WO2022000804A1 - Security check image artifact removing method using region search and pixel value suppression - Google Patents

Abstract

A security check image artifact removing method using region search and pixel value suppression. The method comprises the following steps: calculating the center row coordinates of the head of a human body; determining artifact pixels between the two legs; determining artifact pixels between the arms and the trunk; and suppressing pixel values of the artifact pixels to obtain a security check image without artifacts. According to a fixed posture during human body security check, a region search method is used to search a fixed region for artifacts between the two legs and artifacts between the arms and the trunk, which is conducive to saving on computer resources and search time, and improving the artifact search efficiency.

Description

An Artifact Removal Method for Security Inspection Images Using Region Search and Pixel Value Suppression technical field

The invention relates to the technical field of image processing, in particular to a method for removing artifacts in a security inspection image using area search and pixel value suppression.

Background technique

In recent years, the international security situation has become increasingly severe, and stricter security inspections have been arranged at the entrances of crowded places such as airports and stations. Due to the continuous improvement of millimeter wave imaging technology and the introduction of domestic security inspection product specifications, millimeter wave human body security inspection has huge application prospects. According to the characteristics that millimeter waves can penetrate clothing, dangerous goods hidden on the body can be displayed and identified through millimeter wave imaging and pattern recognition technology. However, due to the multiple scattering of electromagnetic waves on the human body, obvious artifacts will be introduced in the imaging results, mainly between the human body's legs, arms and torso, etc., which seriously affects the display effect of human security images and dangerous goods. Identify the results. The main task of artifact removal is to suppress or remove the artifacts in the image after imaging, so as to ensure the accuracy of the security inspection image display.

There are usually three ways to eliminate the artifact information of human security inspection images. First, use polarization characteristics to eliminate secondary reflection artifacts. American PNNL company eliminates secondary reflection artifacts by sending and receiving the polarization state of millimeter waves. The transmitting antenna of the security detector adopts left-handed polarization, while the receiving antenna adopts right-handed polarization. According to a single reflection, the circular plan direction will be changed and the second The reflection does not change the direction of the circular polarization, and the artifacts caused by the secondary reflection are eliminated through the polarization characteristics; secondly, the posture of the inspected person is standardized to eliminate the artifacts. Since the artifact is caused by the multiple reflections of the two objects at close range, the inspected person is required to check in a fixed posture during the security inspection, with their feet separated and their hands a certain distance from the torso; third, the artifact based on image processing Shadow removal method. Based on the security inspection image, the artifacts on the image are removed by filtering or matting.

There are two main categories of CT image artifact removal methods based on image processing, including CT image correction-based and sinogram-based artifact removal methods. Wu P. et al. used a method based on CT image correction to remove artifacts. This method is aimed at the annular artifact of CT image, transforming the image from the rectangular coordinate system to the polar coordinate system, and the annular artifact is transformed into a straight line. The linear artifacts are removed by global filtering or local filtering, and finally the coordinate system is converted to a rectangular coordinate system to restore the CT image. This type of method needs to transform the coordinate system multiple times, and the interpolation algorithm used affects the image resolution and requires a large amount of computation; Ashrafuzzaman A. et al. The sinogram appears as a straight line or a curve, and the artifacts are removed by filtering, and the projected sinogram is converted into a normal CT image.

Different from CT images, human security inspection images do not have ring artifacts. The artifact shapes are mostly straight lines or curves, and exist between the arms torso and between the legs. The intensity and shape of the artifacts depend on the posture of the inspected person. Most of the existing artifact removal methods are based on CT images, and it is difficult to directly apply the artifact removal methods based on CT images in human security inspection images.

In short, the problems existing in the prior art are:

1. The artifact removal method based on polarization characteristics can only eliminate the artifacts caused by secondary reflection at present, but cannot completely eliminate the artifacts in the image;

2. The artifact removal method based on standardizing the posture of the inspected person cannot completely eliminate the artifact by standardizing the posture of the inspected person due to the different shapes of people and the inability to strictly limit the posture of the security inspection;

3. Artifact removal methods based on image processing, most of these methods focus on the artifact removal of CT images, and have little reference value for the removal of artifacts from human security inspection images.

