Face recognition method for people wearing masks based on interpupillary distance
By collecting and expanding eye coordinates, and using a cascaded classifier and Euclidean model to calculate the distance between human eyes, the problem of low facial recognition accuracy under mask-wearing was solved, achieving efficient and accurate identity recognition.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-09
- Publication Date
- 2026-03-24
AI Technical Summary
Existing facial recognition technology has a lower accuracy rate when people are wearing masks, and traditional methods cannot effectively identify the identity of people wearing masks.
By collecting the coordinates of the two eyes and expanding the area to frame the face, facial key points are obtained using a cascaded classifier and image segmentation technology. The distance between the eyes is then calculated using a Euclidean model for identity matching.
It improves the accuracy of identity recognition when masks are worn, makes up for the shortcomings of traditional methods, and achieves fast and accurate identity recognition.
Smart Images
Figure CN116798106B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of facial recognition technology using masks, and more specifically, to a method for facial recognition using masks based on inter-eye distance. Background Technology
[0002] With the rapid development of social science and technology, facial recognition technology has demonstrated its advantages in identity verification, leading to the development of various facial recognition products that are widely used in scenarios such as mobile phone unlocking, financial payments, attendance tracking, and airport and train station security checks. During periods of widespread respiratory infections, wearing masks has become the norm for epidemic prevention. How facial recognition can function effectively, especially in crowded places like train stations, airports, shopping malls, and residential communities, is a key concern. Masks obscure faces, significantly reducing the number of recognizable facial features and lowering the accuracy of existing facial recognition algorithms. This makes it impossible to properly verify identity information while wearing masks, yet removing masks poses a risk of infection. Therefore, facial recognition while wearing masks has become a social hot topic.
[0003] While existing recognition models have improved the accuracy of small target detection, they also suffer from insufficient extraction of shallow features. Secondly, the accuracy of prediction is determined by the IOU (Intersection over Union) loss function, but the detection accuracy decreases as the value of the IOU loss function increases. In addition, in natural scenes, there are problems such as occlusion, dense crowds, and small-scale target detection, which make it difficult to identify the person. To address these issues, this invention provides a face recognition method for wearing masks based on inter-eye distance. Summary of the Invention
[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a face recognition method based on inter-eye distance for wearing masks, so as to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a face recognition method based on inter-eye distance for people wearing masks, comprising the following steps:
[0006] Step 10: Acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory;
[0007] The preprocessing involves reading the target image, converting it into a grayscale image, and then into a histogram.
[0008] Step 20: Detect the coordinates of the two eyes in the target image using a cascaded classifier, expand the coordinate range to obtain facial contour data, and then perform image segmentation processing.
[0009] Step 30: Obtain the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information;
[0010] Step 40: Obtain facial information and extract the required key points. Analyze the required key points using a convolution matrix to obtain the optimal points.
[0011] Step 50: Using the coordinates of the inner and outer corners of a single eye as a unit, calculate the interocular distance from the obtained optimal point using a Euclidean model and match it with the identity in the database to identify the identity and output the name.
[0012] The Euclidean model is as follows:
[0013] Let there be two points A and B with coordinates A(x1,y1) and B(x2,y2). The formula for the distance between points A and B is as follows: , Indicates the distance between the eyes.
[0014] Preferably, step 20 includes the following:
[0015] S201. Detect the coordinates of a single eye in the target image using a cascaded classifier, and enlarge the obtained coordinates by 4 times to frame the face to obtain the face coordinates. Obtain the face coordinates of the other eye in the same way. In the target image set, if the area of a single eye coordinate is less than one-tenth of the area of the largest eye, the target image will be removed.
[0016] S202. Compare each expanded coordinate range, determine the intersecting coordinates by conjunction, and outline the intersecting part.
[0017] S203. Perform segmentation processing on the target image to obtain a target segmented image. The segmentation processing method includes region-based segmentation and threshold-based segmentation.
