KR101311309B1 - Detecting method of image operation and detecting device of image operation - Google Patents

Abstract

PURPOSE: An image manipulation detection method and a device thereof are provided to detect whether a target image, which is examined, is manipulated or not by extracting Photo Response Non-Uniformity (PRNU) from the target image and extracting a correlation plane based on the extracted value. CONSTITUTION: An image manipulation detection device calculates the change rate of the size of a target image which is examined and restores the target image (S210). The image manipulation detection device extracts Photo Response Non-Uniformity (PRNU) from the target image (S220). The image manipulation detection device calculates Peak to Correlation Energy (PCE) using a correlation plane which is obtained by comparing the extracted PRNU and standard PRNU (S230, S240). The image manipulation detection device determines whether the device which photographed the target image is a standard device or not by comparing the calculated PCE and the standard PCE of the standard device (S250). [Reference numerals] (S210) Calculate the change rate of a target image and restore the target image for the size of a target image to respond to the calculated the change rate of size; (S220) Extract Photo Response Non-Uniformity (PRNU) from the target image; (S230) Use the PRNU value of the standard device to compare the similarity between the extracted PRNU value and the standard PRNU value; (S240) Calculate PCE value by using correlation section calculate through the comparison result; (S250) Determine whether the device which photographed the target image is a standard device or not by comparing the calculated PCE and the standard PCE of the standard device

Description

Image manipulation detection method and image manipulation detection device {DETECTING METHOD OF IMAGE OPERATION AND DETECTING DEVICE OF IMAGE OPERATION}

The present invention relates to an image manipulation detection method and an image manipulation detection apparatus capable of detecting whether an image has been manipulated.

The change to the information society due to the development of information and communication technology has caused the popularization of digital devices. In addition, recently released digital devices perform better than the past digital devices despite being released at a low price. This makes it easier for general users to access digital devices, and as a result, the reliability of data generated by digital devices becomes increasingly important.

 In addition, due to the popularization of professional image editing apparatuses such as Photoshop and Illustrator, easy forgery of images using them is being made more and more, while the development of technology for detecting such forgery images is still at a rudimentary level.

On the other hand, digital image forensic technology is a technology related to the reliability problem of data generated in digital devices, and has been largely divided into image acquisition equipment identification and image forgery detection technology. First, the image acquisition equipment discrimination technique is a technique for determining what equipment is used to acquire the inspection target image from the inspection target image. The image acquisition equipment discrimination technology aims to determine not only the type of equipment used to acquire the inspection target image but also the exact model name level using only the inspection target image.

In addition, the image forgery detection technology is a technology for determining whether the inspection target image has undergone a forgery modulation step. Forgery of digital images leaves no visible clues, but changes the statistical properties that make up the image. It aims to find the forged and altered parts by finding such changed statistical characteristics.

 On the other hand, security images taken by CCTV, etc. exist in the form of infrared image as well as RGB image, and is manipulated by a special operation method. Therefore, the security video is difficult to detect the operation with high accuracy with the conventional forensic technology described above.

An object of the present invention is an image manipulation detection method and an image manipulation detection apparatus capable of detecting a device in which an inspection target image is photographed, or detecting whether a portion of the inspection target image is cut or modified, that is, whether the inspection target image has been manipulated. It is about.

The method for detecting whether an image is manipulated according to an embodiment of the present invention includes calculating a size change rate of an inspection target image and restoring the inspection target image such that the size of the inspection target image corresponds to the calculated size change rate. Extracting a PRNU (Photo Response Non-Uniformity) value from the reconstructed inspection target image, comparing the similarity between the extracted PRNU value and the reference PRNU value using a reference PRNU value of a reference device, Calculating a peak to correlation energy (PCE) value using a correlation plane derived through a comparison result, and comparing the calculated PCE value with a PCE reference value of the reference device, wherein the device to which the test target image is photographed is And determining whether it is a reference device.

The method may further include estimating whether there is an area extracted from the inspection target image by using the position of the peak in the correlation plane.

In the estimating step, when the position of the peak is different from the reference position, it is estimated that at least a part of the inspection target image has been extracted, and the extracted position corresponds to the right side of the position of the peak. It features.

The method may further include estimating a deformed region of the inspection target image by using partial similarity between the extracted PRNU value and the reference PRNU.

