WO2018001092A1

WO2018001092A1 - Face recognition method and apparatus

Info

Publication number: WO2018001092A1
Application number: PCT/CN2017/088219
Authority: WO
Inventors: 朱海涛
Original assignee: 中兴通讯股份有限公司
Priority date: 2016-06-29
Filing date: 2017-06-14
Publication date: 2018-01-04
Also published as: CN107545220A

Abstract

A face recognition method and apparatus, the method comprising: acquiring a user face image (S101); detecting key point coordinates of the face image and correcting the face image using the detected key point coordinates (S102); recalculating the key point coordinates of the corrected face image and extracting a user face feature according to the recalculated key point coordinates (S103); and performing face verification using the face feature and the recalculated key point coordinates (S104). By means of detecting key point coordinates of a face image, correcting the face image by means of the key point coordinates, extracting a user face feature from the corrected face image, and finally performing face verification, precision and accuracy of face recognition are improved.

Description

Face recognition method and device

Technical field

The present disclosure relates to the field of communications technologies, and in particular, to a face recognition method and apparatus.

Background technique

As a biometric feature, human face has the advantages of not being lost, difficult to be copied, convenient to collect, unique, and undetected. It is getting more and more people's attention and has entered various fields of social life. Compared with other human biometric recognition systems such as retina, fingerprint, iris, voice, palm print, etc., face recognition system has a wide application prospect because of its convenience and friendliness, especially in face recognition access control attendance system, people. Face recognition ATM intelligent video alarm system, face recognition, public security offenders, intelligent alarm system identification, video conferencing and medical applications have become a research hotspot in the field of pattern recognition and content retrieval.

Feature extraction and selection are the core issues of face recognition and the basis for subsequent correct recognition. The face image acquisition process is often interfered by factors such as illumination changes and face pose changes. The traditional face recognition process compares the acquired face features with face samples after extracting facial features. To see if it is the same person. In this way, simply extracting facial features for comparison results in low face recognition accuracy and poor accuracy.

Summary of the invention

An object of the present disclosure is to provide a face recognition method, which solves the problem that the face recognition technology in the related art simply extracts facial features for comparison, resulting in low accuracy and poor accuracy of face recognition.

In order to achieve the above object, an embodiment of the present disclosure provides a face recognition method, including: acquiring a user face image; performing key point coordinate detection on the face image, and using the detected key point coordinates to Correcting the face image; recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates; and using the face feature and the recalculated key point Coordinates for face verification.

The embodiment of the present disclosure further provides a face recognition device, including: an acquisition unit, configured to acquire a user face image;

a correction unit configured to perform key point coordinate detection on the face image, and correct the face image using the detected key point coordinates; and extract an unit to recalculate the corrected face image Key point coordinates, and extracting user face features according to the recalculated key point coordinates; and a verification unit configured to perform face verification using the face features and the recalculated key point coordinates.

Embodiments of the present disclosure also provide a computer storage medium having stored therein one or more programs executable by a computer, the one or more programs being executed by the computer to cause the computer to perform as described above A method of face recognition is provided.

One of the above technical solutions has the following advantages or advantages: the face recognition method and apparatus of the present disclosure acquires a user face image; performs key point coordinate detection on the face image, and uses the detected Correcting the face image by key point coordinates; recalculating the key point coordinates of the corrected face image, and according to the re The calculated key point coordinates are used to extract the user face feature; the face feature and the recalculated key point coordinates are used for face verification. In this way, by detecting the key point coordinates of the face image, correcting the face image through the key point coordinates, extracting the user face feature from the corrected face image, and performing face verification to improve the accuracy of the face recognition. Degree and accuracy.

DRAWINGS

FIG. 1 is a method for recognizing a face according to an embodiment of the present disclosure;

FIG. 2 is another method for recognizing a face according to an embodiment of the present disclosure;

FIG. 3 is a device for recognizing a face according to an embodiment of the present disclosure.

detailed description

The technical problems, the technical solutions, and the advantages of the present invention will be more clearly described in conjunction with the accompanying drawings and specific embodiments.

As shown in FIG. 1 , an embodiment of the present disclosure provides a method for face recognition, including the following steps. In step S101, a user face image is acquired. In this step, the device is activated to acquire an image of a face. The method for obtaining a user's face image may be that the user is photographed by using a camera, or the image of the user's face that has been photographed is selected, which is not limited herein.