SUMMARY OF THE INVENTION

In order to solve the above problems, the present invention proposes a method for removing artifact from a security inspection image using area search and pixel value suppression, which includes the following steps:

Calculate the row coordinates of the center of the human head:

Accumulate and sum the pixel values of all row pixels in the set head a priori search area, take the row coordinate of the cumulative sum greater than the threshold as the head row coordinate, and take the average of the head row coordinates as the center of the human head row coordinates;

Determine the artifact pixel between the legs:

Determine the artifact prior search area between the legs, and determine whether the pixels in the artifact prior search area between the legs are artifact pixels between the legs through the first multiple threshold method, so as to obtain the artifact pixel coordinates between the legs; The row search area of the inter-leg artifact prior search area is calculated according to the row coordinates of the head center and the body scale, and the column search area of the inter-leg artifact prior search area is obtained by dividing the center line of the leg area image. The column coordinates are obtained by extending a fixed column of pixels to the left and right, and the column coordinates of the leg center line of the security inspection image are obtained by the flip translation difference method;

Determining artifact pixels between arms and torso:

Determine the prior search area of the artifact between the left and right arm torso according to the head center row coordinate and body proportion, and use the second multiple threshold method and the third multiple threshold method to determine the prior search area of the left and right arm torso artifacts respectively. Whether the pixel of is the artifact pixel between the arm and torso, get the coordinates of the artifact pixel between the arm and torso;

Suppress the pixel value of the artifact pixel:

Take out each artifact pixel and the pixel neighborhood sequence of its row, construct an artifact suppression function for artifact suppression, and then replace the original pixel sequence with the suppressed pixel sequence to obtain an artifact-removed security inspection image; Shadow pixels include artifact pixels between legs and artifact pixels between arms and torso.

Further, the flip-translation difference method is specifically as follows: cyclic shift and left-right flipping are performed on the leg region image, and difference is made, and the image difference after the difference is cumulatively summed to obtain the cumulative sum of image differences with different shift numbers. A vector, and the column coordinates of the center line of the image of the leg area are calculated by the shift number corresponding to the minimum value in the vector.

Further, the row search area of the inter-leg artifact prior search area is [L _RowBgn , N _Row ], where N _Row is the number of rows of the security inspection image, and L _RowBgn is the starting row of the inter-leg artifact Coordinates, L _RowBgn = round(N _Row -(N _Row -H _RowCtr )·ρ), round means rounding, H _{RowCtr is the} center row coordinate of the human head, and ρ is the ratio of leg length to height in the security image;

The column search area of the artifact a priori search area between the legs is [L _ColCtr -L _Width , L _ColCtr +L _Width ], where L _Width is the column coordinate of the leg center line of the security inspection image L _ColCtr to the left and right neighbors The number of extended pixel columns;

The calculation formula of the _{coordinate L ColCtr} of the symmetry axis of the leg image is:

Among them, round means rounding, and arcmin means the n value that makes f( ) get the minimum value;

f(n) is the value of the flip translation difference function, and its specific calculation formula is:

f(n)=sum(abs(circshift(I _Leg ,n)-fliplr(circshift(I _Leg ,n))))

Among them, circshift represents the horizontal cyclic shift operation, fliplr represents the left and right flip operation, sum represents the cumulative summation, abs represents the absolute value operation, and the value range of n is _{an integer in [-N Shift} , N _Shift ], then f(n ) is _{a one-dimensional vector containing 2N Shift} + 1 values, the shift number n corresponding to the minimum value in f(n) can represent the deviation of the symmetry axis of the leg area image from the center line of the security image column, and N _Shift represents the preset the first offset of ;

The described first multiple threshold judgment method is specifically, if the pixel value I _Leg (x, y) of the pixel in the two-leg artifact a priori search area satisfies the following three conditions simultaneously, then the pixel value I _Leg (x, y) ) corresponds to the pixel of the artifact between the legs:

(1) The pixel value I _Leg (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(2) I _Leg (x, y) < Th _Artifact ;

(3) I _Leg (x, y) < V _SmoothLeft +Th _Artifact /3 and I _Leg (x, y) < V _SmoothRight +Th _Artifact /3;

Among them, x and y represent the pixel row coordinates and column coordinates, respectively, and their value range satisfies the coordinate range of the artifact prior search area between the legs; Th _Artifact is the set artifact pixel threshold; V _SmoothLeft and V _SmoothRight are double The mean value of the smooth areas on the left and right sides of the inter-leg artifact prior search area, the row coordinate range of the left and right smooth areas is the same as the row search area of the double-leg artifact prior search area, and the column coordinate ranges of the left and right smooth areas are respectively

[L _ColCtr -L _Width -L _Smooth , L _ColCtr -L _Width -1] and [L _ColCtr +L _Width +1, L _ColCtr +L _Width +L _Smooth ], L _Smooth is the preset second offset.

