A method for automatic passenger face recognition

Abstract

The present invention discloses a passenger face automatic recognition method, which relates to the field of face recognition technology, and comprises the following steps: S1, photographing a verification face image of a passenger; S2, using the overlap coefficient of all pixel points to correct the contour area of ​​the verification face image, obtaining a sensitive contour area, and generating a sensitive comparison coefficient for the sensitive contour area; S3, using the sensitive comparison coefficient to compare the verification face image with a face database, and obtaining a recognition result. The sensitive comparison coefficient and Bhattacharyya distance used in the present invention can reflect the similarity between the sensitive contour area and each standard image in the face database, and can more accurately judge whether the verification face image is consistent with the face in the face database.

Description

Technical Field

[0001] The present invention relates to the technical field of face recognition, and in particular to a method for automatically recognizing a passenger's face. Background Art

[0002] With the development of intelligent transportation systems, the demand for automatic passenger identification in public transportation is growing. Traditional identification methods rely on manual ticket inspection or magnetic card tickets, which have problems such as low efficiency and prone to errors. In recent years, face recognition technology has been widely used in public transportation systems due to its high efficiency and accuracy. However, existing face recognition methods still have shortcomings in processing large-scale passenger data and improving recognition speed and accuracy. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a passenger face automatic recognition method.

[0004] The technical solution of the present invention is: a method for automatic passenger face recognition comprises the following steps:

[0005] S1. Take a verification face image of the passenger;

[0006] S2, using the overlap coefficient of all pixels to correct the contour area of ​​the verification face image, obtain a sensitive contour area, and generate a sensitive comparison coefficient for the sensitive contour area;

[0007] S3. Use the sensitive comparison coefficient to compare the verification face image with the face database to obtain the recognition result.

[0008] Furthermore, S2 includes the following sub-steps:

[0009] S21, calculating the overlap coefficient of each pixel in the verification face image;

[0010] S22, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and correcting the contour area of ​​the verification face image by using the overlap coefficient of each pixel point to obtain a sensitive contour area;

[0011] S23. Generate a sensitive comparison coefficient for the sensitive contour area.

[0012] The beneficial effect of the above further scheme is: in the present invention, the main principle of the grayscale integral projection method is to perform grayscale integral projection on the grayscale image of the face in the horizontal and vertical directions respectively, and roughly determine the position of the face area or facial organs (such as eyes) according to the concave and convex features of the integral projection image. Since the grayscale integral projection method is based on the accumulation operation of grayscale values, its accuracy is affected by multiple factors such as image resolution and lighting conditions. Under complex backgrounds or when the face posture changes greatly, the positioning accuracy of the grayscale integral projection method may be limited. Therefore, the present invention also needs to use the overlap coefficient to correct the contour area. The calculation of the overlap coefficient takes into account the relative position and grayscale value between pixels, so that the algorithm has a certain robustness to interference factors such as lighting changes and noise.

[0013] Further, in S21, the face image is verified. Overlap coefficient of pixels The calculation formula is:

[0014] ;

[0015] In the formula, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, Represents the remainder function.

[0016] Further, S22 includes the following sub-steps:

[0017] S221, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and generating a vertical grayscale integral vector and a horizontal grayscale integral vector;

[0018] S222, generating a distribution feature range using the Euclidean norm of the vertical grayscale integral vector and the Euclidean norm of the horizontal grayscale integral vector;

[0019] S223, eliminating the pixel points whose overlap coefficients in the contour area do not belong to the distribution feature range, and obtaining the sensitive contour area.

[0020] The beneficial effect of the above further scheme is that in the present invention, by calculating the Euclidean norm of the vertical grayscale integral vector and the horizontal grayscale integral vector, the contour feature can be quantified into a specific value. Generating a distribution feature range helps to determine which pixel points have a reasonable overlap coefficient and which may be an abnormal value caused by noise or interference factors, which provides a basis for the subsequent elimination of pixels that do not meet the range. By eliminating pixels that do not belong to the distribution feature range, the noise and interference factors in the contour area can be further reduced, and the accuracy of the contour can be improved.

[0021] Grayscale integral projection method reveals the grayscale distribution characteristics of an image in a certain direction (usually horizontal or vertical direction) by integrally projecting the grayscale values ​​of the image in that direction. Therefore, the vertical grayscale integral vector is a one-dimensional array obtained by summing the grayscale values ​​of each column of the image; the horizontal grayscale integral vector is a one-dimensional array obtained by summing the grayscale values ​​of each row of the image.

[0022] Furthermore, in S222, the upper limit of the distribution feature range The calculation formula is:

[0023] ;

[0024] In the formula, represents the Euclidean norm of the horizontal grayscale integral vector, Represents the Euclidean norm of the vertical grayscale integral vector;

[0025] In S222, the lower limit of the distribution characteristic range The calculation formula is:

[0026] .

