CN106023209A - Blind detection method for spliced image based on background noise - Google Patents

Abstract

The invention relates to a blind detection method for a spliced image based on background noise. Secondary noise adding is carried out on the spliced image, correlation coefficients of a corresponding subblock before and after secondary noise adding are estimated, the correlation coefficients are classified according to a selected threshold, and splicing areas are positioned according to different classifications. Compared with a present algorithm, the method of the invention needs not to know a noise database or information of an original image in advance, and the practicality is high. The method of the invention has the advantages of simple principle and low computation amount. Experiment results show that the method can be used to discriminate splicing images with different background noises effectively, and has high robustness for typical post-treatment.

Description

A Blind Detection Method of Mosaic Image Based on Background Noise

technical field

The invention relates to a method for blind detection of mosaic images based on background noise, and belongs to the technical field of image recognition.

Background technique

With the increasingly powerful functions of image processing software such as Photoshop, people's willful tampering of digital images is increasing day by day, and it is difficult to detect, which brings a serious crisis of trust in digital images. PS (Photoshop) has been deduced from a software name into a verb for modifying and embellishing images. With the help of such image processing tools, users can create almost realistic doctored images using only a stitching operation. Therefore, it is of urgent practical significance to study simple and practical splicing and tampering blind forensics technology.

Most of the existing blind forensics methods for mosaic images start from detecting a certain type of tampering, and realize forensics by identifying similarities and differences of features. For example, most of the blind forensics methods that have been proposed at this stage are based on traceology, and identify the authenticity of images by identifying the abnormal characteristics of different traces. Stitching tampering can be identified by detecting traces of inconsistent resampling factor, blur type, light direction, JPEG quantization coefficient, and sharpening intensity in the background area and the stitching area. However, most of these methods assume that the forensic examiner knows the forensic characteristics in advance. During the imaging process of the image, some inherent feature information of the imaging device will be brought into the natural image, and these inherent features can be regarded as having "ballistic" characteristics. Stitched images can also be detected, for example, by detecting inconsistencies in “ballistic” properties such as camera response functions, lens chromatic aberration, color filters, and pattern noise in different regions of the image. However, this type of method needs the support of comparing the training sample database.

Contents of the invention

The purpose of the present invention is to overcome the defects of the above-mentioned prior art and propose a method for blind detection of mosaic images based on background noise.

In order to achieve the above object, the blind detection method of mosaic image based on background noise proposed by the present invention, the steps are as follows:

Step 1: Perform noise-adding processing on the image to be detected to obtain a noise-added image, the noise used in the noise-adding processing is Gaussian white noise, the mean value of the Gaussian white noise is 0, and the variance is 10 ^-n , n∈{3 ,4,5};

Step 2: Use a window of k*k size to extract the sub-blocks of the image to be detected and the image after adding noise, the distance of each movement of the window is d grid units, and several pairs of sub-blocks of the image to be detected are obtained and the sub-block image pair formed by the image sub-block after adding noise, the two images in the sub-block image pair correspond to the positional relationship, where k ∈ {4, 8, 16}, d ∈ {2, 4, 8} , and d≤k;

Step 3, calculating the correlation coefficients of the two images in all sub-block image pairs;

Step 4: Calculate the variance of the obtained correlation coefficient and compare it with the threshold value V. If the variance is greater than the threshold value, it means that there is a splicing operation in the image. When n=3, V takes 4*10 ^-2 , and when n=4, V takes 5* 10 ^-5 , when n=5, V takes 1*10 ^-4 ;

Step 5, for images with splicing operations, the correlation coefficient is classified according to size, and one major category and at least one small category are obtained;

Step 6: The area corresponding to the small class on the image to be detected is the stitching area.

The further improvement of the present invention is:

1. The correlation coefficient ρ(X, Y) of the two sub-block images corresponding to the positional relationship is obtained by the following formula:

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

In the formula, X _mn is the image signal value of the pixel in row m and column n in the sub-block image before adding noise, is the image signal mean value of the sub-block image before adding noise, Y _mn is the image signal value of the mth row and nth column pixel in the sub-block image after adding noise, Y is the image signal mean value of the sub-block image before adding noise, m, n ∈ {1, 2, ,,, k}.

2. The classification method in the 5th step is to add the correlation coefficient with a value greater than or equal to T to the large class, and add the correlation coefficient less than T to the small class, wherein when n=3, T is 0.3, and when n=4, T is 0.8 , when n=5, T is taken as 0.9;

3. The classification in the fifth step is automatically realized by a clustering method.

4. As a preferred solution, in the first step, the variance of Gaussian white noise is 10 ^-4 , and in the second step, k=8, d=4.

The method for splicing image blind detection based on background noise in the present invention first adds noise twice to be detected, secondly calculates the correlation coefficients of the corresponding sub-blocks before and after the second noise addition, and then classifies the correlation coefficients according to the size of the selected threshold, and finally according to The different taxa locate spliced regions. The invention does not need the prior information of the known noise database and the original image, and has strong practicability. And it has the advantages of simple principle and small amount of calculation. Experimental results show that the method can effectively identify stitched images with different background noises, and at the same time, the method is robust to some typical post-processing operations.