Further, the row coordinate interval of the artifact a priori search area between the arms and torso is [H _RowCtr + S _RowUp , H _RowCtr + S _RowDown ], and the column coordinate interval of the artifact a priori search area between the left and right arms torso is respectively [H RowCtr + S RowUp , H RowCtr + S RowDown ] for

and

where S _RowUp is the distance between the starting row coordinate of the artifact between the arm torso and the head center row coordinate H _RowCtr , and S _RowDown is the distance between the ending row coordinate of the arm torso artifact and the head center row coordinate H _RowCtr ;

The second multi-threshold decision method in particular, if the pixels between the left arm torso artifact prior search area pixel value _I ArmLeft (x, y) satisfies the following three conditions, the pixel value _I ArmLeft (x, y) The corresponding pixel is the artifact pixel between the arm and torso:

(1) The pixel value I _ArmLeft (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(2) I _ArmLeft (x, y) < Th _Artifact ;

(3) I _ArmLeft (x, y) < V′ _SmoothLeft +Th _Artifact /3 and

Among them, x and y represent the pixel row coordinates and column coordinates, respectively, and their value ranges satisfy the coordinate range of the artifact prior search area between the left arm and torso; V′ _SmoothLeft and

is the mean value of the smooth areas on the left and right sides of the left arm and torso artifact prior search area. The row coordinate range of the left and right smooth areas is the same as the row search area of the left arm torso artifact prior search area, and the column coordinate range of the left and right smooth areas is the same. Respectively [L _ColCtr -L _Width -L _Smooth , L _ColCtr -L _Width -1] and [L _ColCtr +L _Width +1, L _ColCtr +L _Width +L _Smooth ];

The third method of multi-threshold decision Specifically, if the inter-pixel right arm torso artifact prior search area pixel value _I ArmRight (x, y) satisfies the following three conditions, the pixel value _I ArmRight (x, y) The corresponding pixel is the artifact pixel between the arm and torso:

(1) I _ArmRight (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(2) I _ArmRight (x, y) < Th _Artifact ;

(3) The pixel value I _ArmRight (x, y) satisfies I _ArmRight (x, y) < VS _SmoothLeft +Th _Artifact /3 and

Further, the coordinate set of the artifact pixels is M, where M={(x _Artifact (i), y _Artifact (i))|i=1, 2, 3, ..., m}, where x _Artifact (i), y _Artifact (i) represents the row and column coordinates of the i-th artifact pixel, respectively, and m represents the total number of artifact pixels;

The value range of the column coordinates of the pixel neighborhood sequence of the row where the artifact pixel is located is

[y _Artifact (i)-L _Width -L _Smooth , y _Artifact (i)+L _Width +L _Smooth ],

By judging the artifact pixels between the legs and the artifact pixels between the arms and torso, the set M of artifact pixel coordinates is obtained, and the expression is M={(x _Artifact (i), y _Artifact (i))|i=1, 2,3, ..., m}, _{where, x Artifact (i), y} Artifact (i) represent the coordinates of the i-th row and dummy column coordinates of the image pixel, m represents the total number of dummy video pixel. For each artifact pixel (x _Artifact (i), y _Artifact (i)), take out the pixel neighborhood sequence of its row, the pixel neighborhood sequence, the pixel neighborhood sequence includes the artifact area and the smooth area, let the ith The pixel neighborhood sequence of the artifact pixels is

The number of pixels included in each pixel neighborhood sequence is 2(L _Width +L _Smooth )+1.

Further, the specific formula of the artifact suppression function is:

Among them, F(t) is the artifact suppression function, t∈[-L _Width -L _Smooth , +L _Width +L _Smooth ], t is an integer, A is the amplitude, σ is the standard deviation of the Gaussian distribution,

represents the background mean value of the ith artifact pixel;

The value of the specified range A is

That is, the difference between the maximum value in the neighborhood of the artifact pixel and the average value of the background, the average value of the image background

That is, the pixel mean of the smooth area on both sides of the ith artifact pixel;

The pixel neighborhood sequence of the i-th artifact pixel after artifact suppression

for:

in,

Express

Divide the pixel corresponding to F(t), and control the value of the artifact pixel within the range of the artifact pixel suppression operation.

about;

finally get the sequence

replace the original sequence

of artifact-removed security images. Compared with the prior art, the present invention has the following significant advantages:

(1) According to the fixed posture of the human body during security inspection, the area search method is used to search for the artifact between the legs and the artifact between the arms and the torso in the fixed area, which is conducive to saving computer resources and search time, and improving the efficiency of artifact search;

(2) Using multiple threshold judgment methods to judge whether the pixels in the search area are artifacts, which is conducive to accurately judging the position of artifacts, and can reduce misjudgments at the same time;

(3) The pixel value suppression method is used to suppress the artifact pixel value, so that the artifact pixel value and its neighboring pixel value have a natural transition, so as to avoid image quality degradation due to excessive value change.

Description of drawings

Figure 1 is a flowchart of a method for suppressing artifacts in human security inspection images.

Figure 2. Original imaging results of human security inspection.