[0018] Preferably, step 30 includes the following:
[0019] S301. Obtain the target segmentation image, detect face coordinates using facial key points, and obtain facial key points;
[0020] S302. Obtain coordinate data through facial key points and extract facial information;
[0021] S303. If the face coordinates are not extracted, return to the previous step.
[0022] Preferably, the facial key points estimate the facial geometry in the space by using facial landmark screen positions. Facial landmarks can mark facial features such as the eyes, nose, and mouth. The facial key points consist of three-dimensional coordinates, including a pose transformation matrix and a triangular mesh.
[0023] The pose transformation matrix is a linear mapping from a normalized face model to each face landmark set, thereby tracking and aligning the face.
[0024] Preferably, the triangular grid refers to setting a list of face landmarks in the target image. Each landmark consists of x, y, and z. x and y are normalized to [0.0, 1.0] by the image width and height, respectively, and z represents the depth. With the center of the head as the origin, the closer the landmark is to the camera, the smaller the value, and vice versa. The key points of the face can be obtained from the three-dimensional coordinates.
[0025] Preferably, step 40 includes the following:
[0026] S401. Obtain facial information, traverse the set of facial key points, and extract the required key points;
[0027] S402. Classify the acquired key points and label them as orbital key point K, inner canthus key point N, and outer canthus key point W;
[0028] S403. Obtain the key points K of the eye socket, smooth them using the convolution matrix of the mean filter function, and then extract the corner points of the eye using corner detection and mark them as corner points D.
[0029] S404. Obtain the key point N at the inner corner of the eye and the key point W at the outer corner of the eye. Loop through all points detected by the corner point detection. If a point is detected that matches the corner point D, obtain the corner point. Perform a sub-pixel level point positioning algorithm on the corner point D, the key point N at the inner corner of the eye, and the key point W at the outer corner of the eye to obtain the optimal point.
[0030] Preferably, step 50 includes the following:
[0031] S501. Obtain key points J at the inner and outer corners of the eyes. Using a single eye as a unit 1, calculate the distance between the two eyes using the Euclidean model based on the obtained best points.
[0032] S502. Obtain four different photos of each person, store them, calculate the interocular distance, calculate the variance of the interocular distance for each person, and store the data values in the database.
[0033] S503: During detection, the database is read and identified. If the two match, the identity will be identified; otherwise, it is unknown.
[0034] Preferably, the interocular distance is compared with a unit of 1; the closer the similarity is numerically to the unit of 1, the more accurate the measurement. When measuring distance, the following method is used: The formula yields the judgment value Jl between the two eyes.
[0035] Preferred options also include:
[0036] The image acquisition module is used to acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory.
[0037] The image segmentation module is used to detect the coordinates of two eyes in the target image through a cascaded classifier, and then perform image segmentation processing after expanding the coordinate range to obtain facial contour data;
[0038] The key point detection module is used to acquire the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information.
[0039] The optimal point analysis module is used to acquire facial information and extract the required key points. The required key points are then analyzed using a convolution matrix to obtain the optimal points.
[0040] The identity recognition module uses the coordinates of the inner and outer corners of each eye as a unit, calculates the distance between the two eyes using the Euclidean model based on the best point obtained, matches it with the identity in the database, identifies the identity, and outputs the name.
[0041] The technical effects and advantages of this invention are as follows:
[0042] (1) This invention collects the coordinates of two eyes and enlarges the frame of the face by a multiple, and frames out the intersecting parts to segment the target image, thus completing the target segmentation image and preparing for the next step of obtaining the coordinates of the key points of the face, which is conducive to improving the image recognition rate; the face coordinates are detected and located by using the pose transformation matrix and triangular mesh, and the face coordinates are merged and optimized to quickly obtain the key point data of the face, realizing parallel data computing and improving the efficiency of detection.
[0043] (2) This invention extracts and classifies facial key points to obtain key points K of the eye socket, key points N of the inner corner of the eye and key points W of the outer corner of the eye. Key points K of the eye socket are smoothed by a convolution matrix and corner points are extracted to obtain each corner point of the eye. Key points N of the inner corner of the eye and key points W of the outer corner of the eye are cyclically focused and processed with corner point D by a sub-pixel level point positioning algorithm. If the focus is consistent, the corner point is recorded as the best point, which is beneficial for identity matching based on the best point.
