KR101785195B1 - Feature Points Based Watermarking Method for Depth-Image-Based Rendering Based 3D Images and Apparatus Therefor - Google Patents

Abstract

A feature point based watermarking method and apparatus for DIBR three-dimensional imaging are disclosed. A feature point-based watermark embedding method according to an embodiment of the present invention includes extracting feature points from an image; Determining an insertion area to insert a watermark based on the extracted minutiae; Transforming the determined insertion region from a spatial domain to a frequency domain, and inserting a predetermined watermark in the insertion region converted into the frequency domain; And transforming the embedded region into which the watermark is inserted into a spatial domain.

Description

[0001] The present invention relates to a feature point based watermarking method and apparatus for DIBR three-dimensional images,

The present invention relates to a feature-based watermarking technique, and more particularly to a feature-based watermarking method for copyright protection of a DIBR (Depth-Image-Based Rendering) 3D image using SIFT (Scale invariant feature transform) And apparatus.

Recently, as the popularity of 3D contents continues, there is a problem of copyright protection for 3D images and video. In particular, the DIBR method based on depth information, which is a typical method for generating and displaying 3D images, is capable of adjusting the stereoscopic effect directly by the content user, requires a small space for storing the content, This is a technique that has the advantage of providing benefits.

Because of these various advantages, it is expected that the interest in deep information based DIBR 3D image and video and the possibility of being used in actual film and contents business will increase continuously. However, there is a limit to utilize the conventional 2D image based watermarking technique to solve the copyright protection problem of the 3D image, and there is a problem that it is difficult to apply it because of the characteristic of the 3D content itself.

Therefore, there is a need for a watermarking method for copyright protection of 3D images.

Embodiments of the present invention provide a feature point-based watermarking method and apparatus for copyright protection of a DIBR 3D image using SIFT (Scale invariant feature transform) feature points of an image.

A feature point-based watermark embedding method according to an embodiment of the present invention includes extracting feature points from an image; Determining an insertion area to insert a watermark based on the extracted minutiae; Transforming the determined insertion region from a spatial domain to a frequency domain, and inserting a predetermined watermark in the insertion region converted into the frequency domain; And transforming the embedded region into which the watermark is inserted into a spatial domain.

The extracting of feature points may extract SIFT feature points including position information, scale information, and orientation information using a predetermined Scale invariant feature transform (SIFT) algorithm.

The step of determining the insertion region may determine the insertion region based on at least one of the position information and the orientation information of the extracted SIFT feature points.

Further, the feature point-based watermark embedding method according to an embodiment of the present invention includes removing SIFT feature points corresponding to corner portions of the image when the SIFT feature points are extracted. Selecting only one orientation information for each of at least one SIFT feature point including a plurality of orientation information among the SIFT feature points; And adjusting the number of SIFT feature points based on a predetermined predetermined scale threshold, wherein the step of determining the insert region may determine the insert region based on the number of adjusted SIFT feature points .

Wherein the step of adjusting the number of SIFT feature points comprises: classifying the SIFT feature points according to orientation information of the SIFT feature points according to bins having a predetermined angle range; and determining the number of SIFT feature points It is possible to adjust the number of SIFT feature points for each of the sorted classes so that the number of SIFT feature points becomes smaller than the threshold number.

The step of inserting the watermark may insert the watermark into the insertion area using an ISS (Improved Spread Spectrum) insertion technique.

The step of inserting the watermark may insert the watermark into a middle band of the inserted region converted into the frequency domain.

A feature point-based watermark detection method according to an exemplary embodiment of the present invention includes extracting feature points from a watermark detection image for detecting a watermark; Determining a detection area for detecting a watermark based on the extracted minutiae points; And converting the determined detection region from the spatial domain to the frequency domain and detecting the watermark embedded in the watermark detection image from the detection region converted into the frequency domain.

The extracting of feature points may extract SIFT feature points including position information, scale information, and orientation information using a predetermined Scale invariant feature transform (SIFT) algorithm.

The determining of the detection region may determine the detection region based on at least one of the position information and the orientation information of the extracted SIFT feature points.

Further, the feature point-based watermark detection method according to an exemplary embodiment of the present invention includes removing SIFT feature points corresponding to corner portions of the watermark detection image when the SIFT feature points are extracted. Selecting only one orientation information for each of at least one SIFT feature point including a plurality of orientation information among the SIFT feature points; And adjusting the number of SIFT feature points based on a predetermined specific scale threshold value, wherein determining the detection region may determine the detection region based on the number of adjusted SIFT feature points .