In step S102, key point coordinates are detected on the face image, and the face image is corrected using the detected key point coordinates. In this step, the key point coordinate detection is performed on the face image, which may be a reference coordinate system set first, then the face image is placed in the set reference coordinate system, and then the key point coordinates are detected. Among them, the key point can be representative points on the face image, such as the facial features, nose and ears of the human face, or the characteristic points such as dimples and scorpions.

After the key point coordinates are detected, the face image is corrected using key point coordinates so that subsequent comparisons are more accurate. For example, on the acquired face image, the face is a bit awkward, or slightly sideways, so in order to improve the accuracy of recognition, the face needs to be squared and then verified. Correcting the face image may be based on the detected key point coordinates, in the set reference coordinate system, the face image is translated or rotated to make the face image in a positive direction, and then the face image is performed, for example. Remove the blur points and other corrections.

In step S103, the key point coordinates are recalculated for the corrected face image, and the user face feature is extracted according to the recalculated key point coordinates. In this step, after the face image is corrected, since the face image has been translated or rotated in the reference coordinate system, the key point coordinates on the face image have changed, and at this time, the face is recalculated. The key point coordinates on the image, the method of recalculating the key point coordinates may be implemented by performing the corresponding transformation on the key point coordinates detected in step S102 according to the transformation performed in the face correction in S102, or in the reference coordinates. The new coordinates of the key points on the face image are re-extracted, and the user face features are extracted according to the recalculated key point coordinates, wherein the user face features are not limited to one, and multiple user face features can be extracted for subsequent Verification to improve recognition accuracy. In order to verify the accuracy and effectiveness, the present disclosure preferentially extracts user facial features with relatively high importance, such as facial features.

In step S104, the face verification is performed using the face feature and the recalculated key point coordinates. In this step, after extracting various parts of the human face, that is, the user's face features, the face samples in the face database are compared to perform face verification.

For example, in order to improve the verification accuracy and the recognition accuracy, in the embodiment, multiple facial features may be verified, but since the information carried by the features of the various parts of the face and the importance of the information are different, the calculation is performed. These feature similarities cannot be simply weighted averaged for overall similarity. In order to make good use of the information volume of different parts of the face and the importance of the information, the present disclosure uses a Support Vector Machine (SVM) algorithm (hereinafter referred to as SVM) to train through a large number of training samples, and finally obtains a SVM classifier. User face verification is performed using the SVM classifier. The SVM classifier can effectively utilize the information carried by different parts of the face and the importance of the information, and comprehensively evaluate the similarity of each part of the face to obtain the final verification result.

The face recognition method of the present disclosure includes: acquiring a user face image; performing key point coordinate detection on the face image, and correcting the face image using the detected key point coordinates; The face image recalculates the key point coordinates, and extracts the user face feature according to the recalculated key point coordinates; and uses the face feature and the recalculated key point coordinates to perform face verification. In this way, by detecting the key point coordinates of the face image, correcting the face image through the key point coordinates, extracting the user face feature from the corrected face image, and performing face verification to improve the accuracy of the face recognition. Degree and accuracy.

As shown in FIG. 2, an embodiment of the present disclosure provides a method for face recognition, including the following steps. In step S201, a user face image is acquired. In step S202, key point coordinate detection is performed on the face image, and the face image is aligned according to the preset direction using the detected key point coordinates, and then the aligned face image is normalized. .

It should be noted that the step S202 is not limited to the step S202. Embodiments may also be implemented by setting a coordinate system on the face image to compare a coordinate system on the face image with a preset reference coordinate system, that is, step 202 may be replaced with the person The coordinate system is set on the face image, and the embodiment in which the coordinate system on the face image is compared with the preset reference coordinate system is replaced or replaced with the feature in step 102. In step 202, the key point coordinate detection is performed on the face image, which may be a reference coordinate system set first, and then the face image is placed in the set reference coordinate system, and then the key point coordinates are detected. Among them, the key point can be representative points on the face image, such as the facial features, nose and ears of the human face, or the characteristic points such as dimples and scorpions.

After the key point coordinates are detected, the face image is first aligned on the face image, and the face image is aligned on the face image. The face image is obtained on the face image. The face is somewhat awkward or slightly side-faced. The face image is squared in the reference coordinate system, so that the face image is presented in the preset positive direction.

The alignment method may be based on the extracted key point coordinates, calculated in the reference coordinate system, and after the translation or rotation, the face image is presented in the positive direction. For example, if the coordinates of the two eyes are obtained, it is found that the line between the eyes is at an angle to the X axis of the reference coordinate system, that is, the face image is awkward, and the face needs to be rotated by a certain angle to make the eyes The connection between the lines is parallel to the X axis, and then the face is translated to a certain extent, so that the center of the face on the face image coincides with the origin, that is, the face alignment is completed.

