CN107346422A - A kind of living body faces recognition methods based on blink detection - Google Patents

Abstract

The invention discloses a kind of living body faces recognition methods based on blink detection, image in the face video that this method is detected by reading camera, and pass through distance value between palpebra inferior characteristic point on the face in calculating per two field picture, when reading frame number reaches frame number threshold value, according in the image read of statistics on face between palpebra inferior characteristic point distance value fluctuation situation, judge whether the face in the face video is live body, if the face in the face video is live body, recognition of face is carried out to the face in the face video.Therefore, computational methods of the present invention are simple, not high to image request, it is possible to increase the accuracy and efficiency of In vivo detection.

Description

A Live Face Recognition Method Based on Blink Detection

technical field

The invention relates to the technical field of face recognition, in particular to a method for detecting living human faces based on blink detection.

Background technique

Nowadays, face recognition systems are more and more used in daily life. In order to ensure the safety of the application, the face recognition system is required to prevent interference and counterfeiting of photos, videos, and 3D face models. Simple blink detection is Unable to meet the protection requirements.

Liveness detection algorithms are mainly divided into three categories: (1) Motion-based methods, which mainly analyze the motion trend of the image frame sequence to determine the background of the image or the unconscious actions of the user, but the calculation process is complicated. (2) Texture-based methods, mainly by finding out the salient distinguishing features of single-frame real faces and deceptive faces for liveness judgment, because deceptive faces will cause quality degradation, blurring, etc. in the process of secondary acquisition Variations in microtexture, but it only handles low-res print photo attacks well, not high-resolution photos. (3) Fusion-based methods, that is, to achieve the purpose of resisting multiple attack forms by fusing at least two living body discrimination methods. This method is divided into a feature layer fusion method and a score layer fusion method. The feature layer fusion method is to fuse multiple features in series and is in the research stage, while the score layer fusion method is to obtain the scores of multiple features and then weight them. The final score is obtained, but for features of different dimensions and different meanings, it cannot be accurately fused, which affects the efficiency and accuracy of living body recognition.

Therefore, the current liveness detection methods have shortcomings such as complex calculation methods and strict requirements on images, which seriously affect the accuracy or efficiency of liveness detection.

Contents of the invention

The purpose of the present invention is to solve the technical problems that the existing living body detection methods have shortcomings such as complex calculation methods and strict requirements on images, which seriously affect the accuracy or efficiency of living body detection.

In order to realize the above-mentioned purpose of the invention, the present invention provides the following technical solutions:

A kind of live face recognition method based on blink detection, it comprises the following steps,

Step 1: Read the image in the face video detected by the camera;

Step 2: Detect the face in the current frame image, and locate the feature points used to calibrate the upper and lower eyelids respectively;

Step 3: Calculate and save the distance value between the feature points of the upper and lower eyelids on the face in the current frame image;

Step 4: Judging whether the number of read frames reaches the frame number threshold; if not, jump to step 1 to read the next frame image in the face video;

Step 5: Count the fluctuation of the distance value between the upper and lower eyelid feature points on the human face in the read frame image, and judge whether the human face in the human face video is a living body according to the statistical fluctuation;

Step 6: If the face in the face video is a living body, perform face recognition on the face in the face video.

According to a specific implementation, in the second step of the living face recognition method based on blink detection in the present invention, the three-dimensional coordinate axes used to describe the head posture are determined, and the head is estimated according to the detected face in the current frame image. Attitude, to obtain the head attitude vector; calculate the angle between the head attitude vector of the current frame image and the head attitude vector of each previous frame image, if any angle value is greater than the angle threshold, then jump to step 1 .

Further, if any included angle value is greater than the angle threshold, a first prompt message is issued to remind the identified person to face the camera directly.

According to a specific embodiment, in step 2 of the living face recognition method based on blink detection in the present invention, the width of the face is calculated according to the detected face in the current frame image, if the width of the face in the current frame image is If it is smaller than the width threshold, go to step 1.

Further, if the width of the face in the current frame image is smaller than the width threshold, a second prompt message is issued to remind the recognized person to move his face closer to the camera.

According to a specific embodiment, in step 2 of the living face recognition method based on blink detection in the present invention, the size of the face is calculated according to the detected face in the current frame image, and the size of the face in the current frame image is calculated. The difference between the size and the face size of each previous frame image, if there is a difference greater than the face size threshold, then jump to step 1.

Further, if there is a difference greater than the face size threshold, a third prompt message is issued to remind the recognized person not to shake his head.

According to a specific implementation, in step five of the present invention's live face recognition method based on blink detection, the fluctuations are characterized by statistical values, and if the statistical values reach the fluctuation threshold, then the determination of the face video in the The human face is a living body; and, the statistical value is calculated using the following formula:

Among them, L represents the statistical value, s ₁ represents the distance value between the feature points of the upper and lower eyelids on the face of the first frame image, _si represents the distance value between the feature points of the upper and lower eyelids on the face of the i-th frame image, and n is the frame number threshold.