A feature point-based watermark embedding apparatus according to an embodiment of the present invention includes: an extracting unit for extracting feature points from an image; A determining unit for determining an insertion area to insert a watermark based on the extracted minutiae points; An insertion unit that converts the determined insertion region from a spatial domain to a frequency domain and inserts a predetermined watermark into the insertion region converted into a frequency domain; And a conversion unit for converting the embedded region into which the watermark is inserted into a spatial domain.

A feature point-based watermark detection apparatus according to an embodiment of the present invention includes an extraction unit for extracting feature points from a watermark detection image to be watermarked; A determining unit that determines a detection area for detecting a watermark based on the extracted minutiae points; And a detector for converting the determined detection domain from the spatial domain to the frequency domain and detecting the watermark embedded in the watermark detection image from the detection domain converted into the frequency domain.

According to embodiments of the present invention, the copyright of the DIBR 3D image can be protected using the Scale invariant feature transform (SIFT) feature point of the image.

According to the embodiments of the present invention, high non-visibility can be ensured because the watermark embedding strength is adjusted in consideration of noise-visible areas using NVF masking.

According to the embodiments of the present invention, since the watermark detection area is selected based on the feature points, which are robust against the depth condition change attack and the depth image preprocessing attack caused by the DIBR system and do not use the entire image domain, And can be robust against geometric attacks such as < RTI ID = 0.0 >

FIG. 1 shows an example of comparing SIFT feature points in a central image, a left eye image, and a right eye image.
FIG. 2 is a flowchart illustrating an operation of a feature point-based watermark embedding method according to an embodiment of the present invention.
FIG. 3 is a diagram illustrating an exemplary process for removing feature points of an image boundary portion.
4 is a flowchart illustrating a process of refining SIFT feature points.
5 is a flowchart illustrating a process of determining a watermark embedding area and a message bit.
FIG. 6 is a flowchart illustrating an operation of a feature point-based watermark detection method according to an embodiment of the present invention.
FIG. 7 shows a configuration of a feature point-based watermark embedding apparatus according to an embodiment of the present invention.
FIG. 8 shows a configuration of a feature point-based watermark detection apparatus according to an embodiment of the present invention.

Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. In addition, the same reference numerals shown in the drawings denote the same members.

Embodiments of the present invention provide a watermarking technique that uses a similarity between SIFT feature points extracted from a central image, left eye images generated through a DIBR system, and SIFT feature points extracted from a right eye image, to provide a watermarking technique for protecting a DIBR 3D image That is the point.

The DIBR system generates the left eye image and the right eye image by moving the pixels of the central image in the horizontal direction using the depth information of the depth image, the DIBR rendering formula, and the depth factor value. The left eye image is an image obtained by moving a pixel of the central image in the right horizontal direction, and the right eye image is an image obtained by moving a pixel of the central image in the left horizontal direction. The hole region generated in the rendering process is filled with new pixel values through the hole filling process, and the central image, the left eye image, and the right eye image have very similar shapes.

When the feature points are extracted using the SIFT feature point extraction algorithm for the three images of the central image, the left eye image, and the right eye image, as shown in FIG. 1, the position, scale, and orientation information of the feature points are similar to each other have.

Through the threshold value for the scale of the minutiae point, it is possible to provide a robust property from the signal distortion attack by utilizing a large scale minutiae representing the characteristic for a relatively large area rather than a very small fine area. In the case of orientation, the whole 360 ° range is divided into bins as many as the number of bits to be inserted, and the same message bit is inserted into the feature points having the orientation corresponding to the angle range of the corresponding bin, It is possible to solve the problem of the difference in fine orientation between extracted minutiae points.

The watermarking method according to an embodiment of the present invention includes a step of selecting a patch area to insert a watermark using refined SIFT feature points satisfying a condition set as a reference, inserting a watermark, and detecting a watermark A patch region is selected using SIFT feature points satisfying the same condition to detect watermark information embedded in the corresponding region.

Hereinafter, a watermark embedding method and a watermark detecting method according to the present invention will be described.

1. Watermarking technique

The present invention provides a watermarking technique for DIBR 3D images using SIFT feature points.

In order to be robust against a depth condition change attack and a preprocessing attack on a depth image, which is an attack applicable to a DIBR system that generates a left eye image and a right eye image, the watermark embedding method according to the present invention is applied to a SIFT feature point To transform the image from the spatial domain to the frequency domain, and then inserts the watermark based on the ISS (Improved Spread Spectrum) insertion technique after DCT transforming the image, for example.

