TWI731461B - Identification method of real face and identification device using the same - Google Patents

Abstract

An identification method of real face and an identification device using the same are provided. The method includes: obtaining a face image of a target face; obtaining depth information of a target region in the face image; analyzing the depth information to obtain at least one characteristic value related to a quadratic curve which reflects a depth distribution status of the target region; determining whether the at least one characteristic value meets a default condition; if the at least one characteristic value meets the default condition, determining that the target face is a face in a photograph; and if the at least one characteristic value does not meet the default condition, determining that the target face is a real face.

Description

Real face recognition method and real face recognition device

The present invention relates to an image recognition technology, and particularly relates to a real face recognition method and real face recognition device.

With the advancement of technology, the use of face recognition technology for login authentication of electronic devices has become more and more popular. As long as the user presents his face in front of the lens of the electronic device, he can log in directly through the face verification mechanism. However, some unscrupulous elements may use photos downloaded from the Internet or photos of real users to scan their faces in an attempt to log in to other people's electronic devices. Therefore, how to improve the recognition efficiency of real human faces in the process of login verification is actually one of the research topics devoted to those skilled in the art.

The invention provides a real face recognition method and a real face recognition device, which can effectively improve the recognition efficiency of recognizing the face in front of the lens as a real face or a face in a photo.

An embodiment of the present invention provides a real face recognition method, which includes: obtaining a face image of a target face; obtaining depth information of a target area in the face image; analyzing the depth information to obtain and secondary At least one characteristic value related to the curve, wherein the quadratic curve reflects the depth distribution state of the target area; judging whether the at least one characteristic value meets a preset condition; if the at least one characteristic value meets the preset condition , Determining that the target human face is a human face in the photo; and if the at least one feature value does not meet the preset condition, determining that the target human face is a real human face.

The embodiment of the present invention further provides a real face recognition device, which includes a depth camera and a processor. The processor is coupled to the depth camera. The processor is used for obtaining the facial image of the target person's face by the depth camera. The processor is further used to obtain the depth information of the target area in the face image by the depth camera. The processor is further configured to analyze the depth information to obtain at least one characteristic value related to the quadratic curve. The quadratic curve reflects the depth distribution state of the target area. The processor is further configured to determine whether the at least one characteristic value meets a preset condition. If the at least one characteristic value meets the preset condition, the processor is further configured to determine that the target human face is a human face in the photo. If the at least one characteristic value does not meet the preset condition, the processor is further configured to determine that the target face is a real face.

Based on the above, when obtaining the facial image of the target person's face, the depth information of the target area in the facial image can be obtained at the same time. By analyzing the depth information, at least one characteristic value related to a quadratic curve can be obtained, and the quadratic curve can reflect the depth distribution state of the target area. Then, by determining whether the at least one characteristic value meets a preset condition, the target face can be effectively identified as a real face or a face in a photo. In this way, the recognition efficiency of recognizing the human face in front of the lens as a real human face or a human face in a photo can be effectively improved.

FIG. 1 is a schematic diagram of an electronic device according to an embodiment of the invention. Please refer to Figure 1, the electronic device (also known as a real face recognition device) 10 can be a notebook computer, a desktop computer, a tablet computer, a smart phone, a game console, or an information service station (Kiosk), etc. The electronic device of the depth camera and the processor, and the type of the electronic device 10 is not limited to the above.

The electronic device 10 includes a depth camera 11, a storage device 12 and a processor 13. The depth camera 11 can be used to shoot images with depth information. For example, when there is a human face (also referred to as a target human face) in front of the lens of the depth camera 11, the captured image may be a human face image, and at least one pixel in the human face image may carry depth information of a corresponding position . For example, the depth camera 11 may include at least one lens, at least one photosensitive element, and/or at least one depth sensor to accomplish the above-mentioned functions.

The storage device 12 is used to store data. For example, the storage device 12 may include a non-volatile memory module and a volatile memory module. The non-volatile memory module can be used to store data non-volatile. For example, the non-volatile memory module may include read only memory (ROM), solid state drive (SSD) and/or traditional hard drive (HDD). The volatile memory module can be used to temporarily store data. For example, the volatile memory module may include dynamic random access memory (RAM). In addition, the non-volatile memory module and/or the volatile memory module may also include other types of storage media, which is not limited by the present invention.

