guan, Y. Zhu, X. Liu, G. Luo, Z. Sun, and L. Zhang, "A digital blind watermarking scheme based on quantization index modulation in depth map for 3D video," ICARCV, We43.6, 2014.  S-C Pei and Y-Y Wang, "Auxiliary metadata delivery in view synthesis using depth no synthesis error model," IEEE Trans. on Multimedia, Vol. 17, No. 1, pp. 128-133, Jan. 2015.  S. Rana and A. Sur, "3D video watermarking using DT-DWT to resist synthesis view attack," EUSIPCO, pp. 46-50. 20, 2015.  I. Prathap and R. Anitha, "Robust and blind watermarking scheme for three dimensional anaglyph images," Elsevier Computers and Electrical Engineering 40 (2014) 51-58.  Y. Gao. G. Cheung, T. Maugey, P. Frossard, and J. Liang, "Encoder-driven inpainting strategy in multiview video compression." IEEE Trans. on Image Processing, Vol. 25, No. 1, pp. 134-149, Jan. 2016.  "Robustness of DWT-based Digital Images by Resolution, Blind Watermarking," Journal of Broadcast Engineering, Vol. 20, No. 6, pp. 888-900, November 2015. (Y. Lee, DW Kim, YH Seo, "A Robust Blind Watermarking for Digital Image Using DWT," J. of Broadcast Engineering, Vol. 20, No. 6, pp. 888-900, Nov. 2015.)  W, Stalling, Cryptography and Network Security, 6th Ed., Pearson, 2014.   http://www.newwaveinstruments.com/resources/articles/ m_sequence_linear_feedback_shift_register_lfsr.htm  http://research.microsoft.com/en-us/downloads/5e4675af-03f4-4b16-b3bc-a85c5bafb21d/  http://vision.middlebury.edu/stereo/data/

SUMMARY OF THE INVENTION The object of the present invention is to solve the above problems, and it is an object of the present invention to create a depth change map using a depth image in a free-view 2D or 3D image, The present invention provides a robust blind watermarking method of a DIBR attack using a depth change map, which determines a position to be watermarked by referring to a depth change map.

According to another aspect of the present invention, there is provided a method of generating a watermarked image by inserting a watermark bit into each pixel using a linear quantizer, determining a size of a quantization step according to an energy value of each subband, To provide a robust blind watermarking method of a DIBR attack using a depth change map.

It is another object of the present invention to provide a method and apparatus for extracting possible candidates using a degree of correlation with an original watermark information in an attacked image when extracting a watermark after an attack including a viewpoint movement, The present invention provides a robust blind watermarking method of a DIBR attack using a depth change map in which a final watermark is extracted and determined.

In order to achieve the above object, the present invention provides a robust blind watermarking method of a DIBR attack using a depth change map for inserting a watermark into a texture and a depth image composed of a texture image and a depth image, Performing a differential in each of the horizontal and vertical directions to generate a depth variation map (DVM); (b) generating a sub-band image by performing a 2-dimensional discrete wavelet transform (2DDWT) on the texture image; (c) sub-sampling the DVM to generate a DVM sub-image having the same resolution as the sub-band image, selecting an insertion region using pixel values in the DVM sub-image, Selecting an area of the image as a watermark embedding area to embed a watermark; (d) encrypting the watermark data; (e) inserting the watermark data into the watermark embedding area of the sub-band image; And (f) generating a watermarked texture image by performing inverse 2DDWT on sub-band images in which watermarked sub-band images have been replaced.

Further, the present invention provides a robust blind watermarking method of a DIBR attack using a depth change map, wherein in the step (a), differentiations are performed in the horizontal direction and the vertical direction on the depth image, (DDM) by accumulating differential values within a predetermined range in a pixel, and generating a depth variation map (DVM) by averaging a predetermined area in each pixel of the DDAM image.

Further, the present invention is a robust blind watermarking method of a DIBR attack using a depth change map, wherein the depth image is differentiated after performing median filtering on the depth image in the step (a) .

According to another aspect of the present invention, there is provided a robust blind watermarking method for a DIBR attack using a depth variation map, wherein in the step (a), differential values of pixels within a predetermined range from the lower right in each pixel of the differentiated result image are cumulatively added And generates a differential cumulative map.

According to another aspect of the present invention, there is provided a robust blind watermarking method of a DIBR attack using a depth change map, wherein when the texture image is an RGB image, the RGB image is converted into a YCbCr image, .

According to another aspect of the present invention, there is provided a robust blind watermarking method of a DIBR attack using a depth change map, the method comprising the steps of: performing three levels of 2DDWT transformation on the texture image; inserting a watermark in the LH sub- .

According to another aspect of the present invention, there is provided a robust blind watermarking method for a DIBR attack using a depth change map, the method comprising the steps of: (c) Is selected as the insertion region, and the region including the pixel having the smallest pixel value in the remaining sub-images excluding the selected insertion region is selected as the insertion region.

In the robust blind watermarking method of the DIBR attack using the depth change map, the number of the watermark embedding areas may be set to be the same as the number of the watermark embedding areas So that the ratio of the number of pixels to be inserted into the pixel is less than or equal to a predetermined ratio.

In another aspect of the present invention, there is provided a robust blind watermarking method for a DIBR attack using a depth change map, wherein in step (d), the watermark data is XORed with the output of a linear feedback shift register (LFSR) .