According to a specific implementation, in the living face recognition method based on blink detection of the present invention, before the number of read frames does not exceed the frame number threshold, a fourth prompt message is sent to prompt the person to be recognized to make a specific action, And identify whether the specific action appears in the subsequent several frames of images, if the specific action does not appear, stop face recognition.

Compared with prior art, the beneficial effect of the present invention:

1. The living face recognition method based on blink detection of the present invention, by reading the image in the face video detected by the camera, and by calculating the distance value between the feature points of the upper and lower eyelids on the face in each frame of the image, when reading the frame When the number of frames reaches the threshold of the frame number, according to the fluctuation of the distance value between the upper and lower eyelid feature points of the face in the read image, it is judged whether the face in the face video is a living body. If the human face is a living body, face recognition is performed on the human face in the human face video. Therefore, the calculation method of the present invention is simple, has low requirements on images, and can improve the accuracy and efficiency of living body detection.

2. The living face recognition method based on blink detection of the present invention, in the process of calculating the distance value between the upper and lower eyelid feature points on the human face in each frame of image, also judges whether the angle between the head pose vectors between the images exceeds the included angle Threshold, whether the width of the face in the image exceeds the width threshold, and whether the difference in face size between images exceeds the face size threshold, and a corresponding prompt message is sent to remind the recognized person to face the camera better to ensure Accurately acquire face detection video and improve the accuracy of face recognition.

3. The living face recognition method based on blink detection of the present invention sends a message to remind the person to make a specific action before the number of read frames does not exceed the threshold of the frame number, and then recognizes whether there is a specific action in subsequent frames of images. This specific action verifies whether the recognized person in the face detection video is a real person. If no specific action occurs, it indicates that the recognized person is not a real person, and face recognition is stopped. Therefore, the present invention can prevent the camera from capturing the human face displayed in the video, thereby affecting the result of human face recognition.

Description of drawings:

FIG. 1 is a flow chart of the living face recognition method based on blink detection in the present invention.

detailed description

The present invention will be further described in detail below in conjunction with test examples and specific embodiments. However, it should not be understood that the scope of the above subject matter of the present invention is limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

As shown in Figure 1, the present invention is based on the flow chart of the live face recognition method of blink detection; wherein, the method of the present invention comprises the following steps:

Step 1: Read the image in the face video detected by the camera.

Step 2: Detect the face in the current frame image, and locate the feature points used to calibrate the upper and lower eyelids. When detecting a human face, first detect all the feature points used for facial features, and then extract the feature points used to calibrate the upper and lower eyelids, and locate these feature points at the same time, lock these feature points in the image location information in .

Step 3: Calculate and save the distance value between the feature points of the upper and lower eyelids on the face in the current frame image. Among them, after locating the feature points used to calibrate the upper and lower eyelids, according to the position information of these feature points in the image, calculate the distance value between the feature points used to calibrate the upper eyelid and the feature points used to calibrate the lower eyelid . Since there are multiple feature points for marking the upper eyelid or lower eyelid, the feature points for marking the upper eyelid and the feature points for marking the lower eyelid have a corresponding relationship, and the distance value is calculated according to the corresponding relationship.

Step 4: Determine whether the number of read frames reaches the frame number threshold; if not, jump to step 1 to read the next frame image in the face video. After calculating the distance value between the feature points of the upper and lower eyelids of the human face in a frame of image, it is judged whether the number of read frames reaches the frame number threshold, and if it reaches the frame number threshold, continue to step five.

Step 5: Count the fluctuation of the distance value between the upper and lower eyelid feature points of the human face in the read frame image, and judge whether the human face in the human face video is a living body according to the statistical fluctuation. If the camera captures a face video with high-definition photos of the face, since the distance value between the feature points of the upper and lower eyelids in the read frame image will not fluctuate, the face in the collected face video is not a living body. Therefore, the present invention can eliminate the interference to the face recognition result brought by the high-definition photos.

Step 6: If the face in the face video is a living body, perform face recognition on the face in the face video. After verifying that the face in the face video collected by the camera is a living body, face recognition is performed on the face in the face video, thereby improving the accuracy and efficiency of face recognition.

Specifically, in the second step of the present invention, the three-dimensional coordinate axes used to describe the head pose are determined first, and then the head pose is estimated according to the detected face in the current frame image to obtain the head pose vector. Then, calculate the angle value between the head pose vector of the current frame image and the head pose vector of each previous frame image, if there is an angle value greater than the angle threshold, then jump to step 1 and read the person The next frame image in the face video. At the same time, once an included angle value is greater than the angle threshold, a first prompt message is issued to remind the identified person to face the camera directly.

Further, in step 2 of the present invention, the width of the face is also calculated according to the detected face in the current frame image, if the width of the face in the current frame image is smaller than the width threshold, then jump to step 1 and read the The next frame image in the face video. At the same time, once the width of the human face in the current frame image is smaller than the width threshold, a second prompt message is issued to remind the recognized person to move his face closer to the camera. In this way, it can avoid that the position of the face in the face video is located at the edge of the image or exceeds the range of the image, so that the face recognition cannot be performed accurately.