(1) SIFT Characteristic point refinement and determination of insertion area (or detection area)

The DIBR system is characterized by generating the left eye image and the right eye image while locating the pixels of the central image horizontally. Therefore, if the watermark is inserted into the central image in synchronization with the synchronization of the central image as in the conventional 2D watermarking technique, regions in which the watermark is embedded in the left eye image and the right eye image after the DIBR system is applied, The watermark can not be correctly detected because it is moved to another position or disappears. In this way, the watermarking technique for 2D image can detect the watermark by synchronizing the insertion area and the detection area constantly when the geometric change is not applied at the position where the watermark is inserted in the spatial and frequency domain Do.

However, since the DIBR system itself is regarded as an attack that breaks the synchronization between the watermark embedding area and the detection area, the watermark embedding method according to the present invention uses a method of inserting a watermark in a specific area, , And SIFT feature points are used in the process of determining the insertion region.

As described above, although the DIBR system breaks the synchronization between the insertion area and the detection area, it basically generates a left eye image and a right eye image, and is a rendering method used to give a stereoscopic effect when a person views the image using a display device. A person feels stereoscopic because the difference of distance between binocular is that each eye recognizes the information of different scenes for the same scene, and the brain synthesizes it to give stereoscopic feeling.

The DIBR left eye and right eye images are based on the principle that the central image is moved horizontally according to the depth information and the two eyes of the real person recognize the scene. The generated left eye image and right eye image are similar to the central image .

The watermark embedding method according to an exemplary embodiment of the present invention can be applied to a watermark embedding method in which a location, a scale, and a size of SIFT feature points extracted from a center image using a SIFT algorithm, Since the difference between the orientation information and the information of the minutiae points extracted from the left eye image and the right eye image is not large, the watermark embedding area can be determined using this.

The method according to an embodiment of the present invention refines SIFT feature points using a specific scale threshold value to provide robustness from signal processing and geometric attack and assigns message bits according to orientation assigning bit to orientation of SIFT points.

When representing the i-th feature point extracted by the SIFT Algorithm to Sift _i (x, y, s, o), and the number of SIFT feature points extracted from the image Z gaera, Sift in the image ₁ (x, y, s, o) to Sift _Z (x, y, s, o) can be extracted.

When the SIFT feature points are extracted from the image, the extracted SIFT feature points are refined. The first step in refining the SIFT feature points is to eliminate the feature points at the edges of the image (eliminating SIFT feature points). When the characteristic point of the image are extracted, as in the example shown in Figure 3, about the extracted characteristic point there is to create a patch of p _s × p _s size, as to the feature point of the blue which is located in the yellow range p _s × because it is not possible to create a patch of p _s size the feature point may be removed.

When x and y, which are position information of Sift ₁ (x, y, s, o), are expressed for a continuous space and w and h respectively represent a horizontal length and a vertical length of the image, By storing feature points, it is possible to select feature points that are not included in the corner area. That is, the feature point corresponding to the corner portion of the image can be removed using Equation (1).

[Equation 1]

The next step is to solve the case where one feature point Sift _i (x, y, s, o) has a plurality of orientation information, that is, a case where two feature points have the same position and scale information and orientations in different directions (refining SIFT orientation). In the watermark embedding process, matching is performed to beans of a specific angle range according to the orientation information of the minutiae, and message bits are allocated. When one minutiae has a plurality of orientation information, a watermark for different message bits When the patterns + Rp and -Rp are inserted, the signal components of each other can be canceled. Therefore, the feature points are refined to have only one orientation o for Sift _i (x, y, s, o).

Sift _i (x, y, s, o), which is a feature point extracted using the SIFT algorithm, generates a scale space and extracts a local extremum through comparison between adjacent DOGs (Difference of Gaussian) Is a feature point selected according to the results of the selection.

In this case, the minutiae having a small scale value are minutiae points extracted from a small small area, and conversely, the larger the scale is, the minutiae points extracted from a large area having a low frequency ingredient. Therefore, if signal distortion such as noise addition and JPEG compression are added to the image, the scale having a relatively small scale is a feature point for a small small area, so that the scale has a disadvantage that it is not robust compared with large feature points .

In the watermarking method according to the present invention, utilizing the SIFT feature points is for extracting SIFT feature points similar to each other before and after insertion of a watermark, and determining a watermark embedding area (or detection area) using the information. If the feature point is easily distorted due to a signal distortion attack or the like, synchronization between the watermark embedding area and the detection area is not performed well. Therefore, the process of refining the feature points using a specific threshold value is performed.

In this case, if a high threshold is set to select relatively strong feature points in the process, the number of feature points for watermark insertion and detection may be reduced, which may act as an obstacle to robustness and detection rate of the watermarking algorithm have. Accordingly, in order to solve such a problem, it is possible to extract a strong feature point through the process shown in FIG. 4, and at the same time ensure a sufficient number of feature points for inserting and detecting a watermark.