)의 분산을 이용하여 산출하고, 상기 분산(

)은 하기의 [수학식 2]인 것을 특징으로 한다.In one embodiment, the rate of change of magnitude is the second derivative of the pixel function p (i, j) corresponding to the inspection subject image represented by Equation 1 below.

Calculated using the variance of

) Is characterized by the following [Equation 2].

[Equation 1]

&Quot; (2) "

(Where R is a value corresponding to the horizontal size of the image)

In an embodiment, the restoring may include performing a Fourier transform of the variance of the second derivative to analyze the periodicity in the variance of the second derivative, and extracting a peak value derived through the Fourier transform. Comparing with the reference value of the Fourier transform and as a result of the comparison, if the peak value is larger than the reference value, determining that the size of the inspection target image has been converted.

)을 이용하여, 하기의 [수학식 3]과 같이 산출되는 것을 특징으로 한다.In one embodiment, the rate of change of magnitude (W) is the peak value (which can be expressed as a frequency)

) Is calculated as shown in Equation 3 below.

&Quot; (3) "

The comparing of the similarity may include calculating a cross correlation between the extracted PRNU value and the reference PRNU value, and deriving the correlation plane from the calculated cross correlation.

In an exemplary embodiment, when the calculated PCE value is smaller than the PCE reference value, it is determined that the inspection target image is manipulated.

In one embodiment, the noise is extracted by applying a wavelet-based filter from the reconstructed inspection target image, and the PRNU of the reconstructed inspection target image is applied by applying a maximum likelihood estimation (MLE) to the extracted noise. It is characterized in that the extraction.

In an embodiment, the reference PRNU value may be converted into a MACE filter in order to minimize energy of the correlation plane and derive a correlation plane having a high peak.

An image manipulation detecting apparatus according to an embodiment of the present invention calculates the size change rate of the inspection target image and restores the inspection target image such that the size of the inspection target image corresponds to the calculated size change rate, and the restoration. A PRNU extraction unit for extracting a PRNU (Photo Response Non-Uniformity) value from the scanned image, a comparison unit comparing the similarity between the extracted PRNU value and the reference PRNU value using a reference PRNU value of a reference device, By using the correlation plane derived through the comparison result, a PCE (peak to correlation energy) value is compared with the PCE calculation unit and the calculated PCE value and the PCE reference value of the reference device. It is characterized in that it comprises a determination unit whether the reference device.

The determining unit may estimate an area extracted from the inspection target image by using a position of a peak in the correlation plane.

The determining unit may estimate a deformed region of the inspection target image by using partial similarity between the extracted PRNU value and the reference PRNU.

In an embodiment, the reference PRNU value may be converted into a MACE filter in order to minimize energy of the correlation plane and derive a correlation plane having a high peak.

The image manipulation detecting method and the image manipulation detecting apparatus according to an embodiment of the present invention may extract a PRNU value from an inspection target image, derive a correlation plane based on the detection image, and detect whether the inspection target image is manipulated. .

In addition, the image manipulation detection method and the image manipulation detection apparatus according to an embodiment of the present invention can detect whether the security image captured by the CCTV has been manipulated, it is possible to prevent crime due to the manipulation of the security image.

1 is a block diagram of an image manipulation detection apparatus according to an exemplary embodiment.
2 is a conceptual diagram illustrating an image manipulation detection method according to an exemplary embodiment.
3, 4 and 5 are conceptual views for explaining the manipulated image.
6 and 7 are conceptual views illustrating the accuracy of the image manipulation detection method according to an embodiment of the present invention.
8 and 9 are conceptual views illustrating a method of detecting whether a security image is manipulated using an image manipulation detection method according to an exemplary embodiment.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, It will be possible. While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and are herein described in detail. It is to be understood, however, that the invention is not intended to be limited to the particular forms disclosed, but on the contrary, is intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

Hereinafter, an image manipulation detection method and an image manipulation detection apparatus according to an exemplary embodiment will be described with reference to the accompanying drawings.

1 is a block diagram of an image manipulation detection apparatus according to an exemplary embodiment.

The image manipulation detection method and the image manipulation detection apparatus according to the present invention detect whether an RGB image or an infrared image is manipulated, and determine whether a source of an image is manipulated, at least a portion of the image is extracted, or at least a portion of the image is deformed. It is possible to detect.

Meanwhile, as illustrated in FIG. 1, the image manipulation detecting apparatus 100 capable of detecting whether an image has been manipulated includes a reconstructor 110, a photo response non-uniformity (PRNU) extractor 120, and a comparator 130. ), The PCE calculation unit 140 and the determination unit 150 may be included.