Still further, in the second step of the present invention, the size of the face is also calculated according to the detected face in the current frame image, and the difference between the size of the face in the current frame image and the size of the face in each previous frame image is calculated. value, if there is a difference greater than the face size threshold, then jump to step 1 to read the next frame image in the face video. At the same time, once a difference is greater than the face size threshold, a third prompt message is issued to remind the identified person not to shake his head. In this way, it can be avoided that the face in the face video is at different positions from the camera at different times, which will affect the result of face recognition.

In the living face recognition method based on blink detection of the present invention, the fluctuation in step 5 is represented by a statistical value, and if the statistical value reaches the fluctuation threshold, it is judged that the human face in the human face video is a living body. And, the statistical value is calculated using the following formula:

Among them, L represents the statistical value, s ₁ represents the distance value between the feature points of the upper and lower eyelids on the face of the first frame image, _si represents the distance value between the feature points of the upper and lower eyelids on the face of the i-th frame image, and n is the frame number threshold.

In the live face recognition method based on blink detection of the present invention, before the number of read frames does not exceed the frame number threshold, a fourth prompt message is sent to prompt the person to be recognized to make a specific action, and to identify the number of subsequent frames of images Whether the specific action occurs, if the specific action does not occur, face recognition is stopped. If the camera captures the human face played by the video, since the present invention randomly prompts the identified person to make a specific action in real time, the played video cannot make timely and adaptive adjustments and present the specific action to the camera. Therefore, the present invention can eliminate the interference caused by the video on the face recognition result. Among them, the specific actions include opening and closing one eye, nodding or shaking the head, etc.

Claims (9)

1. A kind of 1. living body faces recognition methods based on blink detection, it is characterised in that comprise the following steps,

   Step 1：Read the image in the face video of camera detection；

   Step 2：The face in current frame image is detected, and is positioned for demarcating the characteristic point of upper eyelid and palpebra inferior respectively；

   Step 3：Calculate and preserve on the face in current frame image distance value between palpebra inferior characteristic point；

   Step 4：Judge to read whether frame number reaches frame number threshold value；If not up to, step 1 is jumped to, reads the face Next two field picture in video；

   Step 5：In the read two field picture of statistics on face between palpebra inferior characteristic point distance value fluctuation situation, and according to The fluctuation situation of statistics, judge whether the face in the face video is live body；

   Step 6：If the face in the face video is live body, recognition of face is carried out to the face in the face video.
2. 2. the living body faces recognition methods based on blink detection as claimed in claim 1, it is characterised in that in step 2,

   It is determined that the 3-D walls and floor for describing head pose, according to the face detected in current frame image, estimates head appearance State, obtain head pose vector；Calculate current frame image head pose vector respectively with above per two field picture head pose Angle value between vector, if there is angle value to be more than angle threshold, jump to step 1.
3. 3. the living body faces recognition methods based on blink detection as claimed in claim 2, it is characterised in that if there is angle value big In angle threshold, then the first prompt message is sent, remind identified person by camera described in face's face.
4. 4. the living body faces recognition methods based on blink detection as claimed in claim 1, it is characterised in that in step 2,

   According to the face detected in current frame image, the width of face is calculated, if the width of face is less than in current frame image Width threshold value, then jump to step 1.
5. 5. the living body faces recognition methods based on blink detection as claimed in claim 4, it is characterised in that if current frame image The width of middle face is less than width threshold value, then sends the second prompt message, reminds identified person by face close to the camera.
6. 6. the living body faces recognition methods based on blink detection as claimed in claim 1, it is characterised in that in step 2,

   According to the face detected in current frame image, the size of face is calculated, and calculates the size of face in current frame image Respectively with the difference of the above face size per two field picture, if there is difference to be more than face size threshold value, step 1 is jumped to.
7. 7. the living body faces recognition methods based on blink detection as claimed in claim 6, it is characterised in that if there is difference to be more than Face size threshold value, then the 3rd prompt message is sent, remind identified person not rock head.
8. 8. the living body faces recognition methods based on blink detection as claimed in claim 1, it is characterised in that in step 5, institute Fluctuation situation is stated by statistical value to characterize, if the statistical value reaches fluctuation threshold, judges that the face in face video is Live body；Also, the statistical value is calculated using below equation：

   <mrow> <mi>L</mi> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <mrow> <mo>(</mo> <msub> <mi>s</mi> <mi>i</mi> </msub> <mo>-</mo> <msub> <mi>s</mi> <mn>1</mn> </msub> <mo>)</mo> </mrow> </mrow>

   Wherein, L represents the statistical value, s₁Represent in the first two field picture on face distance value, s between palpebra inferior characteristic point_iRepresent Distance value, n are frame number threshold value between palpebra inferior characteristic point on face in i-th two field picture.
9. 9. the living body faces recognition methods based on blink detection as claimed in claim 1, it is characterised in that reading frame number not Before the frame number threshold value, the 4th prompting message is sent, prompts identified person to make specific action, and if identifying follow-up Whether occur the specific action in dry two field picture, if not occurring the specific action, stop recognition of face.