Figure 3. Figure 3. Image of human security inspection image marking artifact area.

Figure 4 is a schematic diagram of calculating the row coordinates of the center of the head.

Fig. 5 Map of the prior search area for marking the artifact between the legs in the security inspection image.

Fig. 6 Cumulative sum curve graph of leg region image after flipping translation difference method.

Figure 7. Security image marking the centerline of the leg area column.

Figure 8 Screening images of artifact pixels marked between the legs.

Figure 9. Screening images of labeled prior search regions for artifact between arms and torso.

Figure 10 Screening images of artifact pixels labeled between arms and torso.

Figure 11 Security image with artifact suppression.

detailed description

Below in conjunction with the accompanying drawings, a specific implementation of a method for removing artifact from a security inspection image using area search and pixel value suppression of the present invention will be described in detail, as shown in FIG. 1 , which specifically includes the following steps:

1. Calculation of the row coordinates of the center of the human head

Figure 2 is the original security inspection image, and the white dotted box in Figure 3 marks the artifact area, including the upper arm torso artifact and the lower leg artifact. Due to the difference in the height of the human body, and the size ratio of the human security inspection image is the same as that of the human body, the row coordinate information of the human head and torso in the security inspection image is also different. In the cooperative security inspection process, the posture of the inspected person is required to be separated from the arms and torso, and the feet are separated from each other, and the security inspection images of different personnel have the same posture pattern. In order to determine the position of the artifact, the artifact search area must be determined by calculating the position of the head in the image and the fixed proportion of each part of the human body.

In order to reduce the amount of calculation and increase the detection accuracy, determine the a priori search area of the human head, and it is required to cover the head information of different heights. It is assumed that the starting row coordinate of the head a priori search area is H _RowUp , and the ending row coordinate is H _RowDown , the starting column coordinate is H _ColLeft , the ending column coordinate is H _ColRight , the pixels in each row in the area are accumulated and summed to obtain a row cumulative projection vector, and the row coordinate _{greater than the threshold Th head is found in the projection vector, and} Average these row coordinates to obtain the row coordinate H _{RowCtr of the} center of the head. Using this coordinate information, the height of the inspected person can be calculated, and the prior artifact between the arms and the torso and the artifact between the legs can be calculated according to the size ratio of the body. Search area. As shown in FIG. 4 , the white dotted box in FIG. 4 is the prior search area, and the curve on the right side of the prior search area is the row cumulative projection obtained after the cumulative summation of each row of pixels in the area.

In this embodiment, the rows and columns are calculated based on the coordinate system of the security inspection image. The coordinate system of the security inspection image is the starting point of the upper left corner of the security inspection image as the coordinate center, and the right direction of the coordinate center is the positive direction of the column coordinate axis. The coordinate system formed by the downward direction of the coordinate center is the positive direction of the row coordinate axis.

2. Artifact pixel judgment between legs

The shape of the artifact between the legs is a vertical straight line, and the length and intensity of the artifact are related to the body height and standing posture. Theoretically, the row coordinate interval of the artifact is all the rows in the area where the leg is located, and the column coordinate of the artifact is on the symmetry axis of the two legs. However, in actual situations, the artifacts between the legs will gradually weaken with the increase of the distance between the legs and finally disappear, and due to the non-standard standing posture, the inspected person may not be in the center of the imaging scene. Therefore, when determining the artifact prior search area, the row coordinate interval is still calculated according to the row interval where the legs are located, and the column coordinates of the center lines of the two legs are calculated by the flip translation difference method. After the prior search area is obtained, the specific coordinates of the artifact pixels between the legs are obtained by using the first multiple threshold decision method.

a. Calculation of the artifact pixel row interval between the legs

Assuming that the number of rows and columns of the human security inspection image are N _Row and N _Col , and the ratio of the leg length to the height in the security inspection image is ρ, the starting row coordinates of the artifact between the legs are calculated according to the row coordinates of the center of the head as L _RowBgn =round(N _Row -(N _Row -H _RowCtr )·ρ), the end row coordinate is N _Row , where round represents rounding.

b. Calculation of the artifact pixel column interval between the legs

Since the personnel need to stand for 3 to 5 seconds during the security check, the obtained security check image is approximately symmetrical, and the artifact between the legs is on the symmetry axis, but the symmetry axis is deviated from the center line of the image column. Therefore, the security check image leg is obtained. The position of the center line of the leg is completed to complete the calculation of the coordinates of the artifact column between the legs.