The restoration unit 110 may calculate a size change rate of the inspection target image, and restore the size of the inspection target image to the original size by using the calculated size change rate.

The rate of change of size uses a pixel function (p (i, j)) corresponding to the inspection target image represented by Equation 1 below, and the pixel value of the pixel function has a value between 0 and 255. Can be. Here, i and j are variables representing positions of specific pixels in the image.

) 분산의 주기적인 확률적 특성을 이용하여, 상기 크기 변화율을 산출할 수 있다.The reconstruction unit 110 performs the second derivative of the pixel function p (i, j) (

The rate of change of magnitude can be calculated using the periodic probabilistic characteristic of the variance.

)를 연산한다.First, the reconstructor 110 performs a second derivative of the pixel function p (i, j) to perform a second derivative (Equation 1) below.

) Is calculated.

[Equation 1]

)를 이용하여, 하기의 분하기의 [수학식 2]와 같이, 상기 2차 도함수(

)의 분산(

)을 추정할 수 있다.And, the restoration unit 110 is the second derivative (

), As shown in [Equation 2] below, the second derivative (

Distribution of

) Can be estimated.

&Quot; (2) "

(Where R is a value corresponding to the horizontal size of the image)

) 에서 주기성을 분석하기 위하여, 분산(

)을 푸리에 변환(Fourier transform)시킨다. Next, the restoration unit 110 is the dispersion (

In order to analyze the periodicity in

) Is Fourier transform.

The Fourier transform is used to find a periodicity, and the restoration unit 110 compares a peak value derived through the Fourier transform with a reference value of the Fourier transform.

)은 주기성을 갖는다고 표현할 수 있고, 나아가 상기 검사대상 영상이 크기변환되었다고 판단될 수 있다.As a result of the comparison, when the derived peak is greater than the reference value, the variance (

) May be expressed as having periodicity, and furthermore, it may be determined that the inspection subject image is scaled.

On the other hand, the restoration unit 110 calculates the size conversion ratio (W) by using the position of the peak (peak) derived through the Fourier transform.

Since the peak position derived from the Fourier transform represents a frequency, the size conversion ratio W of the inspection target image may be calculated as shown in Equation 3 below.

&Quot; (3) "

As described above, the reconstructor 110 may equally adjust the reference image and the sync by reconstructing the inspection target image to the original size using the calculated size conversion ratio W.

Here, the reference image is an image obtained from a reference device that is assumed that the inspection target image has been taken. In the present invention, by comparing the attribute values of the inspection target image with the attribute values of the reference image, It can be detected whether it has been manipulated.

On the other hand, the PRNU extractor 120 extracts a PRNU (Photo Response Non-Uniformity) value from the reconstructed inspection target image corresponding to the calculated size conversion rate (W).

Here, the PRNU (Photo Response Non-Uniformity) value is a value representing the degree of pixel response nonuniformity, and may be extracted by applying a wavelet-based filter to the reconstructed inspection target image.

The PRNU extractor 120 first extracts noise from the reconstructed inspection target image using the wavelet-based filter, and extracts a maximum likelihood estimation method to the extracted noise (MLE). The maximum likelihood estimation is applied to extract the PRNU of the reconstructed test target image as shown in Equation 4 below.

&Quot; (4) "

는 검사대상 영상이 동영상인 경우, 동영상을 이루는 복수개의 정지영상 중 k번째 정지영상의 노이즈 residual이고,

는 k번째 복원된 검사대상 영상에 웨이블릿 기반 필터를 적용한 후 노이즈가 제거된 프레임(frame)이다.From here,

Is the noise residual of the kth still image among the plurality of still images constituting the moving image when the inspection target image is a moving image,

Is a frame from which noise is removed after applying a wavelet-based filter to the k-th reconstructed inspection target image.

In addition, the wavelet-based filter here uses a seismic pulse composed of a plurality of cycles, and the maximum likelihood estimation method is a method of estimating an unobservable parameter (parameter) based on an observed sample. .

On the other hand, the PRNU extractor 120 is a reference PRNU value of the reference device, for the detection accuracy of whether the inspection target image is manipulated, even if the restored inspection target image and the reference image is an infrared image, not RGB image, That is, the reference PRNU value corresponding to the reference image photographed by the reference apparatus may be converted into a MACE (Minimum Average Correlation Energy) filter. Here, the MACE filter may be designed to minimize the correlation plane energy and derive a correlation plane having a high peak.