According to another aspect of the present invention, there is provided a robust blind watermarking method of a DIBR attack using a depth change map, wherein the watermark data is composed of bit data, and in the step (e) The watermark is inserted by replacing the pixel value of the pixel in the insertion area with a value output by the linear quantizer.

According to another aspect of the present invention, there is provided a robust blind watermarking method for a DIBR attack using a depth change map, wherein a quantization step of the linear quantizer is determined according to a magnitude of a weighted energy standard deviation of each level subband image.

Further, the present invention is characterized in that in the robust blind watermarking method of the DIBR attack using the depth change map, the quantization step DELTA _L of the linear quantizer is obtained by the following equation (1).

[Equation 1]

,

,

L denotes a 2DDWT level, c _i denotes a pixel value of a subband, T _{L, l} and T _{L, h} denote a predetermined threshold value T _{L, 1} < T _{L, h} .

According to another aspect of the present invention, there is provided a robust blind watermarking method for a DIBR attack using a depth change map, wherein a size of one watermark embedding area is set to half the size of the watermark data, When inserting watermark data, the watermark data is divided in half and alternately inserted into the watermark embedding area.

The present invention also relates to a robust blind watermarking method of a DIBR attack using a depth change map for extracting a watermark from a watermark embedded texture and a depth image, dimensional discrete wavelet transform to generate a subband image; (h) encrypting the watermark data; (i) extracting watermark values in the sub-band image, and if the similarity between the watermark data and the extracted watermark values is greater than a predetermined threshold, determining the candidate watermark; (j) using the values of the same position of the watermark candidates to determine a final watermark; And (k) decoding the final watermark.

In the robust blind watermarking method of the DIBR attack using the depth change map, the value of the corresponding position is multiplied by the corresponding NCC value for each pixel at the same position of the candidate watermarks, And the pixel value of the corresponding position of the final watermark is determined according to the sign.

The present invention also relates to a computer-readable recording medium on which a program for performing a robust blind watermarking method of a DIBR attack using a depth change map is recorded.

As described above, according to the robust blind watermarking method of the DIBR attack using the depth change map according to the present invention, the nonvisibility of the inserted watermark information is excellent compared with the existing methods, and the BER bit error ratio) is almost 0, which is superior to the conventional methods.

In addition, according to the robust blind watermarking method of the DIBR attack using the depth change map according to the present invention, even in the additional attacks after the viewpoint movement, within a range that does not significantly deteriorate the original image, Effect is obtained.

In addition, according to the robust blind watermarking method of the DIBR attack using the depth change map according to the present invention, it is possible to improve the visibility of the watermark information in the free-view image rendering environment, There is obtained an advantageous effect that can be effectively used for protection of intellectual property rights on image contents which are contents.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a diagram showing a configuration of an overall system for carrying out the present invention; Fig.
2 is a view showing a pinhole camera model used in the present invention.
FIG. 3 is a diagram illustrating an example of a hole filling method used in the present invention, which is an example of (a) an original image, (b) a viewpoint moving image, and (c) a hole-filled image.
FIG. 4 shows an example of the effect of the viewpoint movement according to the present invention on the WM extraction. FIG. 4 (a) shows the original image, (b) (e) WM information extracted from the viewpoint moving image, and (f) example image of the WM information extracted after the viewpoint restoration.
FIG. 5 is a flowchart illustrating a digital watermark embedding process according to an embodiment of the present invention. FIG.
FIG. 6 is a detailed flowchart illustrating a process of inserting a watermark WM according to an embodiment of the present invention; FIG.
7 is a detailed flowchart illustrating a DVM generation process according to an embodiment of the present invention.
8 is a pseudo code illustrating a method of selecting a watermarking position according to an embodiment of the present invention.
9 is a configuration diagram of an LFSR used in accordance with an embodiment of the present invention;
10 is a graph illustrating a quantizer for WM insertion according to an embodiment of the present invention.
11 is a diagram illustrating a quantizer for WM insertion according to an embodiment of the present invention.
12 is a flowchart illustrating a WM extraction process according to an embodiment of the present invention.
13 is a detailed flowchart illustrating a WM extraction process according to an embodiment of the present invention.
FIG. 14 is a pseudo code for explaining a process of extracting WM data according to an embodiment of the present invention; FIG.
15 is a table showing test images used according to the experiment of the present invention.
16 is a table showing parameter values used according to the experiment of the present invention.
17 is a table showing attacks considered according to the experiment of the present invention.
18 is a diagram illustrating a result image of a main process of the method according to an embodiment of the present invention, wherein (a) is a circle color image, (b) is a circle depth image, (c) (d) DVM sub-sampled at each level, (e) baseline distance 5% point-in-time and JPEG quality 70/100 attack, (f) WM data extracted by sub-band, Result image.
19 is a table showing conventional methods to be compared according to the experiment of the present invention.
20 is a table showing the image quality comparison after the WM insertion according to the experiment of the present invention.
FIG. 21 is a comparative graph of experimental results according to the experiment of the present invention. FIG. 21 is a graph showing a comparison of experimental results of (a) a point-of-view attack, (b) (F) point of view + rotation, (g) point of view + matrix affine transformation attack.
FIG. 22 shows an example of the post-attack image according to the experiment of the present invention, in which (a) an original image, (b) JPEG compression image quality 40/100, (c) Result image for filtering.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings.

In the description of the present invention, the same parts are denoted by the same reference numerals, and repetitive description thereof will be omitted.

First, examples of the configuration of the entire system for carrying out the present invention will be described with reference to Fig.