[0027] Further, S23 includes the following sub-steps:

[0028] S231, extracting the convex hull of the sensitive contour area;

[0029] S232, extracting the distance between the convex hull and the sensitive contour area;

[0030] S233. Calculate a sensitive comparison coefficient according to the distance between the convex hull and the sensitive contour area.

[0031] The beneficial effect of the above further scheme is: in the present invention, the convex hull refers to the minimum circumscribed convex polygon of the contour, and the convex hull can reflect the contour shape of the object in the image and the convexity and concavity of the boundary. The distance between the convex hull and the original contour can be obtained by calculating the distance from each contour point to its nearest convex hull corner point and then summing them up. This distance reflects the smoothness of the contour and the degree of approximation of the convex hull. By calculating the distance between the convex hull and the sensitive contour area, the shape difference between the two can be quantified. This provides a reliable basis for subsequent comparison and recognition.

[0032] Furthermore, in S233, the sensitivity comparison coefficient The calculation formula is:

[0033] ;

[0034] In the formula, Represents the distance between the convex hull and the sensitive contour area, represents the perimeter of the sensitive contour area, Represents the base of natural logarithms.

[0035] Furthermore, in S3, the Bhattacharyya distance between the verification face image and each standard image in the face database is calculated, and the Bhattacharyya distance and the sensitive comparison coefficient are averaged to obtain the final similarity of each standard image. If there is a standard image whose final similarity is greater than or equal to the threshold, the passenger's recognition result is passed, otherwise the passenger's recognition result is failed.

[0036] The beneficial effects of the present invention are as follows: the present invention uses the overlap coefficient of all pixels to correct the contour area of ​​the verification face image, so as to obtain a more accurate sensitive contour area. By correcting the contour area, the feature distortion caused by factors such as image noise and illumination changes can be reduced, thereby improving the accuracy of verification; the sensitive comparison coefficient and the Bhattacharyya distance can reflect the similarity between the sensitive contour area and each standard image in the face database, and can more accurately determine whether the verification face image is consistent with the face in the face database. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the automatic passenger face recognition method. DETAILED DESCRIPTION

[0038] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0039] like Figure 1 As shown, the present invention provides a method for automatic passenger face recognition, comprising the following steps:

[0040] S1. Take a verification face image of the passenger;

[0041] S2, using the overlap coefficient of all pixels to correct the contour area of ​​the verification face image, obtain a sensitive contour area, and generate a sensitive comparison coefficient for the sensitive contour area;

[0042] S3. Use the sensitive comparison coefficient to compare the verification face image with the face database to obtain the recognition result.

[0043] In this embodiment of the present invention, S2 includes the following sub-steps:

[0044] S21, calculating the overlap coefficient of each pixel in the verification face image;

[0045] S22, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and correcting the contour area of ​​the verification face image by using the overlap coefficient of each pixel point to obtain a sensitive contour area;

[0046] S23. Generate a sensitive comparison coefficient for the sensitive contour area.

[0047] In the present invention, the main principle of the grayscale integral projection method is to perform grayscale integral projection on the grayscale image of the face in the horizontal and vertical directions respectively, and roughly determine the position of the face area or facial organs (such as eyes) according to the concave and convex features of the integral projection image. Since the grayscale integral projection method is based on the accumulation operation of grayscale values, its accuracy is affected by multiple factors such as image resolution and lighting conditions. Under complex backgrounds or when the face posture changes greatly, the positioning accuracy of the grayscale integral projection method may be limited. Therefore, the present invention also needs to use the overlap coefficient to correct the contour area. The calculation of the overlap coefficient takes into account the relative position and grayscale value between pixel points, so that the algorithm has a certain robustness to interference factors such as lighting changes and noise.

[0048] In the embodiment of the present invention, in S21, the first Overlap coefficient of pixels The calculation formula is:

[0049] ;

[0050] In the formula, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, Represents the remainder function.

[0051] In this embodiment of the present invention, S22 includes the following sub-steps:

[0052] S221, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and generating a vertical grayscale integral vector and a horizontal grayscale integral vector;

[0053] S222, generating a distribution feature range using the Euclidean norm of the vertical grayscale integral vector and the Euclidean norm of the horizontal grayscale integral vector;

[0054] S223, eliminating the pixel points whose overlap coefficients in the contour area do not belong to the distribution feature range, and obtaining the sensitive contour area.

[0055] In the present invention, by calculating the Euclidean norm of the vertical grayscale integral vector and the horizontal grayscale integral vector, the contour feature can be quantified into a specific value. Generating a distribution feature range helps to determine which pixel overlap coefficients are reasonable and which may be abnormal values ​​caused by noise or interference factors, providing a basis for the subsequent elimination of pixels that do not meet the range. By eliminating pixels that do not belong to the distribution feature range, the noise and interference factors in the contour area can be further reduced, and the accuracy of the contour can be improved.