On the other hand, the MACE filter may be expressed as shown in Equation 5 below.

&Quot; (5) "

Here, H denotes a MACE filter, X denotes a one-dimensional vector form of a reference PRNU, + denotes a transpose matrix, u denotes a predefined correlation constant value, and k denotes an index of the X array. Means Fourier transform.

That is, here, if the size of the X array is K, k is 1 ≦ k ≦ K.

Meanwhile, the comparator 130 calculates a cross-correlation between the extracted PRNU value and the reference PRNU value in order to compare the similarity between the PRNU value extracted from the reconstructed inspection target image and the reference PRNU value, and thereby establishes a correlation plane. To derive

Here, the correlation means when one side increases between two variables, the other side tends to increase (or decrease). In addition, a numerical value representing the degree of correlation between two variables x and y is called a correlation coefficient. Here, two variables x and y may be extracted PRNU values and reference PRNU values, respectively.

The PCE calculator 140 may calculate a personal consumption expecditures (PCE) which can be expressed by Equation 6 below in the derived correlation plane.

&Quot; (6) "

는 상관관계 평면에서의 첨두치를 나타낸다.Where CC refers to the correlation plane,

Represents the peak in the correlation plane.

Next, the determination unit 150 compares the PCE value and the reference value calculated through Equation 6 above.

As a result of the comparison, when the calculated PCE value is equal to or larger than the reference value, the determination unit 150 determines that the restored inspection target image is captured by the same device as the reference image. Also, when the comparison result shows that the calculated PCE value is smaller than the reference value, the determination unit 150 may determine that the reconstructed inspection target image is captured by a device that is not the same as the reference image, that is, the manipulation is performed.

In addition, the determination unit 150 may estimate whether there is an area (cut out area or deleted area) extracted from the reconstructed inspection target image by using the position of the peak in the correlation plane.

That is, the determination unit 150 determines whether the position of the peak is at the reference position, and when the position of the peak is not at the reference position, the determination unit 150 may determine that at least a part of the inspection target image is cut out.

Here, the reference position is a position of (1, 1) in the correlation plane, and the determination unit 150 manipulates the inspection target image when the position of the peak is not at the position of (1, 1). Can be determined.

In addition, a portion cut out of the inspection target image may be located at a lower right side of the region where the peak is located.

This uses the characteristic that the position of the upper left of the region partially extracted from the inspection target image is the same as the position of the peak of the correlation plane.

Therefore, the region partially extracted from the inspection target image may be represented by Equation 7 below.

[Equation 7]

은 첨두치의 위치를 의미한다.Here, w is the rate of change of the size of the inspection target image, M is the horizontal size of the inspection target image, N is the vertical size of the inspection image,

Means the position of the peak.

In addition, the determination unit 150 may estimate a partially deformed region of the inspection target image by using partial similarity between the PRNU value extracted from the restored inspection target image and the reference PRNU value.

If the noise of the synthesized portion of the PRNU value is weak or is not synchronized with the reference PRNU value, the determination unit 150 may estimate the partially deformed region by using a very low correlation value.

As described above, the image manipulation detection apparatus according to an embodiment of the present invention uses a PRNU of the inspection target image and the reference image and a correlation plane using the same, and thus the photographing apparatus is changed (source operation) or partially deleted. It is possible to detect whether the inspection target image has been manipulated, for example, or partially deformed.

Hereinafter, an image manipulation detection method and a specific embodiment will be described with reference to FIGS. 2, 3, 4, and 5.

2 is a conceptual diagram illustrating an image manipulation detection method according to an exemplary embodiment, and FIGS. 3, 4, and 5 are conceptual views illustrating an manipulated image.

First, referring to FIG. 2, in the image manipulation detection method according to the present invention, the size change rate of the inspection target image is calculated and the size of the inspection target image is restored so that the size of the inspection target image corresponds to the calculated size change rate. It becomes (S210).

The size change rate may be a pixel function (p (i, j)) of the inspection target image, and the pixel value of the pixel function may have a value between 0 and 255. Here, i and j are variables representing positions of specific pixels in the image.

) 분산의 주기적인 확률적 특성을 이용하여 산출될 수 있다.The rate of change of magnitude is the second derivative of the pixel function p (i, j) (

) Can be calculated using the periodic probabilistic nature of the variance.