[0044] (3) This invention calculates the distance between human eyes using the Euclidean model and identifies the identity by matching and comparing the distance between human eyes, which improves the accuracy of face recognition for people wearing masks and makes up for the fact that traditional recognition methods can only detect masks but cannot identify the identity of people wearing masks. Attached Figure Description
[0045] Figure 1 This is a block diagram of the method structure of the present invention.
[0046] Figure 2 This is a flowchart of the method of the present invention.
[0047] Figure 3 This is a system structure block diagram of the present invention.
[0048] Figure 4 This is a diagram showing the location of key facial features in this invention.
[0049] Figure 5 This is a variance line graph for the present invention.
[0050] Figure 6 This is a grayscale comparison image of the present invention. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] Example 1
[0053] Please see Figure 1-5 As shown, this embodiment provides a face recognition method for wearing masks based on inter-eye distance, including the following steps:
[0054] Step 10: Acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory;
[0055] The preprocessing involves reading the target image, converting it into a grayscale image, and then into a histogram.
[0056] Step 20: Detect the coordinates of the two eyes in the target image using a cascaded classifier, expand the coordinate range to obtain facial contour data, and then perform image segmentation processing.
[0057] In this embodiment, it should be specifically explained that step 20 includes the following:
[0058] S201. Detect the coordinates of a single eye in the target image using a cascaded classifier, and enlarge the obtained coordinates by 4 times to frame the face to obtain the face coordinates. Obtain the face coordinates of the other eye in the same way. In the target image set, if the area of a single eye coordinate is less than one-tenth of the area of the largest eye, the target image will be removed.
[0059] S202. Compare each expanded coordinate range, determine the intersecting coordinates by conjunction, and outline the intersecting part.
[0060] S203. Perform segmentation processing on the target image to obtain a target segmented image. The segmentation processing method includes region-based segmentation and threshold-based segmentation. The specific segmentation method is not limited.
[0061] In this embodiment, the target image is segmented by acquiring the coordinates of the two eyes, enlarging the bounding box around the face, and outlining the intersecting parts. This segmentation of the target image prepares the next step for obtaining the coordinates of the key facial points, which helps to improve the image recognition rate.
[0062] Step 30: Obtain the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information.
[0063] In this embodiment, it should be specifically explained that step 30 includes the following:
[0064] S301. Obtain the target segmentation image, detect face coordinates using facial key points, and obtain facial key points;
[0065] S302. Obtain coordinate data through facial key points and extract facial information;
[0066] S303. If the face coordinates are not extracted, return to the previous step.
[0067] It should be specifically noted that the facial key points are estimated by using facial landmarks on the screen to determine the facial geometry within that space. These facial landmarks can mark features such as the eyes, nose, and mouth. The facial key points consist of three-dimensional coordinates, including a pose transformation matrix and a triangular mesh.
[0068] The pose transformation matrix is a linear mapping from a normalized face model to each set of face landmarks, thereby tracking and aligning the face;
[0069] The triangular grid refers to setting a list of facial landmarks in the target image. Each landmark consists of x, y, and z. x and y are normalized to [0.0, 1.0] by the image width and height, respectively, and z represents the depth. With the center of the head as the origin, the closer the landmark is to the camera, the smaller the value, and vice versa. The facial key points can be obtained from the three-dimensional coordinates.
[0070] For example, after segmenting the image, the 468 points located by facial key point detection are used to determine whether it is a face. If it is, the coordinates are obtained using the facial key points (10,356,152,127) and the face information is extracted. Otherwise, it is removed and the process returns to the previous step.
[0071] In this embodiment, face coordinates are detected and located using a pose transformation matrix and a triangular mesh. The face coordinates are merged and optimized to quickly obtain face key point data, thereby achieving parallel data computation and improving the efficiency of detection.
[0072] Step 40: Obtain facial information and extract the required key points. Analyze the required key points using a convolution matrix to obtain the optimal points.