1A, the overall system for embodying the present invention includes a watermarking apparatus (or a watermark embedding apparatus) 30 for inserting a watermark into a texture and a depth image 10, And a watermark extracting device 60 for extracting a watermark from a video (or watermark video)

1B, the watermark embedding device 30 may be embodied as a program system on the computer terminal 20. [ That is, the robust blind watermarking method of the DIBR attack using the depth change map according to the present invention is a program system on the computer terminal 20 that receives the texture and depth image 10 and performs watermarking on the image 10 . That is, the watermarking method may be implemented by a program and installed in the computer terminal 20 and executed. A program installed in the computer terminal 20 can operate as a single program system 30. [

Meanwhile, as another embodiment, the robust blind watermarking method of the DIBR attack using the depth change map may be implemented by a single electronic circuit such as an ASIC (on-demand semiconductor) other than the general purpose computer which is constituted by a program. Or dedicated computer terminal 20 for exclusive processing of watermarking only. This will be referred to as a digital watermarking device 30. Other possible forms may also be practiced.

The watermark extracting apparatus 60 or the watermark extracting method may be implemented by a program system on a computer terminal, an electronic circuit such as an ASIC, or a dedicated computer terminal.

On the other hand, the texture and depth image 10 consist of consecutive frames in time. One frame has one image. Also, the image 10 may have one frame (or image). That is, the image 10 corresponds to one image.

Before describing the method according to the present invention, a viewpoint moving image rendering method using DIBR used in the present invention and an influence of view point movement on DW will be described first with reference to FIG. 2 to FIG.

First, the viewpoint movement by depth-image-based rendering (DIBR) will be described.

2 is the point M in the real world, showing a simple pin-hole camera model temperature may cause problems in the point m in the image plane _I O of the camera. Denotes a O _P is the plane containing the optical center C, f is the focal length (focal length) in Fig. 2, respectively, (X, Y, Z) , (u, v, Z), (x, y, z) is O _P, O _I, illustrates the coordinate system of the real world, respectively. DIBR obtains the image at the desired point by re-projecting the point m formed on the image back to the point M in the real world and then re-projecting to the corresponding point m 'of the desired image plane O _I '. This projection method can be expressed by the following equation (1) [Non-Patent Document 1].

[Equation 1]

에서 영상평면의 점 m의 좌표

으로 위치시키는 식이다. 여기서

는 카메라 캘리브레이션(calibration) 매트릭스로 카메라 내부 파리미터들을 포함하고 있고,

과

는 회전(rotation)과 수평이동(transition)의 외부 파라미터, 즉 카메라의 움직임과 관련된 파라미터들을 갖고 있다. 이렇게 한 영상의 점을 실세계 좌표계로 역투영한 다음 임의의 영상으로 재투영할 때는 재투영할 영상의 좌표계와 외부 파라미터들로 수학식 1의 역변환을 수행하면 된다.Equation (1) is the coordinate of the point M in the real world

The coordinates of the point m in the image plane

. here

Is a camera calibration matrix that contains parameters within the camera,

and

Has parameters related to the external parameters of the rotation and the horizontal transition, i.e., the movement of the camera. When the point of the image is reversed to the real world coordinate system and then re-projected into an arbitrary image, the inverse transformation of Equation 1 can be performed with the coordinate system and external parameters of the image to be re-projected.

When a viewpoint moving image is rendered by the DIBR, a region (occlusion region) disappearing in the original image but moved in the viewpoint image and a region (non-occluded region) in which the region not present in the original image appears occur. In this case, since the original image has no information to fill the area, the non-occluded area remains as a hole. Therefore, this area must be filled with surrounding information, which is called hole filling or inpainting [Non-Patent Document 19]. Hall filling is one of the areas where 3D imaging has been studied in the past decade. FIG. 3 shows an example. In FIG. 3, the part of the image (b) in which the original image of (a) is shifted toward the right is a hole area, and the image filled with the area is image (c).

Next, the influence of the viewpoint movement on the watermarking will be described.

In order to investigate the effect of the viewpoint movement on DW, WM was inserted in the original image, and the WM was extracted after moving the viewpoint. One example of the experiment is shown in FIG. In order to see the effect of the viewpoint movement, the position where the WM is inserted is globally identical to the WM, and the method of inserting WM into each pixel is the same as that of the non-patent document 20. The global position of the image is three levels of a Mallat-tree 2DDWT (two-dimensional discrete wavelet transform), and one WM bit is inserted for each pixel of four sub-bands. Then, 5% viewpoint movement was performed to the left, and WM was extracted from the resultant image.

4A is a diagram illustrating an example of an original image of 512 × 512 divided by a pixel of LL3, HL3, LH3, and HH3 (resolution of each subband is 64 × 64) after performing a 3-level Mallat-tree 2DDWT 4 WMs are inserted for each subband by inserting one bit of 32 × 32 WM ('DDNT' binary logo image). (C) is an image obtained by moving this image to the depth information of (b), and the result of extracting WM from this image is (e). As you can see in the figure, you can see that most of the WM data is not extracted properly even though no other attack is applied. LH3 was the subband in which the greatest amount was extracted properly, and this tendency was the same in the experiments on other images.