As shown in FIG. 4, the SIFT feature points extracted from the image are represented by Sift _i (x, y, s, o), the feature point removal process located in the boundary region of the image described with reference to FIG. 3, the orientation in one direction is jeonghaeju "represents La _{i (x, y, s,} o) the total number of feature points N, the SIFT feature points subjected to purification Sift expressed in.

At this time, the orientation of the minutiae point has a range from -180 ° (-π) to 180 ° (π). In the watermarking method according to the present invention, a method of allocating a message bit to be inserted according to the orientation information of minutiae points It is necessary to divide the orientation components according to specific criteria.

If the reference angle for dividing the whole orientation is n and the angular intervals divided by the reference angle are expressed as Bin, all m bins can be generated by mbins = 360 / n °.

In this case, the number of SIFT feature points having an orientation value belonging to the i-th bean and the angle range of the bean can be represented by Bin _i and M _bini , respectively.

Accordingly, the refined feature points Sift ' _i (x, y, s, o) can be classified into the following Table 1, and the number N' of all feature points satisfies Equation (2) below.

&Quot; (2) "

We classify the SIFT feature points according to the orientation components and classify them into m bins. To extract the feature points that are robust against the signal distortion attack, we compare the scale information s of Sift ' _i (x, y, s, o) with the scale threshold value Ts Thereby eliminating small scale feature points.

That is, the feature points Sift ' _i (x, y, s, o) corresponding to the condition s <Ts are removed because they represent feature points for very small and small regions. This process can be performed by increasing the threshold value Ts according to a specific parameter value [alpha] _s and Equation (3) below, and then performing the feature point elimination process, it is possible to refine the feature points of the large scale component providing robust properties.

&Quot; (3) &quot;

However, if a high threshold value is set to select relatively strong feature points, the number of feature points for selecting a watermark insertion and detection region is reduced, which can serve as an obstacle to the robustness and detection rate of the watermarking algorithm. Therefore, in the watermarking method according to the present invention, by using the method of comparing the number of feature points corresponding to each bin before refinement and the number of feature points corresponding to each bin after the refinement to the threshold value Tn, The number of feature points to be used is guaranteed.

As shown in FIG. 4, when M _bini 'representing the number of refined minutiae included in a specific bin satisfies the condition of Equation (4) below, the process of removing minutiae of a small scale is continued.

&Quot; (4) &quot;

When the condition of Equation (4) is not satisfied, the process of removing the small-scale feature points is stopped and finally the refined feature points are represented by Sift " _i (x, y, s, o) and the number of refined feature points is denoted by N" Express.

Through this process, a message bit is assigned based on the orientation component o of the refined minutiae Sift " _i (x, y, s, o), and a watermark pattern is generated using the bit information. Will be described in (2) below.

(2) Generate a watermark pattern

To generate a watermark pattern, a reference pattern Rp is generated using a secret key k _s and a reference pattern generator.

In this case, the generated reference pattern may be a random sequence having a length of Lp according to a Gaussian distribution with an average of 0 and a variance of 1. [

If both the watermark inserter and the detector have a common secret key k _s agreed in advance, they can generate the same reference pattern Rp at the watermark embedding stage and the detection stage.

The watermark pattern Wp to be added to the coefficient of the actual DCT domain is generated by multiplying Rp by message bits having a value of -1 or 1. In the present invention, the SIFT feature points Sift _i (x, y , s, o) to the SIFT feature point SIF " _i (x, y, s, o) of the robust component is assigned a message bit according to the orientation component o of the feature point.

As in the method described in (1) above, according to the orientation component of the minutiae, it is classified into an empty Bin _i divided by a specific reference angle, and a message bit b _i is allocated to each bin. Indicates bit allocation result.

Therefore, the watermarking method of the present invention selects (or determines) a patch region (or an insertion region) into which a watermark pattern is to be inserted by using the position information of the refined SIFT feature point, and generates a watermark to be inserted in the patch region The message bit information required in the process is determined according to the orientation component of the minutiae point.

Message bits b _i are assigned to feature points having an orientation component corresponding to the angle range of Bin _i . The watermark pattern to be inserted into the patch is generated by multiplying the reference pattern and the message bit as shown in Equation (5) below, and the message bit inserted in the image is equal to m, which is the number of bin for classifying the orientation of the SIFT feature point .

&Quot; (5) &quot;

(3) Method of inserting watermark

FIG. 5 is a flowchart illustrating a method of determining a patch area in which a watermark is to be inserted using position information and orientation information of Sift " _i (x, y, s, o), which is a feature point generated by extracting SIFT feature points from an image, And a message bit to be inserted into the patch area is allocated.