[0073] In this embodiment, it should be specifically explained that step 40 includes the following:
[0074] S401. Obtain facial information, traverse the set of facial key points, and extract the required key points;
[0075] S402. Classify the acquired key points and label them as orbital key point K, inner canthus key point N, and outer canthus key point W;
[0076] S403. Obtain the key points K of the eye socket, smooth them using the convolution matrix of the mean filter function, and then extract the corner points of the eye using corner detection and mark them as corner points D.
[0077] S404. Obtain the key point N at the inner corner of the eye and the key point W at the outer corner of the eye. Loop through all points detected by the corner point detection. If a point is detected that matches the corner point D, obtain the corner point. Perform a sub-pixel level point positioning algorithm on the corner point D, the key point N at the inner corner of the eye, and the key point W at the outer corner of the eye to obtain the optimal point.
[0078] In this embodiment, it should be specifically explained that the corner detection is achieved by the goodFeaturesToTrack function finding the corner with the largest feature value in the target image, obtaining more refined corner coordinates, and combining it with a sub-pixel level point localization algorithm to calculate the coordinate position of the feature point in the image. The obtained coordinates are then compared with the coordinates of the key points of the face to obtain the optimal coordinate position of the inner and outer corners of the eyes.
[0079] In this embodiment, by extracting and classifying facial key points, key points K of the eye socket, key points N of the inner corner of the eye, and key points W of the outer corner of the eye are obtained. Key points K of the eye socket are smoothed using a convolution matrix and corner points are extracted to obtain various corner points of the eye. Key points N of the inner corner of the eye and key points W of the outer corner of the eye are cyclically focused and processed with corner point D using a sub-pixel level point positioning algorithm. If the corner points are consistent, the corner point is recorded as the best point, which is beneficial for identity matching based on the best point.
[0080] Step 50: Using the coordinates of the inner and outer corners of a single eye as a unit, calculate the distance between the two eyes using the Euclidean model based on the obtained optimal point, match it with the identity in the database, identify the identity, and output the name.
[0081] In this embodiment, it should be specifically explained that step 50 includes the following:
[0082] S501. Obtain key points J at the inner and outer corners of the eyes. Using a single eye as a unit 1, calculate the distance between the two eyes using the Euclidean model based on the obtained best points.
[0083] S502. Obtain four different photos of each person, store them, calculate the interocular distance, calculate the variance of the interocular distance for each person, and store the data values in the database.
[0084] S503: During detection, the database is read and identified. If the two match, the identity will be identified; otherwise, it is unknown.
[0085] In this embodiment, the Euclidean model should be specifically described as follows:
[0086] Let there be two points A and B with coordinates A(x1,y1) and B(x2,y2). The formula for the distance between points A and B is as follows: , Representing interocular distance, in the process of calculating interocular distance, the smaller the Euclidean distance, the greater the similarity between two target objects, and vice versa.
[0087] Furthermore, the interocular distance is compared to a unit of 1; the closer the similarity is numerically to a unit of 1, the more accurate the measurement. When measuring distance, the interocular distance is... The formula yields the judgment value Jl between the two eyes.
[0088] In this embodiment, the distance between human eyes is calculated using the Euclidean model, and the identity is identified by matching and comparing the distance between human eyes. This improves the accuracy of face recognition for people wearing masks and makes up for the fact that traditional recognition methods can only detect masks but cannot identify the identity of people wearing masks.
[0089] Example 2
[0090] Please see Figure 1-5As shown, this embodiment provides a face recognition system for masked individuals based on inter-eye distance, including:
[0091] The image acquisition module is used to acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory.
[0092] The image segmentation module is used to detect the coordinates of two eyes in the target image through a cascaded classifier, and then perform image segmentation processing after expanding the coordinate range to obtain facial contour data;
[0093] The key point detection module is used to acquire the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information.
[0094] The optimal point analysis module is used to acquire facial information and extract the required key points. The required key points are then analyzed using a convolution matrix to obtain the optimal points.