Below is the derivation process of principle of the present invention:

For the sake of brevity, we first assume that only the tampered region (region II) of the mosaic image contains noise, and then extend it to the case where the non-tampered region (region I) and the tampered region (region II) have different background noises.

A. Tampering area (area II) situation:

Before the second noise addition, we regard the region II in the stitched image _y1 as an additive noise model:

y _II =x _f +w (1)

Among them, y _II represents the noisy tampered area, x _f represents the original image, and w represents Gaussian white noise with mean value 0 and variance σ ₁ .

Region II in image y ₂ after overall noise addition can still be regarded as an additive noise model:

y _nII ＝x _f +w+n (2)

In the formula, y _nII is the area of secondary noise addition, and n is the mean value of 0 and variance of _σ2 Gaussian white noise artificially added twice.

Now randomly select the sub-blocks of equal size in the corresponding position in the area II before and after the second noise addition and Its correlation can be described as formula (3).

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

B. Untampered area (area I) situation:

Before overall noise is added, area I is a noise-free area. After secondary noise addition, the region I in the y ₂ image can also be regarded as an additive noise model:

y _nI =x _t +n (4)

Among them, y _nI is the noise-added background area, and x _t is the original background area.

Also extract the sub-blocks corresponding to the positions in the region I before and after the second noise addition and Its correlation can be described as formula (5).

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

In the process of image tampering, in order to make the stitched tampered area not attract visual attention, one of the methods is to make the tampered area have smaller grayscale values. At this time, in formula (3) and It can be approximately regarded as having the distribution characteristics of Gaussian white noise, and it is known that any two Gaussian white noises are independent of each other (the correlation coefficient is 0), so the correlation of formula (3) is also very small, which can be written as formula (6 ).

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; w</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Right\; Arrow;</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

Based on the above formula, we can get:

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; I</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; I</span>}}^{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

It can be seen from formula (7) that the second noise is added to the mosaic image containing background noise in the tampered area, and the correlation of the corresponding tampered area before and after the second noise addition is smaller than that of the untampered area, that is, according to the similarity and difference of the correlation coefficient to identify tampered areas.

In the above analysis, we assume that only the tampered area contains noise in the mosaic image. If the untampered area also has a certain amount of background noise, the difference between the two still exists after the second noise addition. Therefore, the above correlation analysis method is still Similarities and differences between tampered and non-tampered areas can be identified.

Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings.

FIG. 1 is a flow chart of the method for blind detection of mosaic images based on background noise in the present invention.

Figure 2(a) is the original image 1.

Figure 2(b) is the original image 2.

Figure 2(c) is the stitched image.

Figure 3(a)-Figure 3(d) are comparisons of detection results under different blocking strategies.

Figure 4(a)-Figure 4(c) is a comparison of detection results under different noise levels.

Fig. 5(a) is a diagram of the detection result of the method of this embodiment.

Figure 5(b) is a diagram of the detection results of the forensics method based on K-SVD dictionary learning.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

As shown in Figure 1, it is a flow chart of the method for splicing image blind detection based on background noise of the present invention, which specifically includes the following steps:

Step 1: Perform noise-adding processing on the image to be detected to obtain a noise-added image. The noise used in the noise-adding processing is Gaussian white noise with an average value of 0 and a variance of 10 ⁻⁴ . Figure 2(a) is the original image 1. Figure 2(b) is the original image 2. Figure 2(c) is the stitched image. In this step, noise is added to Fig. 2(c).

Step 2: Use a window of size 8*8 to extract the sub-blocks of the image to be detected and the image after adding noise respectively. The moving distance of the window is 4 grid units each time, and several pairs of sub-blocks of the image to be detected are obtained. and the sub-block image pair formed by the noise-added image sub-blocks, and the positional relationship of the two images in the sub-block image pair corresponds.

Step 3, calculating the correlation coefficients of the two images in all sub-block image pairs;

The correlation coefficient ρ(X, Y) of the two sub-block images corresponding to the positional relationship is obtained by the following formula:

$\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; )</span>}{\sqrt{<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}\underset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; Y</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; Y</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}$

In the formula, X _mn is the image signal value of the pixel in row m and column n in the sub-block image before adding noise, is the image signal mean value of the sub-block image before adding noise, Y _mn is the image signal value of the mth row and nth column pixel in the sub-block image after adding noise, Y is the image signal mean value of the sub-block image before adding noise, m, n∈{1, 2,,,,k}, k is the side length of the window, k=8 in this example.

Step 4: Calculate the variance of the obtained correlation coefficient and compare it with the threshold value V. If the variance is greater than the threshold value, it means that there is a splicing operation in the image, and V is 5*10 ^-5 .

Step 5. For images with splicing operations, the correlation coefficient is classified according to size. In this example, the classification is automatically realized by clustering methods (such as K-means clustering, hierarchical clustering with minimum distance, etc.), and a large image is obtained. class and at least 1 subclass. In this example, the image sub-blocks with a correlation coefficient value greater than or equal to 0.8 are added to the main category, and the image sub-blocks with a correlation coefficient value less than 0.8 are added to the sub-category.