5, an original patch P ₀ of size p _s × p _s extracted from the original image and a message bit b _k to be inserted in the patch are used as input values of the watermark inserter, Generates a watermarked image by repeating the process of generating the watermarked image P _w by inserting the watermarked image into the image using the original position information.

The watermark embedding method according to the present invention can insert the generated watermark pattern into the DCT coefficients after converting the watermark into the Discrete Cosine Transform (DCT) domain instead of inserting the watermark directly into the spatial region of the extracted patch.

In this case, the watermark embedding method can embed a watermark in a DCT coefficient using an improved Spread Spectrum Embedding (SMF) technique. In order to insert a watermark relatively strongly into an invisible area, a noise visibility function (NVF) masking.

The watermark embedding method according to the present invention performs a block DCT before converting a watermark into a patch to convert the watermark into a frequency domain. The DCT transform is a transformation technique frequently used for image processing and image compression, (6) < / RTI >

&Quot; (6) &quot;

When the image is DCT-transformed, the upper left portion of the converted image is a portion indicating a low frequency component, and is an area that can visually affect when a watermark is inserted into the corresponding region. In addition, the component located at the coordinates of (1,1) in the upper left region is referred to as a DC component, and represents the average brightness component of the entire image. The other components are referred to as AC components.

Since the high-frequency component at the bottom right is a component that is considerably weak in the lossy compression method of compressing by removing the DCT high-frequency component value as in the JPEG compression, it is disadvantageous in that it is not robust against signal distortion when a watermark is inserted into the DCT high- do.

Therefore, invisibility and robustness can be guaranteed by inserting the watermark into the middle band of the DCT domain. In the present invention, in order to avoid the low frequency band in the DCT block of size p _s x p _s, Is inserted into the middle band of the DCT domain using the ISS (Improved Spread Spectrum) insertion technique.

If the vector of the DCT domain of the middle band in which the watermark is to be inserted is represented by S and the reference pattern to be inserted into the corresponding component is represented by Rp, the vectors S and Rp have a length of Lp. Since the lengths of the two vectors are the same, an equation for obtaining the inner product of S and Rp is as shown in Equation (7) below, and I representing the interference of the original signal S, (8) < / RTI >

&Quot; (7) &quot;

&Quot; (8) &quot;

The ISS insertion technique has an advantage that watermark can be inserted in the direction of improving the robustness of the watermarking method by adjusting the intensity of the interference I.

The watermark embedding method according to the watermark embedding method of the present invention is expressed by Equation (9) below, where b is a message bit,? Is an insertion strength,? Is a parameter used for adjusting interference I, S _w denotes a DCT coefficient in which a watermark is inserted.

라는 것을 확인할 수 있다.Since only a real reference pattern Rp is inserted but a new signal is generated and added to S by various parameters such as a message bit and an insertion intensity as shown in Equation 9 below, Wp, which is a watermark pattern added to S,

.

&Quot; (9) &quot;

Further, in the watermarking method of the present invention, it is possible to insert a watermark with a relatively strong intensity in an invisible area and a weak intensity in a visible area using NVF masking.

This is that based on the Human visual system (HVS) characteristics that the human vision, and to the NVF masking processing on the original patch P ₀ is the patch area ratio is determined by the explicit or the threshold value that the visible region T _NVF patch The watermark can be inserted by adjusting the insertion strength α according to the characteristics.

는 지역분산을 나타낸다.In this case, the NVF function represents the degree of noise added to the image when the noise is added to the image, and the NVF function is expressed by Equation (10) below

Represents the local variance.

&Quot; (10) &quot;

The NVF function is a function that produces a value close to 1 for a region with a better noise and a value closer to 0 for a region that is not visible. The NVF value becomes smaller as the invisible region becomes invisible.

를 계산한 뒤 해당 값을 임계값 T_NVF와 비교하여 워터마크 삽입세기를 결정할 수 있다.Therefore, in the watermarking method of the present invention, the NVF value for the patch P ₀

And comparing the value with the threshold value T _NVF , the watermark embedding strength can be determined.

In this case, as shown in Equation (11) below, the human perception and visual characteristics can be considered by inserting a watermark with a relatively strong intensity? _S in an invisible region and? _W with a weak intensity in a visible region have.

&Quot; (11) &quot;

The watermark-inserted DCT coefficients S _w are combined with the DCT coefficients of the remaining bands, transformed into the spatial domain by the inverse discrete cosine transform (IDCT) transformation, and a watermark-embedded patch P _w is generated.

The watermark inserted patch is added to the original image by using the position coordinates of the SIFT feature point used when generating the patch, and when the watermark insertion is completed for the patch determined through all the feature points, A watermarked image can be obtained.