In one embodiment, the storage device 12 stores a deep learning model 101. The deep learning model 101 is also called an artificial intelligence model. The deep learning model 101 can have a neural network-like architecture and can be used for image recognition. In an embodiment, the deep learning model 101 can be used to recognize human faces. In an embodiment, the deep learning model 101 can be used to identify at least one facial organ (for example, eyes (or pupils), nose, mouth, and/or ears) in a human face. In addition, the deep learning model 101 can gradually improve the accuracy of image recognition through training. In an embodiment, the deep learning model 101 can also be implemented as a hardware circuit (such as a chip), which is not limited by the present invention.

The processor 13 is coupled to the depth camera 11 and the storage device 12. The processor 13 may be a central processing unit (CPU), a graphics processing unit (GPU), or other programmable general-purpose or special-purpose microprocessors, digital signal processors (DSP), and programmable Controllers, Application Specific Integrated Circuits (ASIC), Programmable Logic Device (PLD) or other similar devices or a combination of these devices. The processor 13 can control the whole or part of the operation of the electronic device 10. For example, the processor 13 may run the deep learning model 101 to perform image recognition.

In one embodiment, the electronic device 10 further includes at least one input/output interface to receive signals or output signals. For example, the input/output interface can include the screen, touch screen, touchpad, mouse, keyboard, physical buttons, speakers, microphone, wired network card and/or wireless network card, and the type of input/output interface is not Limited to this.

When there is a human face in front of the lens of the depth camera 11 (ie, the target human face), the processor 13 can obtain the facial image of the target human face through the depth camera 11. The processor 13 can also obtain the depth information of a specific area (also referred to as a target area) in the face image through the depth camera 11. It should be noted that the present invention does not limit the number and/or shape of target regions in a single face image. The processor 13 can analyze the depth information to obtain at least one characteristic value related to a certain quadratic curve. Wherein, the quadratic curve may reflect the depth distribution state of the target area.

After obtaining the characteristic value, the processor 13 may determine whether the characteristic value meets a preset condition. If the feature value meets a preset condition, the processor 13 may determine that the target face is a face in the photo. In addition, if the feature value does not meet the preset condition, the processor 13 may determine that the target human face is a real human face (that is, not a human face in the photo). For example, if a user presents a face in a photo displayed on the screen of a mobile phone (or a face in a paper photo) in front of the lens of the depth camera 11, the processor 13 can determine the face in front of the current lens according to the above operation It is the face in the photo, not the real face. In this way, it is possible to reduce erroneous actions performed because the human face in the photo is misjudged as a real human face.

In an embodiment, the processor 13 may analyze the facial image by the deep learning model 101 to obtain the position of at least one facial organ of the target person's face. Then, the processor 13 may determine the target area according to the position of the at least one facial organ.

FIG. 2 is a schematic diagram of a human face image according to an embodiment of the invention. Please refer to FIG. 1 and FIG. 2, a target face 22 is presented in the face image 21. The processor 13 can use the deep learning model 101 to identify facial organs in the target face 22, such as eyes, nose, mouth, and/or ears.

In an embodiment, the processor 13 may set the reference points 201 to 205 according to the positions of at least part of the recognized facial organs. For example, the reference point 201 may be set at the location of the left eye in the target face 22, the reference point 202 may be set at the location of the right eye in the target face 22, and the reference point 203 may be set at the location of the target face 22. The position of the nose and the reference point 204 may be set on the left side of the mouth in the target face 22 and the reference point 205 may be set on the right side of the mouth in the target face 22. It should be noted that in other embodiments, the reference points 201 to 205 can also be set at other positions in the target face 22 and/or the number of reference points 201 to 205 can also be more or less. The present invention does not Be restricted.