[0095] The identity recognition module uses the coordinates of the inner and outer corners of each eye as a unit, calculates the distance between the two eyes using the Euclidean model based on the best point obtained, matches it with the identity in the database, identifies the identity, and outputs the name.
[0096] Experimental example:
[0097] Please see Figure 6 As shown, when the camera is shooting, the mask recognition methods in Examples 1 and 2 can both identify the identity in the target image. The corner detection used in this experiment is the goodFeaturesToTrack function, which has the function of identifying strong corners in the image and can find the corner with the largest feature value in the image. This function uses Harris and Shi-Tomasi corner detection algorithms to determine the strong corners of the image.
[0098] The Shi-Tomasi corner detection method is an improvement upon Harris. To accurately detect strong corners, this paper applies the Shi-Tomasi detection method. Further refining corner coordinates are obtained, and combined with a sub-pixel-level point localization algorithm, the coordinate positions of feature points in the image can be calculated. Comparing the obtained coordinates with the coordinates of facial key points yields the optimal coordinate positions of the inner and outer corners of the eyes. Specific data are shown in Tables 1 and 2 below.
[0099] Table 1 shows the interocular distance data for the target photograph.
[0100]
[0101] Table 2 shows the calculated variance values.
[0102]
[0103] Therefore, the mask recognition method provided by this invention can quickly and accurately identify individuals.
[0104] In summary, it has the following effects:
[0105] By acquiring the coordinates of two eyes and enlarging the bounding box around the face, and outlining the intersecting parts, the target image is segmented. This segmentation prepares the next step for obtaining the coordinates of facial key points, which helps improve the image recognition rate. The face coordinates are detected and located using a pose transformation matrix and a triangular mesh. The face coordinates are then merged and optimized to quickly obtain facial key point data. This enables parallel data computation and improves the efficiency of detection.
[0106] By extracting and classifying facial key points, key points K of the eye socket, key points N of the inner corner of the eye, and key points W of the outer corner of the eye are obtained. Key points K of the eye socket are smoothed using a convolution matrix and corner points are extracted to obtain the corner points of the eye. Key points N of the inner corner of the eye and key points W of the outer corner of the eye are repeatedly focused and processed with corner points D using a sub-pixel level point localization algorithm. If the focus is consistent, the corner point is recorded as the best point, which is beneficial for identity matching based on the best point.
[0107] By calculating the distance between human eyes using the Euclidean model and identifying individuals through matching and comparison of these distances, the accuracy of facial recognition for people wearing masks is improved. This overcomes the limitation of traditional recognition methods that can only detect masks but cannot identify the identity of the person wearing the mask.
[0108] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.
[0109] Other embodiments or specific implementations of the face recognition method for wearing masks based on inter-eye distance in this application can be referred to the above method embodiments, and will not be repeated here.
[0110] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A face recognition method based on inter-eye distance for face recognition while wearing a mask, characterized in that... This includes the following steps: Step 10: Acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory; The preprocessing involves reading the target image, converting it into a grayscale image, and then into a histogram. Step 20: Detect the coordinates of the two eyes in the target image using a cascaded classifier, expand the coordinate range to obtain facial contour data, and then perform image segmentation processing. Step 30: Obtain the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information; Step 40: Obtain facial information and extract the required key points. Analyze the required key points using a convolution matrix to obtain the optimal points. Step 50: Using the coordinates of the inner and outer corners of a single eye as a unit, calculate the interocular distance from the obtained optimal point using a Euclidean model and match it with the identity in the database to identify the identity and output the name. The Euclidean model is as follows: Let there be two points A and B with coordinates A(x1,y1) and B(x2,y2). The formula for the distance between points A and B is as follows: , Indicates interocular distance; Step 20 includes the following: S201. Detect the coordinates of a single eye in the target image using a cascaded classifier, and expand the obtained coordinate range by 4 times to frame the face to obtain the face coordinates. Obtain the face coordinates of the other eye in the same way. In the target image set, if the area of a single eye coordinate is less than one-tenth of the area of the largest eye, the target image will be removed. S202. Compare each expanded coordinate range, determine the intersecting coordinates by conjunction, and outline the intersecting part. S203. Perform segmentation processing on the target image to obtain a target segmented image. The segmentation processing method includes region-based segmentation and threshold-based segmentation.