)를 연산한다.That is, as described above with reference to FIG. 1, in the restoring operation (S210), the second derivative of the pixel function p (i, j) is second-derived to obtain a second derivative (

) Is calculated.

[Equation 1]

)에 대한 분산(

)을 추정하고, 상기 분산(

) 의 주기성을 이용하여, 검사대상 영상의 크기 변화율을 분석한다.Further, in the restoring step (S210), as shown in Equation 2 below, the second derivative (

For variance (

), And the variance (

), The rate of change of the size of the inspection target image is analyzed.

&Quot; (2) "

(Where R is a value corresponding to the horizontal size of the image).

) 에서 주기성을 분석하기 위하여, 분산(

)을 푸리에 변환(Fourier transform)시킨다. On the other hand, the restoring step (S210) the dispersion (

In order to analyze the periodicity in

) Is Fourier transform.

)이 주기성이 있는지 알아낼 수 있다.The Fourier transform is used to find a periodicity, and by comparing the peak value derived through the Fourier transform with a reference value of the Fourier transform,

) Can determine if it is periodic.

)은 주기성을 갖는다고 표현할 수 있고, 나아가 상기 검사대상 영상이 크기변환되었다고 판단될 수 있다.As a result of the comparison, when the derived peak is greater than the reference value, the variance (

) May be expressed as having periodicity, and furthermore, it may be determined that the inspection subject image is scaled.

On the other hand, in the step of restoring (S210) using the position of the peak (peak) derived through the Fourier transform, the size conversion rate (W) is calculated.

Since the peak position derived from the Fourier transform represents a frequency, the size conversion ratio W of the inspection target image may be calculated as shown in Equation 3 below.

&Quot; (3) "

As described above, the reconstructor 110 may equally adjust the reference image and the sync by reconstructing the inspection target image to the original size using the calculated size conversion ratio W.

Here, the reference image is an image obtained from a reference device that is assumed that the inspection target image has been taken. In the present invention, by comparing the attribute values of the inspection target image with the attribute values of the reference image, It can be detected whether it has been manipulated.

In step S210, when the process of restoring the size of the inspection target image is completed, extracting a PRNU value from the restored inspection target image (S220) is performed.

In the extracting of the PRNU value (S220), a PRNU (Photo Response Non-Uniformity) value is extracted from the inspection target image reconstructed corresponding to the calculated size conversion rate (W).

Here, the PRNU (Photo Response Non-Uniformity) value is a value representing the degree of pixel response nonuniformity, and may be extracted by applying a wavelet-based filter to the reconstructed inspection target image.

In the extracting of the PRNU value (S220), the noise is extracted from the reconstructed inspection target image by using the wavelet-based filter, and a maximum likelihood estimation method (MLE) is extracted to the extracted noise. By applying a maximum likelihood estimation, the PRNU of the reconstructed inspection target image may be extracted as shown in Equation 4 below.

&Quot; (4) "

는 검사대상 영상이 동영상인 경우, 동영상을 이루는 복수개의 정지영상 중 k번째 정지영상의 노이즈 residual이고,

는 k번째 복원된 검사대상 영상에 웨이블릿 기반 필터를 적용한 후 노이즈가 제거된 frame이다.From here,

Is the noise residual of the kth still image among the plurality of still images constituting the moving image when the inspection target image is a moving image,

Is a frame from which noise is removed after applying a wavelet-based filter to the k-th reconstructed inspection target image.

In addition, the wavelet-based filter here uses a seismic pulse composed of a plurality of cycles, and the maximum likelihood estimation method is a method of estimating an unobservable parameter (parameter) based on an observed sample. .

On the other hand, in the step of extracting the PRNU value (S220), even if the restored inspection target image and the reference image is an infrared image instead of an RGB image, for the detection accuracy of whether the inspection target image is manipulated, the reference PRNU of the reference apparatus A value, that is, a reference PRNU value corresponding to the reference image photographed by the reference apparatus, may be converted into a MACE (Minimum Average Correlation Energy) filter.

Here, the MACE filter may be designed to minimize the correlation plane energy and derive a correlation plane having a high peak.

On the other hand, the MACE filter may be expressed as shown in Equation 5 described above with reference to FIG.

&Quot; (5) "

Here, H is a MACE filter, X is a one-dimensional vector form of a reference PRNU, + is a transpose matrix, u is a predefined correlation constant value, k is an index of an X array, and is represented in capital letters. Means that it is Fourier transformed.