After the face alignment is completed, the face image is normalized to remove the blur points. Where normalization can be For coordinate image centering, shear invariance (x-shearing) normalization, scaling normalization, and rotation normalization.

In step S203, the key point coordinates are recalculated for the corrected face image, and the user face feature is extracted according to the recalculated key point coordinates.

In step S204, face verification is performed using the face feature and the recalculated key point coordinates.

Step S201, step S203, and step S204 are the same as steps S101, S103, and S104 in the first embodiment of the present disclosure, and are not described herein.

Optionally, step S203 includes: recalculating the key point coordinates of the corrected face image, and dividing the face feature area by using the coordinate position of the retrieved key points to extract the user face feature. In this step, since the face image is previously corrected, the position of the user's face image is changed. At this time, the key point coordinates need to be recalculated in the reference coordinate system, and the coordinate position of the retrieved key point is used. A face feature area is divided on the face image, and a user face feature is extracted in the divided face feature area.

In this embodiment, the face feature area mainly includes a first feature area and a second feature area, wherein the first feature area is an area composed of an eyebrow, an eye and a nose, and an eyebrow, an eye and a surrounding part of the nose on the face image, The second feature area is an area composed of a mouth and a peripheral portion of the mouth on the face image.

Optionally, the step S203 further includes: extracting the feature 1 of the first feature area, the feature 2 of the second feature area, the key point position coordinate of the facial face of the face as the feature 3 of the feature, and the key point position coordinate of the outer contour of the face as Feature four of the features.

In this embodiment, in order to improve the accuracy and accuracy of the face recognition, optionally, four features are extracted for face verification, but it is not limited thereto, and more features and Feature points.

In step S203, the facial features include feature 1, feature two, feature three, and feature four, wherein feature one is an eyebrow, an eye and a nose on the face image, and a first feature region composed of an eyebrow, an eye, and a surrounding portion of the nose. The Gabor feature, feature 2 is the Gabor feature of the second feature region composed of the mouth and the surrounding part of the mouth, the feature 3 is the coordinate position of the retrieved key face of the face, and the feature 4 is the outline of the face on the face image. The coordinate position of the key points, such as the coordinate positions of the cheeks and chin.

In this embodiment, the first feature region and the second feature region are characterized by a Gabor feature, but are not limited thereto. In other embodiments, the first feature region and the second feature region may also be characterized by an LBP feature and an HOG. Features, etc. In the present embodiment, the features of the first feature region and the second feature region use the same feature, which are both Gabor features, but are not limited thereto. In other embodiments, the first feature region and the second feature region are Features may also use different features, such as the first feature being a Gabor feature, the second feature being an LBP feature, or other combinations.

Optionally, step S204 includes: comparing feature 1, feature 2, feature 3, and feature 4 with a preset face sample, respectively, and obtaining similarity 1 of feature 1, similarity 2 of feature 2, and feature 3 The similarity degree of similarity three and feature four; four-dimensional feature vectors composed of similarity 1, similarity 2, similarity 3 and similarity are classified by using a preset algorithm, and the user face and the preset person are calculated and verified. Whether the face sample belongs to the same person.

In this step S204, after extracting the features of each part of the person's face, that is, after extracting the feature one, the feature two, the feature three, and the feature four, respectively, the feature one, the feature two, the feature three, and the feature four are respectively Corresponding to the face sample The feature of the part is compared, and the similarity of feature one, the similarity of feature two, the similarity of feature three, and the similarity of feature four are obtained.

It should be noted that the similarity degree 1, the similarity degree 2, the similarity degree 3, and the similarity degree 4 can be compared and calculated at the same time, and can be separately compared and calculated, but the order is not distinguished.

Then, using the preset algorithm, the four-dimensional feature vectors composed of the similarity one, the similarity two, the similarity three and the similarity four are classified, and it is calculated whether the user face and the preset face sample belong to the same person.

In this embodiment, the similarity degree 1, the similarity degree 2, the similarity degree 3, and the similarity degree 4 may be cosine similarity, but is not limited thereto. In other embodiments, the Euclidean distance similarity and the Mahalanobis distance may be used. Similarity or Hamming distance similarity.