FIG. 3 is a schematic diagram of a target area drawn according to an embodiment of the invention. 1 to 3, in an embodiment, at least one of the line segments 301 to 306 can be determined according to the set reference points 201 to 205. For example, the processor 13 may set the line segment 301 as a connection line between the midpoint between the reference points 201 and 202 and the midpoint between the reference points 204 and 205. For example, the processor 13 may set the line segment 302 as the connection between the reference points 201 and 205. For example, the processor 13 may set the line segment 303 as the connection between the reference points 202 and 204. For example, the processor 13 may set the line segment 304 as a line between the midpoint between the reference points 201 and 204 and the midpoint between the reference points 202 and 205. For example, the processor 13 may set the line segment 305 as the connection line between the reference points 201 and 204. For example, the processor 13 may set the line segment 306 as the connection line between the reference points 202 and 205. The path taken by at least one of the line segments 301 to 306 may be determined as the target area. In other words, the target area may include pixel points (or pixel positions) passed or covered by at least one of the line segments 301 to 306. In addition, at least one pixel in the target area can be regarded as a sampling point. Each sampling point may have a depth information (for example, a depth value) to reflect the depth of the location of the sampling point.

In an embodiment, the target area may be divided into at least one first area and at least one second area. The first area includes the position of the nose of the target face. For example, the path taken by the line segments 301 to 304 in FIG. 3 can be regarded as the first area. The second area does not include the position of the nose of the target face. For example, the paths traversed by the line segments 305 and 306 in FIG. 3 can be regarded as the second area. The processor 13 can obtain at least one characteristic value by analyzing the depth information of the first area and/or the second area.

FIG. 4 is a schematic diagram of a curve reflecting the depth distribution state drawn according to an embodiment of the present invention. Please refer to Figures 1 to 4, assuming that the sampling points 1~100, 101~200, 201~300, 301~400, 401~500, and 501~600 are respectively located in the target area passed by the line segments 301~306. The depth values corresponding to sampling points 1~100, 101~200, 201~300, 301~400, 401~500, and 501~600 can be represented by curves 401~406. In other words, the curve 401 can reflect the depth distribution state of multiple sampling points 1 to 100 on the path passed by the line segment 301, and the curve 406 can reflect the depth distribution state of multiple sampling points 501 to 600 along the path passed by the line segment 306 ,So on and so forth.

It should be noted that in the embodiment of FIG. 4, it is assumed that the face image 21 in FIG. 2 is obtained by shooting a real face (that is, the target face 22 is a real face). Therefore, the path traversed by the line segments 301-304 includes the position of the nose in the target face 22 (the depth value of the position of the nose is small), so the curves 401-404 will be curved like a quadratic curve, and The opening direction of the curve 401 is upward. In addition, since the paths traversed by the line segments 305 and 306 do not include the position of the nose in the target face 22 (that is, the line segments 305 and 306 pass through the cheeks of the real human face, and the depth changes are small), so the curves 405 and 406 It will be relatively flat.

However, the curves 401 to 406 in FIG. 4 are only exemplary, and are not used to limit the present invention. In other embodiments, the depth value corresponding to any one of the curves 401 to 406 may also be different and/or the number of sampling points corresponding to any one of the curves 401 to 406 may also be different, which is not included in the present invention. limit. Or, in another embodiment of FIG. 4, if the face image 21 in FIG. 2 is obtained by taking a face in the photo (that is, the target face 22 is not a real face), the curves 401~406 The depth distribution state will also be significantly different.

In an embodiment, the processor 13 may use a quadratic curve to simulate or approximate at least one of the curves 401 to 406 to obtain a characteristic value related to at least one of the curves 401 to 406. In an embodiment, the characteristic value includes a first characteristic value and a second characteristic value. The first characteristic value reflects the opening direction of the quadratic curve and the degree of curvature of the quadratic curve. The second characteristic value reflects the position (or relative position) of the extreme value of the quadratic curve in the quadratic curve.

FIG. 5 is a schematic diagram of a quadratic curve drawn according to an embodiment of the present invention. 1 to 5, taking the curve 401 as an example, the processor 13 can use the quadratic curve 501 to simulate or approximate the curve 401. The quadratic curve 501 can be described by the following equation (1.1).

y=a(xb) ² +c (1.1)

In equation (1.1), the parameter y represents the depth value of the conic 501 in the vertical axis direction, the parameter x represents the sampling point of the conic 501 in the horizontal axis direction, and the parameter a reflects the opening direction of the conic 501 and the quadratic The degree of curvature of the curve 501, the parameter b reflects the position of the extreme value of the quadratic curve 501 in the quadratic curve 501, and the parameter c is a constant. In the embodiment of FIG. 5, a positive value of parameter a can reflect that the opening direction of the quadratic curve 501 is upward, the value of parameter a is positively correlated with the degree of curvature of the quadratic curve 501, and the value of parameter b can reflect the conic curve. The smallest depth value in 501 occurs at the b-th sampling point.