The watermark embedding method according to the present invention is not a watermark embedding method in all areas of an image but embeds a watermark only in an area partially extracted as a SIFT minutiae point and uses a NVF masking to generate a weak intensity in a conspicuous area By inserting a watermark, high invisibility can be guaranteed.

The watermark embedding method according to the present invention not only inserts a watermark by dividing the image into whole or part of an image, but also inserts a watermark into a patch around a minutiae point. Therefore, the attack caused by the DIBR system in which the pixel information of the watermarked image changes It can be robust against geometric attacks.

2 watermark detection technique

As described with reference to Table 1, the watermark detection is performed by performing informa- tion detection that requires additional information used in the process of inserting the original image or watermark in the detection process, blinds that can detect the watermark without the original image and the additional information, Detection.

Although in-deed detection has the advantage that it has higher reliability than blind detection because it utilizes the original image in the detection process, it is limited to be applied to the actual industry due to a serious disadvantage that the original image is necessarily required in the detection process . On the other hand, blind detection has a merit of requiring a complicated design in the process of inserting and detecting a watermark compared with the informed detection method, but does not require an original image in the detection process.

The watermark detection method according to the present invention can detect blinds and detect watermarks not only in the center image but also in the newly generated left eye image and right eye image considering the feature of the DIBR rendering technique.

As described above, the watermark detection method according to an embodiment of the present invention is a watermark detection method in which a location, scale, orientation information, left eye image, and right eye image of a feature point extracted from a central image using a SIFT algorithm The difference between the information of the minutiae extracted from the watermark detection area is not large, so that it is possible to determine the watermark detection area.

That is, the watermark detection method according to the present invention commonly utilizes the SIFT feature point refinement process in the watermark embedding method. The watermark detection region is determined by selecting the patch region using the positional information of the SIFT feature points Sift " _i (x, y, s, o) extracted through the minutiae point refinement process and watermark detection is performed on the region Since the SIFT feature point refinement process is described in detail in the watermark embedding method, the description thereof is omitted here.

FIG. 6 shows a watermark detection process according to the present invention, which will be described in detail below.

(1) Watermark detection method

As shown in FIG. 6, the watermark detection process also uses the refined SIFT feature points Sift " _i (x, y, y) as well as the region in which the watermark is inserted using the position information of the refined SIFT feature points in the watermark embedding process. s, o), the patch region is selected and the watermark is detected.

That is, according to the watermark detection method of the present invention, a patch having a size of p _s x p _s is extracted and then DCT transformed. After skipping the low frequency band for the invisibility in the DCT block by the length of Ls, the DCT domain vector s _s extracts.

의 부호에 따라 해당 패치에 삽입된 메시지 비트 b'_i를 추출한다.A method of extracting a watermark pattern inserted by the ISS insertion method is to calculate a normalized inner product of the extracted vector S _s and the reference pattern Rp as shown in Equation (13) below,

The message bit b ' _i inserted in the corresponding patch is extracted.

In the watermark detection method of the present invention, an image capable of detecting a watermark can include not only a central image but also a left eye image and a right eye image.

The watermark detection equations are as shown in Equations (12), (13), and (14) below, and n represents parameters for noise that can be generated during transmission and reception of an image and noise that may occur in DIBR rendering.

&Quot; (12) &quot;

&Quot; (13) &quot;

&Quot; (14) &quot;

In the watermark embedding step, a patch image is DCT-transformed for invisibility and robustness to extract a vector of the length Lp of the middle band, and a watermark is inserted according to Equation (9). In this process, A watermark pattern component is added.

After the detection process in extracting the insertion process and the vector S _s in the DCT domain, the middle band in the same manner, and generated using the private key and the reference pattern generator, the components of the reference pattern Rp added to the band, S _s and normalization of Rp The message bit information inserted in the corresponding patch can be extracted.

In the watermarking method, since the whole orientation is divided into m bins and message bits are allocated according to the orientation component of the SIFT feature point, a message sequence of length m can be extracted from one image.

The robustness against attack of the watermarking technique according to the present invention can be evaluated by calculating the bit error rate (BER) between the actually inserted message sequence and the extracted message sequence as shown in Equation (15) below.

&Quot; (15) &quot;

As described above, the method according to embodiments of the present invention can protect the copyright of the DIBR 3D image using the SIFT (Scale invariant feature transform) feature point of the image, and considers the noise-conscious region using the NVF masking Thereby adjusting the watermark embedding strength, thereby ensuring high invisibility.