2. The face recognition method based on inter-eye distance for wearing masks according to claim 1, characterized in that: Step 30 includes the following: S301. Obtain the target segmentation image, detect face coordinates using facial key points, and obtain facial key points; S302. Obtain coordinate data through facial key points and extract facial information; S303. If the face coordinates are not extracted, return to the previous step.
3. The face recognition method based on inter-eye distance for wearing masks according to claim 2, characterized in that: The facial keypoints estimate the facial geometry in space by using facial landmark screen positions. Facial landmarks mark the eyes, nose, and mouth of the face. The facial keypoints consist of three-dimensional coordinates, including a pose transformation matrix and a triangular mesh. The pose transformation matrix is a linear mapping from a normalized face model to each face landmark set, thereby tracking and aligning the face.
4. The face recognition method based on inter-eye distance for wearing masks according to claim 3, characterized in that: The triangular grid refers to setting a list of facial landmarks in the target image. Each landmark consists of x, y, and z. x and y are normalized to [0.0, 1.0] by the image width and height, respectively, and z represents the depth. With the center of the head as the origin, the closer the landmark is to the camera, the smaller the value, and vice versa. The facial key points can be obtained from the three-dimensional coordinates.
5. The face recognition method based on inter-eye distance for wearing masks according to claim 1, characterized in that: Step 40 includes the following: S401. Obtain facial information, traverse the set of facial key points, and extract the required key points; S402. Classify the acquired key points and label them as orbital key point K, inner canthus key point N, and outer canthus key point W; S403. Obtain the key points K of the eye socket, smooth them using the convolution matrix of the mean filter function, and then extract the corner points of the eye using corner detection and mark them as corner points D. S404. Obtain the key point N at the inner corner of the eye and the key point W at the outer corner of the eye. Loop through all points detected by the corner point detection. If a point is detected that matches the corner point D, obtain the corner point. Perform a sub-pixel level point positioning algorithm on the corner point D, the key point N at the inner corner of the eye, and the key point W at the outer corner of the eye to obtain the optimal point.
6. The face recognition method based on inter-eye distance for wearing masks according to claim 1, characterized in that: Step 50 includes the following: S501. Obtain key points at the inner and outer corners of the eyes. Using a single eye as a unit, calculate the distance between the two eyes using the Euclidean model based on the best obtained points. S502. Obtain four different photos of each person, store them, calculate the interocular distance, calculate the variance of the interocular distance for each person, and store the data values in the database. S503: During detection, the database is read and identified. If the two match, the identity will be identified; otherwise, it is unknown.
7. The face recognition method based on inter-eye distance for wearing masks according to claim 1, characterized in that: The interocular distance is compared to a unit of 1; the closer the similarity is numerically to a unit of 1, the more accurate the measurement. When measuring distance, the formula is used. The judgment value Jl between the two eyes is obtained.
8. A face recognition method based on inter-eye distance for wearing masks according to claim 1, characterized in that: include: The image acquisition module is used to acquire several sets of target images through the camera, preprocess the several sets of target images, and create an image data set to be stored in the cloud or memory. The image segmentation module is used to detect the coordinates of two eyes in the target image through a cascaded classifier, and then perform image segmentation processing after expanding the coordinate range to obtain facial contour data; The key point detection module is used to acquire the target segmentation image, detect face coordinates using facial key points, determine whether it is a face, obtain coordinates through facial key points, and extract face information. The optimal point analysis module is used to acquire facial information and extract the required key points. The required key points are then analyzed using a convolution matrix to obtain the optimal points. The identity recognition module uses the coordinates of the inner and outer corners of each eye as a unit, calculates the distance between the two eyes using the Euclidean model based on the best point obtained, matches it with the identity in the database, identifies the identity, and outputs the name.
Citation Information
Patent Citations
2D and 3D fused face recognition method and system
CN110458142A
Intelligent face recognition method
CN112800937A