FIG. 4 (f) shows the result of moving the point-in-time image of FIG. 4 (b) back to the original point by using the depth image and extracting WM as the resultant image. From this result, it can be seen that the WM information extracted from the right side of the left side in all the subbands is much more accurate because the position of the pixel does not change even if the distance of the objects on the left side of the original image is large. 4 (e) and (f), it can be seen that WM information can be obtained more accurately by extracting WM after moving to the original point. In order to do this, not only depth image is necessary but also depth After moving the image difference to the viewpoint of (b), the image can be returned back to the original viewpoint. Since the blind DW method is considered in the present invention, WM can not be extracted by this method.

Next, a robust blind watermarking method of a DIBR attack using a depth change map according to an embodiment of the present invention will be described. The DW method according to the present invention comprises a digital watermark (WM) insertion method for inserting a watermark into an image and a digital watermark extraction method.

First, a robust blind digital watermark (WM) insertion method of a DIBR attack using a depth change map according to the first embodiment of the present invention will be described with reference to FIG. 5 to FIG. 5 and 6 show an overall flowchart of a digital watermark (WM) embedding method according to the present invention.

5, a watermark embedding method according to the present invention includes the steps of (a) generating a depth variation map (DVM) (S10), (b) performing a 2DDWT on a texture image to obtain a sub- A step S30 of selecting the position of the watermark WM, a step S40 of encrypting the WM data, a step S50 of inserting the watermark WM, (f) restoring the watermarked image in the sub-band image (S60).

A detailed method is shown in Fig.

In the digital watermark embedding process according to the present invention, a region in which the position is most unchanged at the time of viewpoint movement is searched using the depth image. On the other hand, the original texture image performs 2DDWT at three levels, and the WM information is inserted in a small area of view movement using depth information at each level.

First, a depth variation map (DVM) is generated (S10). That is, the depth image is used to find an area where the position will not change even when the viewpoint is moved. The detailed steps for creating a depth variation map (DVM) are shown in FIG.

 First, median filtering is performed on the depth image in B × B units (B is a natural number of 3 or more) (S11). This may basically be assumed that the depth image is complete, but in some cases, the depth image may contain holes, which are used to eliminate the holes. However, even when there is no hole, the performance is better than that in the case where the experiment result filtering is performed.

Preferably, filtering of 9x9 units is performed. Experiments show that the filtering of 9 × 9 units in B × B filtering is the best.

The filtered resultant image is subjected to differential (two-dimensional differential) in each direction of horizontal and vertical directions (S12). Equation (2) represents a two-dimensional differentiation result expression at (x, y) position.

&Quot; (2) &quot;

Next, a depth derivative accumulation map (DDAM) is formed from the differential result (S13). That is, the DDAM value DDAM (x, y) of the specific position (x, y) is calculated by accumulating the differential values g (i, j) . Where p x q is the resolution of the binary WM.

&Quot; (3) &quot;

As shown in the above equation, the differential cumulative map is a cumulative addition for one pixel to represent the characteristics of a certain region. At this time, the positions of accumulated pixels are not required to be left-upper single of a certain region, but are determined in consideration of the convenience of technical implementation for restoring area coordinates for watermark insertion.

Finally, a depth variation map (DVM) is generated (S14).

As shown in Equation (4), the DDAM value DVM (x, y) at a specific position is a DDAM average value in a 3x3 region centering on the position.

&Quot; (4) &quot;

Next, sub-band images are generated by performing 2DDWT (2-dimensional discrete wavelet transform) on the texture image (S20).

Preferably, the original RGB image is converted into a YCbCr image, and the 2D image is performed on the Y image. And the watermark is inserted into the LH subbands (LH1, LH2, LH3) of the 2DDWT results of each level. LH1, LH2, and LH3 denote LH subband regions of 1-level, 2-level, and 3-level, respectively. The reason for selecting the LH subband among the LL, HL, LH, and HH subbands derived from 2DDWT is that the LH subband region is robust to the horizontal noise as compared with the other subband regions.

Next, the DVM is used to select the watermark (WM) position (or watermark embedding area) (S30).

The location to insert the watermark is selected using the DVM, where the watermark is inserted in the LH subband of each level as a result of 1/2 sub-sampling of the previous DVM horizontally and vertically (or DVM sub-picture). Assume, for example, that a Y image of M × N resolution and a depth image of the same resolution. DVM (DVM ₀ ) has the same resolution as the original image or depth image. The DVM ₁ (DVM ₁ ) having a resolution of M / 2 × N / 2 and 1/2 sub-sampling both horizontally and vertically is also used for the M / And has a resolution of 2 x N / 2. Similarly, the 2-level 2DDWT the LH subbands (LH2) and the DVM ₁ sub-sampling (sub-sampling) a DVM ₂ is M / 4 × N / 4, 3- a level 2DDWT LH subbands (LH3) and DVM ₂ The DVM ₃ sub-sampled has the same resolution of M / 8 x N / 8.

The method of selecting the watermarking position is shown in pseudo code in Fig. In the present invention, binary pxq data such as a binary logo is used as the WM to be inserted. 8 is a method for selecting a position to insert WM among the LHi as a result of performing the i-th 2DDWT, where the resolution is M _i × N _i = M / 2 ⁱ × N / 2 ⁱ (1≤i≤3) And outputs the set pos_WM of the WM insertion position with the DVM _i as an input.

A case configured in the former stage of the DVM DDAM to a value that represents the characteristics of the lower right-side area, while x≤M _i -p / 2 as the DVM search area and the pseudo-code determined by y≤N _i -q-in area. And sequentially searches for a region having the smallest depth change value among the predetermined regions. This is because the meaning of the small depth change is changed by surrounding pixels and groups in the DIBR execution, and the possibility of cracking due to the difference of the pixel distance travel is small. The fact that the depth change value is small does not necessarily mean the background but also includes an area inside the object.