For example, since the amount of information carried by the features of each part of the face and the importance of the information are different, the weighted average of these feature similarities cannot be simply calculated when calculating the overall similarity. Therefore, in the present embodiment, the similarity one, the similarity two, the similarity three, and the similarity four are combined into one four-dimensional feature vector, and the four-dimensional feature vector is classified by using a preset SVM classifier, wherein the SVM The classifier is trained in advance through a large number of training samples through the SVM algorithm, and finally the SVM classifier is obtained. The SVM classifier can effectively utilize the information carried by different parts of the face and the importance of the information, and comprehensively evaluate the similarity of each part of the face to obtain the final verification result.

In order to improve the verification accuracy and the recognition accuracy, in the embodiment, the four-dimensional feature vectors composed of four feature similarities are classified, but the present invention is not limited thereto. In other embodiments, for example, five-dimensional features may be extracted as needed. Vector, six-dimensional feature vectors or more dimensional feature vectors for face recognition verification.

The face recognition method of the present disclosure includes: acquiring a user face image; performing key point coordinate detection on the face image, and using the detected key point coordinates, aligning the face image according to the preset direction, and then aligning The subsequent face image is normalized; the key point coordinates are recalculated for the corrected face image, and the user face feature is extracted according to the recalculated key point coordinates; the face feature and the recalculated key point are used. Coordinates for face verification. In this way, by detecting the key point coordinates of the face image, correcting the face image through the key point coordinates, extracting the user face feature from the corrected face image, and performing face verification to improve the accuracy of the face recognition. Degree and accuracy.

As shown in FIG. 3, an embodiment of the present disclosure provides an apparatus for recognizing a face, including: an obtaining unit 301 configured to acquire a user face image; and a correcting unit 302 configured to perform key point coordinate detection on the face image, and Correcting the face image using the detected key point coordinates; extracting unit 303, recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates; and the verification unit 304 , set to use face features and recalculated keypoint coordinates for face verification.

Optionally, the correcting unit 302 is configured to perform face alignment on the face image according to the preset direction using the detected key point coordinates, and then normalize the aligned face image.

Optionally, the extracting unit 303 is configured to recalculate the key point coordinates of the corrected face image, and divide the face feature area by using the coordinate position of the retrieved key points to extract the user face feature.

Optionally, the face feature area includes a first feature area composed of an eyebrow, an eye and a nose and a surrounding portion thereof, and a second feature area composed of a mouth and a surrounding portion thereof, and the extracting unit 303 is configured to extract features of the first feature area One, The feature of the second feature area is the key point position coordinate of the facial features of the face as the feature three of the feature and the key point position coordinate of the outer contour of the face as the feature four of the feature.

Optionally, the verification unit 304 is configured to compare the feature 1, the feature 2, the feature 3, and the feature 4 with the preset face samples respectively, and obtain the similarity of the feature one, the similarity of the feature two, and the feature three. Similarity 3 and similarity 4 of feature 4;

The four-dimensional feature vector consisting of similarity 1, similarity 2, similarity 3 and similarity 4 is classified by using a preset algorithm, and it is calculated whether the user face and the preset face sample belong to the same person.

a device for recognizing a face of an embodiment of the present disclosure, the acquiring unit is configured to acquire a user face image; the correcting unit is configured to perform key point coordinate detection on the face image, and perform face image detection using the detected key point coordinates Correcting; extracting unit, recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates; the verification unit is set to use the face feature and the recalculated key point coordinates Perform face verification. In this way, by detecting the key point coordinates of the face image, correcting the face image through the key point coordinates, extracting the user face feature from the corrected face image, and performing face verification to improve the accuracy of the face recognition. Degree and accuracy.

It will be understood by those skilled in the art that all or part of the steps of implementing the above embodiments may be performed by hardware associated with program instructions, which may be stored in a computer readable medium, including when executed, including The following steps: obtaining a user face image; performing key point coordinate detection on the face image, and correcting the face image using the detected key point coordinates; recalculating the key point coordinates on the corrected face image, and according to the re The calculated key point coordinates are used to extract the user face features; the face features and the recalculated key point coordinates are used for face verification.

Optionally, the face image is corrected by using the detected key point coordinates, including: using the detected key point coordinates, performing face alignment on the face image according to a preset direction, and then performing the aligned face image. Normalized processing.

Optionally, recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates, including: recalculating the key point coordinates on the corrected face image, and using the retrieved The coordinate position of the key point divides the face feature area to extract the user face feature.

Optionally, the face feature area includes a first feature area composed of an eyebrow, an eye and a nose and a surrounding portion thereof, and a second feature area composed of the mouth and the surrounding portion thereof, and recalculating key point coordinates on the corrected face image And extracting the user face feature according to the recalculated key point coordinates, including: extracting the feature 1 of the first feature area, the feature 2 of the second feature area, and the coordinate position coordinates of the facial features of the face as the feature three of the feature And the key point position coordinates of the outer contour of the face are taken as the feature four of the feature.