Step 6: The area corresponding to the small class on the image to be detected is the stitching area. The test results are shown in Figure 5(a).

As shown in Figure 3(a)-Figure 3(d), they are comparison charts of detection results under different block strategies.

It can be seen from the comparison that when the sub-block is large (16*16), the resolution of the detection result is low and the error is large, as shown in Figure 3(c); when the sub-block is small (4*4) the detection The result is finer, as shown in Fig. 3(d), but the operation time is long. When the middle size sub-block is 8*8, the stitching area can also be accurately located, as shown in Figure 3(a) and Figure 3(b), but the detection result is more accurate when overlapping blocks are selected than when non-overlapping blocks are selected. accurate. Therefore, it is necessary to choose a relatively compromised block strategy according to the reality of the image to be detected and the tampering target.

Similarly, choosing different variances will also affect the detection results of the algorithm. Fig. 4(a)-Fig. 4(c) show the comparison of the results when the variances are selected in turn as 10 ^-2 , 10 ^-4 , and 10 ^-6 under the overlapping block strategy (k=8, d=4).

It can be seen from the figure that when the variance is 10 ^-2 , the interference error of the result map is relatively large; when the ^variance is 10 ^-4 , the detection result is the most accurate; The area is less obvious. Therefore, the value of the variance should be fine-tuned continuously for splicing tampered images with different background noises. Usually, the variance of 10 ^-5 -10 ^-3 is more appropriate.

Tests show that the present invention still has a relatively high recognition success rate for compressed and spliced spliced images with blurred edges, and the method of the present invention is not sensitive to compression and blurred edges, and has strong practicability.

The spliced image detection method of this embodiment and the evidence collection method based on K-SVD dictionary learning (Wang Wei, Zeng Feng, Duan Xintao, etc. Blind detection of synthetic images based on K-SVD dictionary learning. Journal of Wuhan University (Natural Science Edition), 2013, 59(5):499-504.) (see Figure 5(b) for the detection results), it does not require prior information of the known noise database and the original image, and has strong practicability. It can be seen from the comparison results that when the quadratic noise variance in the algorithm in this paper and the noise standard deviation in the comparison algorithm are selected properly, the detection results of the two algorithms are equivalent. However, the algorithm of the method of the invention is simple, the amount of calculation is small, and the detection speed is faster.

In addition to the above-mentioned embodiments, the present invention can also have other implementations. All technical solutions formed by equivalent replacement or equivalent transformation fall within the scope of protection required by the present invention.

Claims (5)

1. a stitching image blind checking method based on background noise, it is characterised in that: specifically include following steps:

1st step, image to be detected is added process of making an uproar, obtains the image added after making an uproar, described in add process of making an uproar noise used be height This white noise, the average of this white Gaussian noise is 0, and variance is 10^-n, n ∈ { 3,4,5}；

2nd step, use the window of k*k size respectively to image to be detected with add the image after making an uproar and carry out sub-block extraction, described window Mouthful every time the distance of movement is d grid cell, it is thus achieved that some to by image subblock to be detected with add make an uproar after image subblock constitute Sub-image pair, two width image positional relationship of described sub-image centering are corresponding, wherein, k ∈ 4,8,16}, d ∈ 2,4, 8}, and d≤k；

3rd step, calculate the correlation coefficient of all sub-image centering two width images；

4th step, the variance of calculating gained correlation coefficient, and compare with threshold value V, if variance exists more than threshold value explanatory diagram picture Concatenation, during n=3, V takes 4*10^-2, during n=4, V takes 5*10^-5, during n=5, V takes 1*10^-4；

5th step, for there is the image of concatenation, described correlation coefficient is classified according to size, it is thus achieved that 1 big class and extremely Few 1 group；

6th step, the described group region corresponding on image to be detected is splicing regions.

Stitching image blind checking method based on background noise the most according to claim 1, it is characterised in that: position relationship The correlation coefficient ρ (X, Y) of two corresponding sub-images is tried to achieve by below equation:

In formula, X_mnFor add make an uproar before the image signal value of m row the n-th row pixel in sub-image,For add make an uproar before sub-image Picture signal average, Y_mnFor add make an uproar after the image signal value of m row the n-th row pixel in sub-image,For add make an uproar before sub-block figure The picture signal average of picture, m, n ∈ 1,2,, k}.

Stitching image blind checking method based on background noise the most according to claim 1, it is characterised in that: the described 5th Sorting technique in step is, will value more than or equal to T correlation coefficient add big class, the addition group less than T, wherein during n=3, T When taking 0.3, n=4, when T takes 0.8, n=5, T takes 0.9.

Stitching image blind checking method based on background noise the most according to claim 1, it is characterised in that: the described 5th Classification in step is automatically obtained by clustering method.

Stitching image blind checking method based on background noise the most according to claim 1, it is characterised in that: the described 1st In step, the variance of white Gaussian noise is 10^-4, in the 2nd step, k=8, d=4.