In the present invention, p / 2 x q WM pixels obtained by vertically bisecting WM from one pixel of DVM _i are inserted. Therefore, the search range (res_DVMi) for the location selection is horizontal direction x M _i -p / 2 and vertical direction y N _i -q. When one position is selected, horizontal p / 2 pixels and vertical q pixels Are excluded from the next search range. The number of WM positions is such that the ratio f _j of the number of pixels into which the WM is inserted does not exceed a certain ratio y _i to the total number of pixels of the corresponding subband, This is to limit it properly. The value of? _i depends on each level and is determined experimentally.

Next, the WM data is encrypted (S40).

In other words, WM data is first encrypted for security. A common method of encrypting data is to use a cryptographic algorithm [Non-Patent Document 21], in which the decoding of these algorithms produces completely different results even if only one bit of the encrypted data changes. If the WM data extracted by the attack can be partially or substantially changed as in the present invention, such an algorithm can not encrypt the WM data. Therefore, in the present invention, a linear feedback shift register (LFSR) [Non-Patent Document 22] shown in FIG. 9 is used.

The LFSR is characterized by the first value (I ₁ , I ₂ , ... I _k ) of each memory element and whether the output of each memory element is returned (b ₁ , b ₂ , ... b _k ) It is different. Therefore, the present invention uses a set of 2k bits of the two values as the encryption key (key) (key = {I 1, I 2, ... I k, b 1, b 2, ... b k}) . However, in the present invention, the serial output f _{k, i} of the LFSR and the WM data w _i are exclusive-ORed with each other bit by bit so that the encrypted WM data w _i ' ).

&Quot; (5) &quot;

Where k is the register number of the LFSR. That is, the value output from the k-th register f _k. The watermark is binary p × q, and the index i = 1, 2,..., P × q in the watermark data w _i .

Next, the watermark WM is inserted (S50).

As described above, binary p × q data such as a binary logo is used for WM to be inserted.

In the present invention, one bit w _i 'of WM data is inserted into one pixel value. As described above, for each DVM selected by the WM position, vertically bisected WM data, that is, p / 2 × q WM bits are inserted in the lower right p / 2 × q pixel including the bit. As it proceeds along the selected WM position, the left and right sides of the WM data are inserted alternately.

The method of inserting one bit of WM into one pixel uses the modified linear quantizer of FIG. In the graph of Fig. 10 the horizontal axis, and pixel values (c _i), after the WM bit is inserted before (c _i), and the vertical axis WM bit is inserted 0 'and' 1 'is inserted into the WM bit (w _i' ). How to insert the WM bit by using a quantization is based on the value of the pixel value c _i is a quantization interval _{(k △ L, k = ±} 1,2,3, ...) and the WM bit w _i 'to be inserted in And it is summarized as shown in FIG.

In FIG. 10, the quantization step? _L is determined according to the magnitude of the weighted energy standard deviation of each level (L = 1, 2, 3) LH subband. A weighted standard deviation E _L in the equation (6) with two threshold values as compared with the (T _{L, l::} low value, T _{L, h} high value), is less than T _{L, l} T _{L, l,} T _{L, h} is greater than the use to T _{L, h,} two threshold values is a weighted standard deviation E _L respectively between △ _L. Where _{T L, l, T L,} h, weight △ _L is determined experimentally.

&Quot; (6) &quot;

&Quot; (7) &quot;

Here, M and N denote the horizontal and vertical sizes of the subband region, respectively, and L denotes the level.

Next, inverse 2DDWT is performed on the sub-band images in which the watermark is replaced with the sub-band image, and a watermarked texture image is generated (S60).

After inserting WM into each LH subband of each level, it performs 3-level 2DDWT inversion with LH subbands and WMB inserted subbands, and combines the Y component and the Cb and Cr components of the result into RGB As shown in FIG.

Next, a robust blind digital watermark (WM) extraction method of a DIBR attack using a depth change map according to a second embodiment of the present invention will be described with reference to FIG. 12 to FIG. 12 and 13 show an overall flowchart of a digital watermark (WM) extraction method according to the present invention.

As shown in FIG. 12, a watermark extraction method according to the present invention includes the steps of (a) generating a subband image by performing 2DDWT on a texture image (S110), (b) encrypting a watermark, (c) a step (S30) of extracting watermark data from sub-band images, (d) a step of forming a watermark (S140), and (e) a step of decrypting the watermark formed (S150).

A detailed method is shown in Fig.

Since the object of the present invention is blind watermarking, only the RGB color image and the original WM data which are subjected to the malicious / non-malicious attack including the viewpoint movement are used as the input of the extraction process, and neither the original image nor the original depth image is used. In the process of FIG. 12 or FIG. 13, the step of extracting the Y component from the input image, performing the three-level 2DDWT (S110), and encrypting the WM (S120) are the same as the inserting process. Only WM extraction, WM formation, and De-scrambling are described.

First, a texture image in which a watermark is embedded is subjected to 2DDWT to obtain a subband image (S110). In particular, when the texture image is an RGB color image, the RGB color image is converted into a YCbCr image. Then, a 3-level 2DDWT is performed on the Y (luminance) image to generate a sub-band image.

Next, the watermark is encrypted to generate an encrypted watermark (S120). This is the same as the watermark encryption method of the first embodiment. The generated encryption watermark is p x q encrypted WM data.