The invention uses the flip translation difference method to obtain the leg centerline column coordinates of the security inspection image. Let the leg region image be I _Leg (as shown in the dotted box in Figure 5), and the cyclic shift number is n, then the calculated flip translation difference function value is:

f(n)=sum(abs(circshift(I _Leg ,n)-fliplr(circshift(I _Leg ,n)))) (1)

Among them, circshift represents the horizontal cyclic shift operation, fliplr represents the left and right flip operation, sum represents the cumulative summation, abs represents the absolute value operation, and the value range of n is _{an integer in [-N Shift} , N _Shift ] (wherein, N _{Shift )} represents the preset first offset), then f(n) is _{a one-dimensional vector containing 2N Shift} + 1 values, and the shift number n corresponding to the minimum value in f(n) can represent the center line of the leg of the security inspection image Deviation from the centerline of the security image column. The curve of the absolute value sum of the image differences obtained in the shift interval is shown in Fig. 6 .

Calculate the column coordinates of the leg centerline column coordinates of the security inspection image as:

As shown in Figure 7, the white line marks the center line of the leg of the security inspection image; among them, round means rounding, and arcmin means the n value that makes f(·) take the minimum value. Since there is one column or multiple columns of _{artifact pixels, extend L Width} columns of pixels in the left and right neighborhoods of _{L ColCtr} as the search area for artifact column coordinates (that is, L _Width is the column coordinate of the leg center line of the security inspection image, L _ColCtr The number of pixel columns extending to the left and right neighborhoods). So far, the row coordinate range of the prior artifact search area between the legs is: [L _BowBgn , N _Row ], and the column coordinate range is: [L _ColCtr -L _Width , L _ColCtr +L _Width ].

c. Artifact pixel decision

After determining the a priori search area, a threshold is used to determine whether the pixels in the area are artifact pixels, and the artifact pixel threshold is set as Th _Artifact . In the column coordinate range [L _ColCtr -L _Width , L _ColCtr +L _{Width ] of the determined artifact area, then extend L Smooth} columns of pixels to the left and right sides respectively _{(where L Smooth} is the preset second offset) _{, and assuming that each L Smooth} column pixel on both sides of the artifact area is a smooth area, the mean values of the smooth areas on the left and right sides are calculated as V _SmoothLeft and V _{SmoothRight , respectively} . It is assumed that the pixel value of the artifact prior search area between the legs is represented as I _Leg (x, y), where x and y represent the pixel row and column coordinates, respectively, and the value range satisfies the coordinate range of the artifact prior search area. . Traverse each pixel in the prior search area, and if I _Leg (x, y) satisfies the following conditions simultaneously, it is considered to be an artifact pixel:

(a) The pixel value I _Leg (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(b) The pixel value is greater than the artifact threshold I _Leg (x, y) < Th _Artifact ;

(c) The pixel value I _Leg (x, y) satisfies I _Leg (x, y) < V _SmoothLeft + Th _Artifact /3 and I _Leg (x, y) < V _SmoothRight + Th _Artifact /3.

The coordinate information of the artifact pixels is recorded, as shown in FIG. 8 . The white dots in FIG. 8 mark the artifact pixels between the legs.

3. Artifact pixel judgment between arms and torso

The shape of the artifact between the arms and torso is a vertical straight line or a curve, and the length and intensity of the artifact are related to the placement of the human arm. First, determine the row interval information and column interval information of the prior search area according to the head center row coordinates and body proportion information, and then use the second and third multiple threshold judgment methods to judge each pixel in the search area to obtain a pseudo Shadow pixel coordinates.

a. A priori search interval determination

According to the fixed relationship between the position of the human arm and the head, it is assumed that the distance _{between the starting and ending row coordinates of the artifact between the arm torso and the center row coordinate H RowCtr of the head is S RowUp} and S _{RowDown respectively.} The row coordinate interval of the verification search area is [H _RowCtr +S _RowUp , H _RowCtr +S _RowDown ]. It is stipulated that the left arm in the security inspection image is the right arm of the human body, and the right arm in the image is the left arm of the human body. According to the fixed ratio information of the position of the human arm and the size of the security inspection image, the column coordinate intervals of the artifact prior search area between the left and right arms and torso of the human body are determined as follows:

and

According to the range of the search interval, the prior search interval images of the left and right arms of the human body are obtained as I _ArmLeft and I _{ArmRight respectively} , and the area shown by the dotted box in Figure 9 is the prior search area for the artifact between the arms and torso.

b. Artifact pixel decision

After the search area is determined, the second multiple threshold method is used to determine whether the pixels in the area are artifact pixels. Similar to the judgment method of the artifact between the legs, assuming that the column interval of the _{artifact pixel I ArmLeft (x, y) between the left arm and torso is [yL Width} , y+L _Width ], and then expand the L _Smooth column on the left and right sides. Pixels are used to represent smooth areas, and the mean values of pixels in the smooth areas on the left and right sides are V′ _SmoothLeft and