[0056] Grayscale integral projection method reveals the grayscale distribution characteristics of an image in a certain direction (usually horizontal or vertical direction) by integrally projecting the grayscale values ​​of the image in that direction. Therefore, the vertical grayscale integral vector is a one-dimensional array obtained by summing the grayscale values ​​of each column of the image; the horizontal grayscale integral vector is a one-dimensional array obtained by summing the grayscale values ​​of each row of the image.

[0057] In the embodiment of the present invention, in S222, the upper limit of the distribution feature range The calculation formula is:

[0058] ;

[0059] In the formula, represents the Euclidean norm of the horizontal grayscale integral vector, Represents the Euclidean norm of the vertical grayscale integral vector;

[0060] In S222, the lower limit of the distribution characteristic range The calculation formula is:

[0061] .

[0062] In this embodiment of the present invention, S23 includes the following sub-steps:

[0063] S231, extracting the convex hull of the sensitive contour area;

[0064] S232, extracting the distance between the convex hull and the sensitive contour area;

[0065] S233. Calculate a sensitive comparison coefficient according to the distance between the convex hull and the sensitive contour area.

[0066] In the present invention, the convex hull refers to the smallest circumscribed convex polygon of the contour, and the convex hull can reflect the contour shape of the object in the image and the convexity and concavity of the boundary. The distance between the convex hull and the original contour can be obtained by calculating the distance from each contour point to its nearest convex hull corner point and then summing them up. This distance reflects the smoothness of the contour and the degree of approximation of the convex hull. By calculating the distance between the convex hull and the sensitive contour area, the shape difference between the two can be quantified. This provides a reliable basis for subsequent comparison and recognition.

[0067] In the embodiment of the present invention, in S233, the sensitivity comparison coefficient The calculation formula is:

[0068] ;

[0069] In the formula, Represents the distance between the convex hull and the sensitive contour area, represents the perimeter of the sensitive contour area, Represents the base of natural logarithms.

[0070] In the embodiment of the present invention, in S3, the Bhattacharyya distance between the verification face image and each standard image in the face database is calculated, and the Bhattacharyya distance and the sensitive comparison coefficient are averaged to obtain the final similarity of each standard image. If there is a standard image whose final similarity is greater than or equal to the threshold, the passenger's recognition result is passed, otherwise the passenger's recognition result is failed.

[0071] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A passenger face automatic recognition method, characterized in that: The following steps are involved: S1. Take a verification face image of the passenger; S2, using the overlap coefficient of all pixels to correct the contour area of ​​the verification face image, obtain a sensitive contour area, and generate a sensitive comparison coefficient for the sensitive contour area; S3, using the sensitive comparison coefficient to compare the verification face image with the face database to obtain the recognition result; The S2 comprises the following sub-steps: S21, calculating the overlap coefficient of each pixel in the verification face image; S22, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and correcting the contour area of ​​the verification face image by using the overlap coefficient of each pixel point to obtain a sensitive contour area; S23, generating a sensitive comparison coefficient for the sensitive contour area; In the step S21, the face image is verified. Overlap coefficient of pixels The calculation formula is: ； In the formula, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, Indicates the first The gray value of a pixel, represents the remainder function; The S22 comprises the following sub-steps: S221, extracting the contour area of ​​the verification face image by using the grayscale integral projection method, and generating a vertical grayscale integral vector and a horizontal grayscale integral vector; S222, generating a distribution feature range using the Euclidean norm of the vertical grayscale integral vector and the Euclidean norm of the horizontal grayscale integral vector; S223, removing pixels in the contour area whose overlap coefficients do not fall within the distribution feature range, to obtain a sensitive contour area; In S222, the upper limit of the distribution characteristic range The calculation formula is: ； In the formula, represents the Euclidean norm of the horizontal grayscale integral vector, Represents the Euclidean norm of the vertical grayscale integral vector; In S222, the lower limit of the distribution characteristic range is The calculation formula is: ； The S23 comprises the following sub-steps: S231, extracting the convex hull of the sensitive contour area; S232, extracting the distance between the convex hull and the sensitive contour area; S233, calculating a sensitive comparison coefficient according to the distance between the convex hull and the sensitive contour area; In S233, the sensitivity comparison coefficient The calculation formula is: ； In the formula, Represents the distance between the convex hull and the sensitive contour area, represents the perimeter of the sensitive contour area, represents the base of natural logarithms; The convex hull refers to the minimum convex polygon of the contour. The distance between the convex hull and the original contour is obtained by calculating the distance from each contour point to its nearest convex hull corner point and then summing them.

2. The automatic passenger face recognition method according to claim 1, characterized in that: In S3, the Bhattacharyya distance between the verification face image and each standard image in the face database is calculated, and the Bhattacharyya distance and the sensitive comparison coefficient are averaged to obtain the final similarity of each standard image. If there is a standard image whose final similarity is greater than or equal to the threshold, the passenger's recognition result is passed, otherwise the passenger's recognition result is failed.

Images