Next, WM data is extracted from the generated sub-band image (S130).

The process of extracting WM data from the attacked image is shown in pseudo code in FIG. This step inputs the Y component (I _test ) extracted from the attacked M × N image and the original WM data (W _scr ) encrypted with p × q. W _scr is divided by p / 2 × q to use the left (W _{scr, l} ) and right (W _{scr, r} ) data separately. Since the left and right halves of WM are inserted alternately when WM is inserted, WM _scratches are performed using W _{scr, l} and W _{scr, r} , respectively (FIG. 14, variable s). In addition, when inserted, all of the LH sub-bands of 2DDWT levels 1 to 3 are inserted, and WM data is also extracted from each of these sub-bands (variable L in FIG.

A method for extracting WM data corresponding to W _{scr, l} or W _{scr, r} (W _{0 in} Fig. 14) for each LH subband is as follows. First, the pixel values of each subband are quantized by the quantizer of FIG. 10 and WM values according to the quantization positions are extracted (Q _L (i, j)). The watermarks representing '0' and '1' are different according to the quantization interval of the pixel values of each subband. This is shown in Fig. Therefore, the pixel values are quantized according to the quantization step. Then, the watermark bit is extracted according to the interval of each quantized pixel value, and is represented by the extracted bit image QL (i, j).

Then _{W scr, l (W scr,} r) with the same size of p / 2 × q raster scan from upper left upper of the search window that subband _{Q L (i, j) (} raster scan) and W ₀ and the search window, (V _NCC ) between the pixel values Q _L (i to i + p / 2, j to j + q). If v _NCC is greater than the predetermined threshold (T _NCC ), the pixels of the search window are set as WM candidates (W _{ext, k} ) and stored separately ({W _ext }). At this time, the corresponding v _NCC value (v _{NCC, k} ) is also stored ({v _NCC }). If v _NCC is 100%, discard all W _{ext, k} and store the WM candidate in {W _ext } and terminate the search. Here, the NCC is a normalized cross correlation, and is an index for measuring the similarity between data.

Next, the final WM data is formed from the candidate WM data (S140).

And forms the final WM data with the extracted WM data candidates. First, all values extracted as' 0 'from each WM data candidate are replaced with' -1 '(W' _{ext, k} ). The left WM and right WM candidates of W ' _{ext, k} are classified into a left candidate group and a rear candidate group, and the final WM data of the corresponding part is determined in each candidate group.

As shown in Equations (8) and (9), the determination method is such that the position (i, j) of the WM is firstly calculated at the corresponding position of each W ' _{ext, k} And the corresponding NCC value v _{NCC, k} (q (i, j)). Determine the final WM data value w "(i, j) to '0' if this value is negative and to '1' if it is a positive value.

&Quot; (8) &quot;

&Quot; (9) &quot;

Next, the WM data is decoded (S150).

The final determined WM data is de-scrambled and finally extracted WM (w "'). The decryption uses the LFSR and encryption key of FIG. 9 used in encryption, XOR (exclusive-OR) between the output f _{k, i} and the finally determined WM data w ".

&Quot; (10) &quot;

Next, the effects of the present invention will be described through experiments.

For the method according to the present invention, an experiment was conducted to test robustness against various attacks including invisibility and viewpoint movement for WM insertion.

First, the experimental environment is explained.

The method according to the present invention is implemented in C / C ++ in a 64-bit Windows 7 Ultimate environment on a PC equipped with an Intel Core i7-2700K CPU and 16GB RAM. The test images used in the experiment are shown in the table of FIG. Among them, Breakdancer and Ballet are 3D video data of Microsoft Corporation (non-patent document 23), and the rest are stereoscopic images of Middlebury (non-patent document 24). Among them, Breakdancer and Ballet use only the first image of the video clip to avoid the homogeneity of the test image.

The values of the parameters experimentally determined are listed in the table of FIG. WM uses 32 × 32 binary logo images, and the ratio (γ _i ) of the WM pixels among the LH subband pixels at each level of 2DDWT is 13% for LH1 and 7% for LH2 based on the non- , And LH3 was 5%. The upper and lower thresholds T _{L, l} and T _{L, h} of the energy standard deviation, which determine the quantization step size of the quantizer for WM insertion, are 40 and 46.7, respectively _, for all subband levels. The NCC threshold T _NCC used for extracting WM data is two. First, when the 32 × 32 (1,024 bit) WM used in the present invention is used as it is, it is set to 0.159. For comparison with the conventional method described later, the value is set to 0.21 when the 64-bit WM is used.

The types and intensities of attacks considered in the present invention are listed in the table of FIG. First, a point-of-view attack was applied to adjust the baseline distance and the intensity was from 3% to 7%. Other attacks were JPEG compression, Gaussian noise addition, median filtering, and scaling attacks. These attacks attacked after moving the baseline distance 5% to the right. During the attack, the matrix affine transform attack is an attack attempted in [Non-Patent Document 13], which is an attack that distorts the image by varying the horizontal and vertical coordinate system of the image. Therefore, all of the eight matrices considered in this paper are considered.

Next, experimental results and comparison with existing methods will be described.