That is, here, if the size of the X array is K, k is 1 ≦ k ≦ K.

As described above, when the PRNU value of the reconstructed test target image is extracted in the step of extracting the PRNU value (S220), the similarity between the extracted PRNU value and the reference PRNU value is compared using the reference PRNU value of the reference device. The process proceeds (S230).

In the comparing step (S230), in order to compare the similarity between the PRNU value extracted from the reconstructed inspection target image and the reference PRNU value, the cross correlation between the extracted PRNU value and the reference PRNU value is calculated, and the correlation plane. Can be derived.

Here, the correlation means when one side increases between two variables, the other side tends to increase (or decrease). In addition, a numerical value representing the degree of correlation between two variables x and y is called a correlation coefficient. Here, two variables x and y may be extracted PRNU values and reference PRNU values, respectively.

As such, when the correlation plane is derived in the comparing step 230, calculating the PCE value using the correlation vudausdmf is performed (S240).

In the calculating of the PCE value (S240), personal consumption expecditures (PCE), which may be represented by Equation 6 described above with reference to FIG. 1, may be calculated.

&Quot; (6) "

는 상관관계 평면에서의 첨두치를 나타낸다.Where CC refers to the correlation plane,

Represents the peak in the correlation plane.

As described above, when the PCE value of the reconstructed inspection target image is calculated, comparing the calculated PCE value and the PCE value of the reference apparatus, and determining whether the device on which the reconstructed inspection target image is photographed is the reference apparatus. It proceeds (S250).

In the determining operation S250, when the calculated PCE value is equal to or larger than a reference value, it may be determined that the reconstructed inspection target image is captured by the same device as the reference image. In addition, on the contrary, when the calculated PCE value is smaller than the reference value, it may be determined that the reconstructed inspection target image is captured by the device that is not the same as the reference image, that is, the manipulation.

For example, referring to FIG. 3, the first image 210 is an inspection target image and the second image 220 is a reference image of the reference device. Although the illustrated first and second images 210 and 220 appear to be the same image, it can be seen that the positions of the chairs shown in the first and second regions 210a and 220a are different.

That is, the first and second images are images captured by different photographing apparatuses (cameras or CCTVs) at similar places, that is, images having different sources.

In the determining step S250, as illustrated in FIG. 3, it may be determined whether the inspection target image is captured by a device different from the reference image.

In the determining step S250, the position of the peak value in the correlation plane may be used to estimate whether there is an area (cut out area or deleted area) extracted from the restored inspection target image.

That is, in the determining step S250, it is determined whether the position of the peak is at the reference position, and when the position of the peak is not at the reference position, it is determined that at least a part of the inspection target image is cut out.

Here, the reference position is a position of (1, 1) in the correlation plane, and the determination unit 150 manipulates the inspection target image when the position of the peak is not at the position of (1, 1). Can be determined.

In addition, a portion cut out of the inspection target image may be located at a lower right side of the region where the peak is located. This uses the characteristic that the position of the upper left of the region partially extracted from the inspection target image is the same as the position of the peak of the correlation plane.

Therefore, the region partially extracted from the inspection target image may be represented by Equation 7 below.

[Equation 7]

은 첨두치의 위치를 의미한다.Here, w is the rate of change of the size of the inspection target image, M is the horizontal size of the inspection target image, N is the vertical size of the inspection image,

Means the position of the peak.

For example, referring to FIG. 4, the first image 210 is a reference image of the reference device, and the second image 220 is an inspection target image. It can be seen that the first and second regions 210a and 220a of the illustrated first and second images 210 and 220 are different.

That is, the partial image of the first region 210a of the first image is cut out in the second image 220. In the determining operation S250, it may be determined that the second image 220 has been manipulated. The position of the manipulated second region 220a can also be grasped.

In operation S250, the partially deformed region of the inspection target image may be estimated using partial similarity between the PRNU value extracted from the restored inspection target image and the reference PRNU value.

That is, in the determining operation S250, when the noise of the synthesized portion of the PRNU value is weak or is not synchronized with the reference PRNU value, the partially deformed region may be estimated by using a very low correlation value. Can be.

For example, referring to FIG. 5, the first image 210 is a reference image of the reference device, and the second image 220 is an inspection target image. It can be seen that the first and second regions 210a and 220a of the illustrated first and second images 210 and 220 are different.