Optionally, the step of performing face verification using the face feature and the recalculated key point coordinates comprises: comparing the feature 1, the feature 2, the feature 3, and the feature 4 with the preset face sample respectively, and obtaining The similarity degree of feature one, the similarity degree of feature two, the similarity degree of feature three, and the similarity degree of feature four are four; using the preset algorithm to combine similarity 1, similarity 2, similarity 3 and similarity 4 The four-dimensional feature vector is classified, and it is calculated whether the user face and the preset face sample belong to the same person.

The storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

The above is a preferred embodiment of the present disclosure, it should be noted that for those of ordinary skill in the art, Numerous modifications and improvements may be made without departing from the principles of the present disclosure, and such modifications and refinements are also considered to be within the scope of the present disclosure.

Claims

A face recognition method includes:

Obtaining a user face image;

Performing key point coordinate detection on the face image, and correcting the face image using the detected key point coordinates;

Recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates;

The face verification is performed using the face features and the recalculated key point coordinates.
The method of claim 1 wherein said correcting said face image using said detected keypoint coordinates comprises:

Using the detected key point coordinates, the face image is aligned according to a preset direction, and the aligned face image is normalized.
The method of claim 1, wherein the recalculating key point coordinates for the corrected face image and extracting user face features based on the recalculated key point coordinates comprises:

The key point coordinates are recalculated for the corrected face image, and the face feature area is divided by the coordinate position of the retrieved key points to extract the user face feature.
The method according to any one of claims 1 to 3, wherein said face feature region comprises a first feature region composed of an eyebrow, an eye and a nose and a surrounding portion thereof, and a second feature region composed of a mouth and a surrounding portion thereof, And the step of recalculating the key point coordinates of the corrected face image and extracting the user face feature according to the recalculated key point coordinates, including:

Extracting the feature 1 of the first feature region, the feature 2 of the second feature region, the key point position coordinate of the facial face of the face as the feature 3 of the feature, and the key point position coordinate of the outer contour of the face as the feature 4 of the feature.
The method of claim 4, wherein the step of performing face verification using the face features and the recalculated keypoint coordinates comprises:

Comparing the feature 1, the feature 2, the feature 3, and the feature 4 with a preset face sample respectively, and obtaining the similarity of the feature one, and the similarity of the feature two , the similarity degree 3 of the feature 3 and the similarity 4 of the feature 4;

And classifying, by using a preset algorithm, the four-dimensional feature vector consisting of the similarity 1, the similarity 2, the similarity 3, and the similarity 4, and calculating whether the user face and the preset face sample are verified Belong to the same person.
A face recognition device comprising:

Obtaining a unit, configured to obtain a user face image;

a correction unit configured to perform key point coordinate detection on the face image, and correct the face image by using the detected key point coordinates;

Extracting unit, recalculating the key point coordinates of the corrected face image, and extracting the user face feature according to the recalculated key point coordinates;

The verification unit is configured to perform face verification using the face feature and the recalculated key point coordinates.
The apparatus according to claim 6, wherein the correcting unit is configured to perform face alignment on the face image in a preset direction using the detected key point coordinates, and then to align the face image Perform normalization.
The apparatus according to claim 6, wherein said extracting unit is configured to recalculate the key point coordinates of the corrected face image, and divide the face feature area using the coordinate position of the retrieved key point to extract the user Face features.
The device according to any of claims 6-8, wherein said facial feature region comprises a first feature region composed of an eyebrow, an eye and a nose and a surrounding portion thereof, and a second feature region composed of a mouth and a surrounding portion thereof, The extracting unit is configured to extract the feature 1 of the first feature region, the feature 2 of the second feature region, the key point position coordinate of the facial face of the face as the feature 3 of the feature, and the key point position coordinate of the outer contour of the face as Feature four of the features.
The apparatus according to claim 9, wherein the verification unit is configured to compare the feature one, the feature two, the feature three, and the feature four with a preset face sample, respectively The similarity of the feature one, the similarity of the feature two, the similarity of the feature three, and the similarity of the feature four;

And classifying, by using a preset algorithm, the four-dimensional feature vector consisting of the similarity 1, the similarity 2, the similarity 3, and the similarity 4, and calculating whether the user face and the preset face sample are verified Belong to the same person.
A computer storage medium storing execution instructions for performing the method of any one of claims 1 to 5.