Similarly, the average pixel values of the smooth areas on the left and right sides of the _{artifact pixel I ArmRight} (x, y) between the right arm and torso _{are V″ SmoothLeft} and

Traverse each pixel in the prior search area, if the pixel value I _ArmLeft (x, y) of the pixel in the left arm prior search area satisfies the following conditions at the same time, it is considered to be an artifact pixel:

(a) The pixel value I _ArmLeft (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(b) The pixel value is greater than the artifact threshold I _ArmLeft (x, y) < Th _Artifact ;

(c) The pixel value I _ArmLeft (x, y) satisfies I _ArmLeft (x, y) < V′ _SmoothLeft +Th _Artifact /3 and

Similarly, according to the third multiple threshold method, if the pixel I _ArmRight (x, y) in the right arm prior search area satisfies the following conditions, it is considered as an artifact pixel:

(a) The pixel value I _ArmRight (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

(b) The pixel value is greater than the artifact threshold I _ArmRight (x, y) < Th _Artifact ;

(c) The pixel value I _ArmRight (x, y) satisfies I _ArmRight (x, y) < V″ _SmoothLeft +Th _Artifact /3 and

Record the coordinate information of the left and right arm artifact pixels. The pixels marked with white dots in Figure 10 are the artifact pixels between the arms and torso.

4. Artifact pixel value suppression

By judging the artifact pixels between the legs and the artifact pixels between the arms and torso, the set M of artifact pixel coordinates is obtained, and the expression is M={(x _Artifact (i), y _Artifact (i))|i=1, 2,3, ..., m}, _{where, x Artifact (i), y} Artifact (i) represent the coordinates of the i-th row and dummy column coordinates of the image pixel, m represents the total number of dummy video pixel. For each artifact pixel (x _Artifact (i), y _Artifact (i)), take out the pixel neighborhood sequence of its row, and the value range of the column coordinates of the pixel neighborhood sequence is [y _Artifact (i)-L _Width -L _Smooth , y _Artifact (i)+L _Width + L _Smooth ], the pixel neighborhood sequence includes the artifact area and the smooth area, and the pixel neighborhood sequence of the i-th artifact is

The number of pixels included in each pixel neighborhood sequence is 2(L _Wrdth +L _Smooth )+1.

Construct the artifact suppression function F(t) whose expression is:

Among them, t∈[-L _Width -L _Smooth , +L _Width +L _Smooth ], t is an integer, A is the amplitude, σ is the standard deviation of the Gaussian distribution,

represents the background mean of the ith artifact pixel. The value of the specified range A is

That is, the difference between the maximum value in the neighborhood of the artifact pixel and the average value of the background. Image background average

That is, the pixel mean of the smooth region on both sides of the ith artifact pixel. Therefore, the maximum value of the obtained suppression function is the maximum value of the artifact pixel area, and the minimum value is the value of the image background, and the parameters of the suppression function need to be recalculated for each artifact pixel.

Suppressed pixel neighborhood sequence of artifact pixels

for:

in,

Express

Divide the pixel corresponding to F(t), and control the value of the artifact pixel within the range of the artifact pixel suppression operation.

left and right, and then sequence

replace the original sequence

The security inspection image after artifact suppression is obtained.

The security inspection image after artifact suppression is shown in Figure 11.

Compared with the prior art, the present invention has the following significant advantages:

(1) According to the fixed posture of the human body during security inspection, the area search method is used to search for the artifact between the legs and the artifact between the arms and the torso in the fixed area, which is conducive to saving computer resources and search time, and improving the efficiency of artifact search;

(2) Using multiple threshold judgment methods to judge whether the pixels in the search area are artifacts, which is conducive to accurately judging the position of artifacts, and can reduce misjudgments at the same time;

(3) The pixel value suppression method is used to suppress the artifact pixel value, so that the artifact pixel value and its neighboring pixel value have a natural transition, so as to avoid image quality degradation due to excessive value change.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection of the present invention. within the range.

Claims (6)

1. A security inspection image artifact removal method utilizing area search and pixel value suppression, characterized in that it comprises the following steps:

   Calculate the row coordinates of the center of the human head:

   Accumulate and sum the pixel values of all row pixels in the set head a priori search area, take the row coordinate of the cumulative sum greater than the threshold as the head row coordinate, and take the average of the head row coordinates as the center of the human head row coordinates;

   Determine the artifact pixel between the legs:

   Determine the artifact prior search area between the legs, and determine whether the pixels in the artifact prior search area between the legs are artifact pixels between the legs through the first multiple threshold method, so as to obtain the artifact pixel coordinates between the legs; The row search area of the inter-leg artifact prior search area is calculated according to the row coordinates of the head center and the body scale, and the column search area of the inter-leg artifact prior search area is obtained by dividing the center line of the leg area image. The column coordinates are obtained by extending a fixed column of pixels to the left and right, and the column coordinates of the leg center line of the security inspection image are obtained by the flip translation difference method;