In addition, the method according to embodiments of the present invention is robust against a depth-condition change attack and a depth image preprocessing attack caused by the DIBR system, and since a watermark detection area is selected based on a feature point instead of using an entire image domain, And geometric attacks such as position shifting.

FIG. 7 illustrates a configuration of a feature point-based watermark embedding apparatus according to an embodiment of the present invention, and illustrates a configuration of an apparatus for embedding feature point-based watermark embedding.

7, a watermark embedding apparatus 700 according to an embodiment of the present invention includes an extracting unit 710, a refining unit 720, a determining unit 730, an inserting unit 740, and a converting unit 750, .

The extraction unit 710 extracts feature points from a central image, for example, an image to which a watermark is to be inserted.

At this time, the extracting unit 710 may extract SIFT feature points including position information, scale information, and orientation information using a predetermined Scale invariant feature transform (SIFT) algorithm.

When the SIFT feature points of the image are extracted, the refinement unit 720 refines the SIFT feature points.

More specifically, the refinement unit 720 removes the SIFT feature points corresponding to the corner portions of the image, and performs refinement such that only one orientation information is included for each of at least one SIFT feature points including a plurality of orientation information among the SIFT feature points (Or selects) the SIFT feature points, and refines the SIFT feature points by adjusting the number of the SIFT feature points based on the predetermined specific scale threshold value.

At this time, the refinement unit 720 classifies the SIFT feature points according to the orientation information of the SIFT feature points according to bins having a predetermined angle range, and when the number of the SIFT feature points for each of the classified wise images is smaller than a predetermined threshold number It is possible to adjust the number of SIFT feature points for each bin that is classified so as to be classified.

The determination unit 730 determines an insertion area to insert a watermark based on the SIFT feature points.

At this time, the determination unit 730 can determine the insertion area based on at least one of the position information and the orientation information of the SIFT feature points. Of course, the determination unit 730 can determine the insertion area based on the adjusted number of SIFT feature points when the number of SIFT feature points is adjusted by the refinement unit 720.

The insertion unit 740 transforms the determined insertion region from the spatial domain to the frequency domain, and inserts the predetermined watermark into the insertion region converted into the frequency domain.

At this time, the inserting unit 740 can insert a watermark into the insertion region using an ISS (Improved Spread Spectrum) insertion technique, and insert the watermark into the middle region of the insertion region converted into the frequency domain.

The converting unit 750 converts the embedded region into which the watermark is inserted into the spatial domain, thereby generating an image in which the watermark is embedded.

The feature point-based watermark embedding device 700 of the present invention can include all the operations and functions for the feature point watermark embedding process described above even if the description is not described in FIG. 7, it will be understood by those skilled in the art It is clear to you.

FIG. 8 shows a configuration of a feature point-based watermark detection apparatus according to an embodiment of the present invention, and shows the configuration of an apparatus for performing the feature point-based watermark detection process.

8, a watermark detection apparatus 800 according to an embodiment of the present invention includes an extraction unit 810, a refinement unit 820, a determination unit 830, and a detection unit 840.

The extracting unit 810 extracts feature points from any one of a watermark detection image to detect a watermark, for example, a central image, a left eye image, and a right eye image.

At this time, the extracting unit 810 can extract SIFT feature points including the position information, the scale information, and the orientation information by using a predetermined Scale invariant feature transform (SIFT) algorithm.

When the SIFT feature points of the watermark detection image are extracted, the refinement unit 820 refines the SIFT feature points.

Specifically, the refinement unit 820 removes the SIFT feature points corresponding to the corner portions of the watermark detection image, and only one orientation information is provided for each of at least one SIFT feature points including a plurality of orientation information among the SIFT feature points (Or selects) the SIFT feature points, and adjusts the number of the SIFT feature points based on a predetermined specific scale threshold value to refine the SIFT feature points.

At this time, the refinement unit 820 classifies the SIFT feature points according to orientation information of SIFT feature points according to bins having a predetermined angle range, and if the number of SIFT feature points for each of the classified wise images is smaller than a predetermined threshold number It is possible to adjust the number of SIFT minutiae for each bin that is classified so as to be classified according to the situation.

The determination unit 830 determines a detection area to detect a watermark based on the SIFT feature points.

At this time, the determination unit 830 can determine the detection region based on at least one of the position information and the orientation information of the SIFT feature points. If the number of SIFT feature points is adjusted by the refinement unit 820, the determination unit 830 may determine the detection region based on the adjusted SIFT feature points.

The detection unit 840 converts the determined detection region from the spatial domain to the frequency domain, and detects the watermark embedded in the watermark detection image from the detection domain converted into the frequency domain.