First, the main process results of the method according to the present invention are shown in FIG. (a) and (b) are the original color image and the depth image, and (c) the result of 2DDWT of the Y component of (a) at three levels. (d) shows the DVM sub-sampled according to the 2DDWT level and is shown in the corresponding LH subband. The PSNR value of the image reconstructed as WM by this WM is 49.05 (non-patent document dB). (a) the PSNR value for the original image is 13.86 [non-patent document dB (b)], and ]. (e) The WM data extracted from the image is shown by each subband (f). The final extraction WM formed by this data is shown in the subband of Fig. HH1. The original WM is shown in the HL1 subband of this figure. The bit error rate (BER) of the WM extracted for the original WM is 0.1064 and the NCC value is 0.7624.

The experimental results of the method according to the present invention are shown in comparison with the experimental results of the conventional methods. The comparison target method is listed in the table of FIG. 19, and basically, the two methods are [Non-Patent Document 12] and [Non-Patent Document 13]. [Non-Patent Document 12] The method uses WM of 5,926 bits and 73 bits, which is a method of WM for an image and an original image shifted to the right and left by predicting the maximum viewpoint movement amount. [Non-Patent Document 13] proposes a conversion method for emphasizing the six-directionality called DT-CWT, and inserts 64-bit WM information into the block by the column direction quantization method of the three-level conversion result. Since there is a difference in performance depending on the number of WM bits, the proposed method basically uses 32 × 32 1,024 bit WM, but the experiment for 64 bit WM is added for comparison. The results of the other methods in the reference literature were worse than those of the two methods, and [Non-Patent Documents 14] to [Non-Patent Document 18] were excluded from the comparison for the following reasons. As mentioned in the background art, [Non-Patent Document 14] does not perform an experiment for attacking to erase WM maliciously against an image after moving the viewpoint by applying an attack other than the viewpoint movement to the image before the viewpoint movement, Document 15] performed WM on the depth image, but this also can not protect the copyright on the color image, so it is not compatible with the intention of the present invention and is excluded. [Non-Patent Document 16] The method is a non-blind method requiring a depth image in extraction, and the method of [Non-Patent Document 17] assumes that left and right view images are given, 18] were excluded from the comparison because they were energized stereo images.

First, the image quality of the restored image after WM insertion is compared with the PSNR value and the SSIM value of the original image in the table of FIG. The method according to the present invention has the same value irrespective of the number of WM bits and shows better image quality than the method of Kim or the method of Lin using 64 bits. Therefore, it can be seen that the proposed method is superior to the conventional methods in nonvisibility of the embedded watermark information.

FIG. 21 is a graph showing the BER for each attack including the point-of-view attack for all methods to be compared. In this figure, (b) to (g) are the results of adding the attack after baseline distance 5% of the point move attack. First, in the case of the point-of-view attack in (a), both methods of Lin have good characteristics at a certain baseline distance, but the BER increases significantly as they deviate from that point. Kim's method showed good performance in the previous moving range. However, in the two cases of the proposed method, the BER is almost 0 in all the intervals, and the best performance is obtained in all the intervals as the moving point.

For the scaling attack (e) and the rotation attack (f) in the additional attack after moving the viewpoint, the proposed method showed much better performance than the existing methods. In particular, for the rotation attack, the proposed method showed a BER of almost zero at all intervals in both 1024-bit and 64-bit WM cases. For the Matrix affine transform attack, the proposed method using 64-bit WM showed the lowest BER in all transforms. In the case of using 1024-bit WM, especially high BER was shown in matrix 2 and 5, but the remaining matrix showed BER similar to Kim's method using 64-bit.

In the other attacks, the method according to the present invention showed very good performance for weak attacks, and the BER tended to be rapidly increased when the attack strength was increased to some degree or more. In the case of the JPEG compression attack (b), the image quality is 40/100 or more. In the case of the noise adding attack (c), 2.5% or more, and in the case of the median filtering attack (d) The attacked images are shown in FIG. 22 to show the degree of attack of these. (B), which shows a 40/100 JPEG compression attack, is significantly blurred at the boundary or high-frequency part of the object as compared with the original image of (a) ) Shows the noise in the low-frequency component, and (d) in the case of the 9 × 9 median filtering shows that the high-frequency component inside the object is almost lost. In other words, although these attacks are good attacks for eliminating WM information, they are considered to be unsuitable for reuse of attacked images because they are attacks that strongly cause deterioration in image quality. The PSNR values of (b), (c) and (d) were 28.85 [dB], 28.72 [dB] and 28.28 [dB] respectively.

Therefore, the method according to the present invention is superior to the conventional method in robustness against malicious / non-malicious attack that can reuse the image after moving the viewpoint.

In the present invention, a 2D image or a 3D image, which is obtained by receiving an original image and its depth image on a receiver side and viewing an arbitrary view image on a receiver side, is targeted to a digital image having high visibility and non- We propose a watermarking method. This method is a method of performing 3-level 2DDWT and inserting a watermark into the LH subband of each level. The watermark insertion position generates a depth variation map (DVM) I found a safe place to go. In the method of inserting watermark, the quantization step size is determined according to the energy of each subband, and the pixel value is changed according to the value of the original pixel and the watermark to be inserted. The method of extracting the watermark after the attack including the viewpoint movement extracts the possible candidates by using correlation with the original watermark information and determines the watermark bits by statistical method of the values of the watermark positions.

Experimental results show that the visibility of embedded watermark information is superior to that of conventional methods. The bit error ratio (BER) is close to zero for the point-in-time attacks considered. In addition, the proposed method showed better performance than the previous methods in the range of not destroying the original image even in the additional attacks after moving the viewpoint.