   Determining artifact pixels between arms and torso:

   Determine the prior search area of the artifact between the left and right arm torso according to the head center row coordinate and body proportion, and use the second multiple threshold method and the third multiple threshold method to determine the prior search area of the left and right arm torso artifacts respectively. Whether the pixel of is the artifact pixel between the arm and torso, get the coordinates of the artifact pixel between the arm and torso;

   Suppress the pixel value of the artifact pixel:

   Take out each artifact pixel and the pixel neighborhood sequence of its row, construct an artifact suppression function for artifact suppression, and then replace the original pixel sequence with the suppressed pixel sequence to obtain an artifact-removed security inspection image; Shadow pixels include artifact pixels between legs and artifact pixels between arms and torso.
2. The method for removing artifact from a security inspection image using area search and pixel value suppression according to claim 1, characterized in that:

   The flip translation difference method is specifically as follows: the leg region images are cyclically shifted and left and right flipped, and then the difference is made, and the image difference after the difference is accumulated and summed, so as to obtain the image difference accumulated sum vector of different shift numbers, and by The number of shifts corresponding to the minimum value in the vector calculates the column coordinates of the centerline of the leg area image.
3. The method for removing artifact of a security inspection image using area search and pixel value suppression according to claim 2, wherein the row search area of the artifact a priori search area between the legs is [L _RowBgn , N _Row ], Among them, N _Row is the line number of the security inspection image, L _RowBgn is the starting line coordinate of the artifact between the legs,

   L _RowBgn = round(N _Row -(N _Row -H _RowCtr )·ρ), round means rounding, H _{RowCtr is the} center row coordinate of the human head, and ρ is the ratio of leg length to height in the security image;

   The column search area of the artifact a priori search area between the legs is [L _ColCtr -L _Width , L _ColCtr +L _Width ], where L _Width is the column coordinate of the leg center line of the security inspection image L _ColCtr to the left and right sides of the neighborhood The number of extended pixel columns;

   The calculation formula of the _{coordinate L ColCtr} of the symmetry axis of the leg image is:

   

   Among them, round means rounding, arcmin means the n value that makes f( ) get the minimum value;

   f(n) is the value of the flip translation difference function, and its specific calculation formula is:

   f(n)=sum(abs(circshift(I _Leg ,n)-fliplr(circshift(I _Leg ,n))))

   Among them, circshift represents the horizontal cyclic shift operation, fliplr represents the left and right flip operation, sum represents the cumulative summation, abs represents the absolute value operation, and the value range of n is _{an integer in [-N Shift} , N _Shift ], then f( n) is _{a one-dimensional vector containing 2N Shift} + 1 values, the shift number n corresponding to the minimum value in f(n) can represent the deviation of the symmetry axis of the leg area image from the center line of the security image column, and N _Shift represents the pre- Set the first offset;

   The described first multiple threshold judgment method is specifically, if the pixel value I _Leg (x, y) of the pixel in the two-leg artifact a priori search area satisfies the following three conditions simultaneously, then the pixel value I _Leg (x, y) ) corresponds to the pixel of the artifact between the legs:

   (1) The pixel value I _Leg (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

   (2) I _Leg (x, y) > Th _Artifact ;

   (3) I _Leg (x, y) > V _SmoothLeft +Th _Artifact /3 and I _Leg (x, y) > V _SmoothRight +Th _Artifact /3;

   Among them, x and y represent the pixel row coordinates and column coordinates, respectively, and their value range satisfies the coordinate range of the artifact prior search area between the legs; Th _Artifact is the set artifact pixel threshold; V _SmoothLeft and V _SmoothRight are double The mean value of the smooth areas on the left and right sides of the inter-leg artifact prior search area, the row coordinate range of the left and right smooth areas is the same as the row search area of the double-leg artifact prior search area, and the column coordinate ranges of the left and right smooth areas are respectively

   [L _ColCtr -L _Width -L _Smooth , L _ColCtr -L _Width -1] and

   [L _ColCtr +L _Width +1, L _ColCtr +L _Width +L _Smooth ], L _Smooth is the preset second offset.
4. The method for removing artifact of a security inspection image using area search and pixel value suppression according to claim 3, wherein,

   The row coordinate interval of the artifact prior search area between the arms and torso is:

   [H _RowCtr +S _RowUp , H _RowCtr +S _RowDown ], the column coordinate interval of the artifact prior search area between the left and right arms and torso is respectively