That is, in the second image 220, the object included in the first area 210a of the first image is erased, that is, a part of the image is deformed. In the determining operation S250, the second image ( 220 may be determined to have been manipulated.

As described above, the image manipulation detection method according to an embodiment of the present invention may detect whether the inspection target image is manipulated by using the PRNU of the inspection target image and the reference image and a correlation plane using the same.

Hereinafter, an experimental result of detecting whether an inspection target image is manipulated using the image manipulation detection method according to the present invention will be described with reference to FIGS. 6, 7, 8, and 9. 6 and 7 are conceptual views illustrating the accuracy of the image manipulation detection method according to an embodiment of the present invention.

6 and 7 illustrate an accuracy of discriminating a photographing apparatus of a security image photographed with RGB and infrared rays, respectively.

In order to measure the accuracy of discriminating the photographing apparatus, the experiment was performed by increasing the size conversion ratio by 0.1 from 1.8 to 1.8.

The size conversion ratio was estimated by the method disclosed herein, and the similarity measurement was performed by applying the normalized cross correlation (NCC), the NCC after applying the codec noise reduction filter proposed by Jessica, and the minimum average correlation energy (MAC) method disclosed in the present invention.

In addition, the threshold value of the PCE value for whether two images are acquired by the same imager is set to be less than 1.0% of the false positive rate (FPR).

Looking at the experimental results shown in Figures 6 and 7 it can be seen that the present invention achieves up to 50% more than the NCC method, 10% or more better than the NCC-Filter.

8 and 9 are conceptual views illustrating a method of detecting whether a security image is manipulated using an image manipulation detection method according to an exemplary embodiment.

First, FIG. 8 shows the results of the partial extraction region estimation experiment. The experiment was carried out by resizing the captured image with size conversion ratio 1.2 and partially extracting the original image size at random position. The accuracy was measured by averaging the difference between the actual coordinates and the estimated coordinates of the upper left coordinate of the partial extraction region. Experimental results show that the proposed technique estimates the partial extraction region with an error of less than 1 pixel.

Next, FIG. 9 shows the results of the partial deformation region estimation experiment. In order to measure the accuracy of the experimental results, the concept of Precision and Recall used in the field of radiation transfer deformation detection was used. Precision indicates how precisely the partial strain area was found. And Recall shows how many percentages of partially deformed regions were found. Since Precision and Recall are trade-offs, we measure the accuracy of the proposed algorithm with the harmonic mean of the two factors. The experimental results show that the partial deformation region is detected with a harmonic mean of 81%.

As described above, the image manipulation detecting method and the image manipulation detecting apparatus according to the present invention extract a PRNU value from the inspection target image, derive a correlation plane based on the detection, and detect whether the inspection target image is manipulated. Can be.

In addition, the image manipulation detection method and the image manipulation detection apparatus according to an embodiment of the present invention can detect whether the security image photographed by the CCTV has been manipulated, thereby preventing crime caused by the manipulation of the security image.

100: image manipulation detection device 110: restoring unit
120: PRNU extraction unit 130: comparison unit
140: PCE calculation unit 150: determination unit

Claims (5)

In the method of detecting the operation of the image,
Reconstructing the inspection target image by calculating a size change rate of the inspection target image so that the size of the inspection target image corresponds to the calculated size change rate;
Extracting a Photo Response Non-Uniformity (PRNU) value from the restored inspection target image;
Comparing a similarity between the extracted PRNU value and the reference PRNU value using a reference PRNU value of a reference device;
Calculating a peak to correlation energy (PCE) value using the correlation plane derived through the comparison result; and
And comparing the calculated PCE value with the PCE reference value of the reference device to determine whether the device on which the inspection target image is captured is the reference device. The method of claim 1,
And estimating whether there is an area extracted from the inspection target image by using a position of a peak in the correlation plane. The method of claim 1,
Estimating a deformed region in the inspection target image by using the extracted PRNU value and the partial similarity of the reference PRNU. The method of claim 1,
The rate of change of size
The second derivative of the pixel function p (i, j) corresponding to the inspection subject image represented by Equation 1 below (

Is calculated using the variance of and calculated using the Equation 2 below.

And estimating the rate of change of the size of the inspection target image by analyzing the periodicity.
[Equation 1]

&Quot; (2) "

(Where R is a value corresponding to the horizontal size of the image) The method of claim 1,
And the reference PRNU value is converted into a MACE filter in order to minimize energy of the correlation plane and derive a correlation plane having a high peak.

Images