From the experimental results, it can be seen that the method according to the present invention has robustness against invisibility, viewpoint movement and additional attack of excellent watermark information in the environment of free-view image rendering and protects the intellectual property rights It will be useful.

Although the present invention has been described in detail with reference to the above embodiments, it is needless to say that the present invention is not limited to the above-described embodiments, and various modifications may be made without departing from the spirit of the present invention.

10: texture and depth image 20: computer terminal
30: program system 40: watermark image
60: Watermark extraction device 70: Watermark

Claims (16)

A robust blind watermarking method of a DIBR attack using a depth change map for inserting a watermark into a texture and a depth image composed of a texture image and a depth image,
(a) generating a depth variation map (DVM) by performing a derivative on the depth image in each of horizontal and vertical directions;
(b) generating a sub-band image by performing a 2-dimensional discrete wavelet transform (2DDWT) on the texture image;
(c) sub-sampling the DVM to generate a DVM sub-image having the same resolution as the sub-band image, selecting an insertion region using pixel values in the DVM sub-image, Selecting an area of the image as a watermark embedding area to embed a watermark;
(d) encrypting the watermark data;
(e) inserting the watermark data into the watermark embedding area of the sub-band image; And
(f) generating a watermarked texture image by performing an inverse 2DDWT on sub-band images that are replaced with a watermark-inserted sub-band image. The robustness of a DIBR attack using a depth- Watermarking method.
The method according to claim 1,
In the step (a), differentials are performed in the horizontal and vertical directions on the depth image, and a differential accumulation map (DDAM) is generated by cumulatively adding differential values within a predetermined range in each pixel of the differential result image , And a depth variation map (DVM) is generated by averaging a predetermined area in each pixel of the DDAM image, wherein the depth variation map (DVM) is generated.
3. The method of claim 2,
Wherein the depth image is differentiated after performing median filtering on the depth image in the step (a). The robustness blind watermarking method of the DIBR attack using the depth variation map.
3. The method of claim 2,
The differential accumulation map is generated by cumulatively accumulating differential values of a predetermined range of pixels in a lower right corner of each pixel of the differentiated result image in the step (a). The robustness of the DIBR attack using the depth variation map. Marking method.
The method according to claim 1,
And converting the RGB image to a YCbCr image to generate a sub-band image from a Y (luminance) image when the texture image is an RGB image, wherein the sub-band image is generated from the Y (luminance) image.
The method according to claim 1,
And performing a 2DDWT transformation on the texture image in all three levels to insert a watermark in the LH sub-band image among the 2DDWT results of each level. The robustness blind watermarking method of the DIBR attack using the depth change map.
The method according to claim 1,
In the step (c), the pixel and a predetermined range including the pixel are selected as an insertion region in the DVM sub-image in the order of the smallest pixel value, and in the remaining sub-images excluding the selected insertion region, And a region including a small pixel is selected as an insertion region. A robust blind watermarking method of a DIBR attack using a depth variation map.
The method according to claim 1,
Wherein the number of watermark embedding areas is determined such that a ratio of the number of pixels inserted in the watermark embedding area to a total number of pixels of the subband image is equal to or less than a predetermined ratio in step (c) Robustness of DIBR Attack using Mapping Blind Watermarking Method.
The method according to claim 1,
Wherein the watermark data is XORed with an output of a linear feedback shift register (LFSR) to encrypt the watermark data in step (d).
The method according to claim 1,
Wherein the watermark data is composed of bit data,
Wherein in the step (e), a pixel value of a pixel of the watermark embedding area is replaced with a value output by a linear quantizer in accordance with a bit of the watermark data, and a watermark is inserted Robustness of DIBR Attack Using Blind Watermarking Method.
11. The method of claim 10,
Wherein the quantization step of the linear quantizer is determined according to a magnitude of a weighted energy standard deviation of each level subband image, and the robustness blind watermarking method of the DIBR attack using the depth variation map.
12. The method of claim 11,
Wherein the quantization step DELTA _L of the linear quantizer is obtained by the following Equation 1. < EMI ID = 1.0 &gt;
[Equation 1]

,

L denotes a 2DDWT level, c _i denotes a pixel value of a subband, T _{L, l} and T _{L, h} denote a predetermined threshold value T _{L, 1} &lt; T _{L, h} .
The method according to claim 1,
When a watermark embedding area is set to half the size of the watermark data and the watermark data is inserted into a plurality of watermark embedding areas, Wherein the depth map is embedded in a region of the image.
A robust blind watermarking method of a DIBR attack using a depth change map for extracting a watermark from a texture and a depth image in which a watermark is embedded by the method of claim 1,
(g) generating a sub-band image by performing a 2-dimensional discrete wavelet transform (2DDWT) on the texture image;
(h) encrypting the watermark data;
(i) extracting watermark values in the sub-band image, and if the similarity between the watermark data and the extracted watermark values is greater than a predetermined threshold, determining the candidate watermark;
(j) using the values of the same position of the watermark candidates to determine a final watermark; And
(k) decoding the final watermark, wherein the robustness blind watermarking method of the DIBR attack using the depth variation map.
15. The method of claim 14,
The pixel value of the corresponding position of the final watermark is determined according to the sign of the sum value by adding all the values of the position and the NCC value for each pixel at the same position of the candidate watermark, Robustness of DIBR Attack using Mapping Blind Watermarking Method.
A computer-readable recording medium on which a program for performing a robust blind watermarking method of a DIBR attack using the depth change map of any one of claims 1 to 15 is recorded.

Images