   

   and

   

   where S _RowUp is the distance between the starting row coordinate of the artifact between the arm torso and the head center row coordinate H _RowCtr , and S _RowDown is the distance between the ending row coordinate of the arm torso artifact and the head center row coordinate H _RowCtr ;

   The second multi-threshold decision method in particular, if the pixels between the left arm torso artifact prior search area pixel value _I ArmLeft (x, y) satisfies the following three conditions, the pixel value _I ArmLeft (x, y) The corresponding pixel is the artifact pixel between the arm and torso:

   (1) The pixel value I _ArmLeft (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is [yL _Width , y+L _Width ];

   (2) I _ArmLeft (x, y)>Th _Artifact ;

   (3) I _ArmLeft (x, y) > V′ _SmoothLeft +Th _Artifact /3 and

   I _ArmLeft (x, y)>V′ _{SmoothRig t} +Th _Artifact /3;

   Among them, x and y represent the pixel row coordinates and column coordinates, respectively, and their value ranges meet the coordinate range of the artifact prior search area between the left arm and torso; V′ _SmoothLeft and V′ _{SmoothRig t} are the artifact between the left arm and torso. The average value of the smooth areas on the left and right sides of the experimental search area. The row coordinate range of the left and right smooth areas is the same as the row search area of the left arm and torso artifact prior search area. The column coordinate ranges of the left and right smooth areas are [L _ColCtr -L _Width -L _Smooth , L _ColCtr -L _Width -1] and [L _ColCtr +L _Width +1, L _ColCtr +L _Width +L _Smooth ];

   The third method of multi-threshold decision Specifically, if the inter-pixel right arm torso artifact prior search area pixel value _I ArmRight (x, y) satisfies the following three conditions, the pixel value _I ArmRight (x, y) The corresponding pixel is the artifact pixel between the arm and torso:

   (1) I _ArmRight (x, y) is the maximum value in the neighborhood of the row where it is located, and the coordinate range of the neighborhood column is

   _{[yL Width, y + L Wiidth} ];

   (2) I _ArmRight (x, y) > Th _Artifact ;

   (3) The pixel value I _ArmRight (x, y) satisfies I _ArmRight (x, y) > V″ _SmoothLeft +Th _Artifact /3 and

   I _ArmRight (x, y)>V″ _{SmoothRig t} +Th _Artifact /3.
5. The method for removing artifact of a security inspection image using area search and pixel value suppression according to claim 4, wherein,

   The coordinate set of the artifact pixels is M, where M={(x _Artifact (i), y _Artifact (i))|i=1, 2, 3, ..., m}, where x _Artifact (i) , y _Artifact (i) represents the row and column coordinates of the ith artifact pixel, respectively, and m represents the total number of artifact pixels;

   The value range of the column coordinates of the pixel neighborhood sequence of the row where the artifact pixel is located is

   [y _Artifact (i)-L _Width -L _Smooth , y _Artifact (i)+L _Width +L _Smooth ],

   By judging the artifact pixels between the legs and the artifact pixels between the arms and torso, the set M of artifact pixel coordinates is obtained, and the expression is M={(x _Artifact (i), y _Artufact (i))|i=1, 2,3, ..., m}, _{where, x Artifact (i), y} Artifact (i) represent the coordinates of the i-th row and dummy column coordinates of the image pixel, m represents the total number of dummy video pixel. For each artifact pixel (x _Artifact (i), y _Artifact (i)), take out the pixel neighborhood sequence of its row, the pixel neighborhood sequence, the pixel neighborhood sequence includes the artifact area and the smooth area, let the ith The pixel neighborhood sequence of the artifact pixels is

   

   The number of pixels included in each pixel neighborhood sequence is 2(L _Width +L _Smooth )+1.
6. The method for removing artifact of a security inspection image using area search and pixel value suppression according to claim 4, wherein,

   The specific formula of the artifact suppression function is:

   

   Among them, F(t) is the artifact suppression function, t∈[-L _Width -L _Smooth , +L _Width +L _Smooth ], t is an integer, A is the amplitude, σ is the standard deviation of the Gaussian distribution,

   

   represents the background mean value of the ith artifact pixel;

   The value of the specified range A is

   

   That is, the difference between the maximum value in the neighborhood of the artifact pixel and the average value of the background, the average value of the image background

   

   That is, the pixel mean of the smooth area on both sides of the ith artifact pixel;

   The pixel neighborhood sequence of the i-th artifact pixel after artifact suppression

   

   for:

   

   in,

   

   Express

   

   Divide the pixel corresponding to F(t), and control the value of the artifact pixel within the range of the artifact pixel suppression operation.

   

   about;

   finally get the sequence

   

   replace the original sequence

   

   of artifact-removed security images.

Images