Although the feature point-based watermark detection apparatus 800 of the present invention can include all the operations and functions for the feature point watermark detection process described above even if the description thereof is not described in FIG. 8, it will be apparent to those skilled in the art It is clear to you.

The system or apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the systems, devices, and components described in the embodiments may be implemented in various forms such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array ), A programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to embodiments may be implemented in the form of a program instruction that may be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI &gt; or equivalents, even if it is replaced or replaced.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (22)

Extracting SIFT feature points from an image using a predetermined Scale invariant feature transform (SIFT) algorithm;
Removing the SIFT feature points corresponding to corner portions of the image when the SIFT feature points are extracted;
Selecting only one orientation information for each of at least one SIFT feature point including a plurality of orientation information among the SIFT feature points;
Classifying the SIFT minutiae points according to orientation information of the SIFT minutiae points according to bins having a predetermined angular range and setting the number of SIFT minutiae to be smaller than a predetermined threshold number Adjusting the number of SIFT feature points of the image;
Determining an insertion area to insert the watermark based on the number of SIFT feature points adjusted;
Transforming the determined insertion region from a spatial domain to a frequency domain, and inserting a predetermined watermark in the insertion region converted into the frequency domain; And
Transforming the embedded region into which the watermark is inserted into a spatial domain
Based watermark embedding method.
The method according to claim 1,
The step of extracting the feature points
Wherein the SIFT feature points including position information, scale information, and orientation information are extracted using the predetermined Scale invariant feature transform (SIFT) algorithm.
3. The method of claim 2,
The step of determining the insert region
Wherein the insertion region is determined based on at least one of the position information and the orientation information of the extracted SIFT feature points.
delete delete The method according to claim 1,
The step of inserting the watermark
And inserting the watermark into the insertion region using an ISS (Improved Spread Spectrum) insertion technique.
The method according to claim 1,
The step of inserting the watermark
And inserting the watermark into a middle band of the insertion region converted into the frequency domain.
Extracting feature points from a watermark detection image to be watermarked;
Determining a detection area for detecting a watermark based on the extracted minutiae points; And
Converting the determined detection region from a spatial domain to a frequency domain, and detecting a watermark embedded in the watermark detection image from the detection domain converted into the frequency domain
Lt; / RTI &gt;
The step of extracting the feature points
Extracts SIFT feature points including position information, scale information, and orientation information using a predetermined Scale invariant feature transform (SIFT) algorithm,
The step of determining the detection region
Determines the detection area based on at least one of the position information and the orientation information of the extracted SIFT feature points,
Removing the SIFT feature point corresponding to a corner portion of the watermark detection image when the SIFT feature points are extracted;
Selecting only one orientation information for each of at least one SIFT feature point including a plurality of orientation information among the SIFT feature points; And
Classifying the SIFT minutiae points according to orientation information of the SIFT minutiae points according to bins having a predetermined angular range and setting the number of SIFT minutiae to be smaller than a predetermined threshold number Adjusting the number of SIFT feature points of the image;
Further comprising:
The step of determining the detection region
And the detection region is determined based on the number of the adjusted SIFT feature points.
delete delete delete An extracting unit for extracting feature points from an image;
A determining unit for determining an insertion area to insert a watermark based on the extracted minutiae points;
An insertion unit that converts the determined insertion region from a spatial domain to a frequency domain and inserts a predetermined watermark into the insertion region converted into a frequency domain; And
A transform unit for transforming the inserted region into which the watermark is inserted into a spatial domain,
Lt; / RTI &gt;
The extracting unit
Extracts SIFT feature points including position information, scale information, and orientation information using a predetermined Scale invariant feature transform (SIFT) algorithm,
The SIFT minutiae corresponding to the corner portion of the image is removed and only one orientation information is selected for each of at least one SIFT minutiae including a plurality of orientation information among the SIFT minutiae points, And for adjusting the number of SIFT feature points based on the determined specific scale threshold value,
Further comprising:
The determination unit
Determines the insertion area based on the number of adjusted SIFT feature points,
The purification unit
Classifying the SIFT minutiae points according to orientation information of the SIFT minutiae points according to bins having a predetermined angular range and setting the number of SIFT minutiae to be smaller than a predetermined threshold number Wherein the feature point-based watermark embedding device adjusts the number of SIFT feature points.
delete 13. The method of claim 12,
The determination unit
Wherein the insertion region is determined based on at least one of the position information and the orientation information of the extracted SIFT feature points.
delete delete 13. The method of claim 12,
The insert
And inserting the watermark into the insertion area using an ISS (Improved Spread Spectrum) insertion technique.
13. The method of claim 12,
The insert
And inserting the watermark into the intermediate band of the insertion region converted into the frequency domain.
delete delete delete delete

Images