In an embodiment, the processor 13 may obtain the first characteristic value related to the curve 401 (or the quadratic curve 501) according to the parameter a, and obtain the first characteristic value related to the curve 401 (or the quadratic curve 501) according to the parameter b. Two eigenvalues. In an embodiment, the processor 13 can also obtain the characteristic value related to any one of the curves 402 to 406 in FIG. 4 in the same manner, and details are not repeated here. The processor 13 may determine that the target face 22 in FIG. 2 is a real face or a face in a photo according to the first feature value and the second feature value.

In an embodiment, the processor 13 may determine the parameter a as the first characteristic value. In an embodiment, the processor 13 may divide the parameter b by the total number of sampling points corresponding to the curve 401 (for example, 100) and determine the calculation result as the second characteristic value. Therefore, in an embodiment, the first characteristic value may be a parameter a, and the second characteristic value may be a parameter p. Among them, the parameter p=b/(the total number of sampling points=100). It should be noted that in other embodiments, the first characteristic value and the second characteristic value may also be obtained by performing other logical operations according to the parameters a and b, respectively, and the present invention is not limited.

In an embodiment, the processor 13 may determine whether the first characteristic value (represented by the parameter C1) and/or the second characteristic value (represented by the parameter C2) meets a preset condition. In an embodiment, the preset conditions corresponding to different target regions (ie, line segments) can be represented in Table 1 below. Table 1 Line segment (target area) First eigenvalue C1 Second eigenvalue C2 301 C1> V1 ︱C2- V4︱> V5 302 C1> V1 ︱C2- V4︱> V5 303 C1> V1 ︱C2- V4︱> V5 304 C1>V2 ︱C2- V4︱> V6 305 C1> V3 306 C1> V3

In one embodiment, the parameter V1 can be 0.015, the parameter V2 can be 0.03, the parameter V3 can be 0.02, the parameter V4 can be 0.5, the parameter V5 can be 0.3 and/or the parameter V6 can be 0.2. However, in another embodiment, the parameters V1 to V6 may also be other values, which is not limited by the present invention. In an embodiment, the processor 13 may use a plurality of face images for training to train the deep learning model 101. According to the training result, the processor 13 can generalize the parameters V1 to V6 that can be used to distinguish between the human face in the photo and the real human face.

In an embodiment, as long as any one of the conditions listed in Table 1 is met, it can be determined that the target face 22 in FIG. 2 is a face in the photo. Or, in an embodiment, only when at least two conditions listed in Table 1 are met, can the target face 22 in FIG. 2 be determined to be the face in the photo. Or, in an embodiment, only when all the conditions listed in Table 1 are met, can the target face 22 in FIG. 2 be determined to be the face in the photo.

For example, in one embodiment, assuming that the first characteristic value C1 related to the curve 401 meets the condition of C1>V1 corresponding to the line segment 301 in Table 1, then in response to the satisfaction of this condition, the processor 13 may determine The target human face 22 is a human face in the photo, not a real human face. Or, in an embodiment, assuming that the first characteristic value C1 related to the curve 405 meets the condition of C1>V3 corresponding to the line segment 305 in Table 1, then in response to the satisfaction of this condition, the processor 13 may determine The target human face 22 is a human face in the photo, not a real human face. Or, in an embodiment, assuming that the second characteristic value C2 related to the curve 402 meets the condition of ︱C2-V4︱>V5 corresponding to the line segment 302 in Table 1, then in response to the satisfaction of this condition, the processor 13 may It is determined that the target human face 22 in FIG. 2 is a human face in the photo, rather than a real human face.

In an embodiment, if it is determined that the target human face is a real human face, the processor 13 may allow subsequent operations related to face verification or facial image registration to continue. For example, after determining that the target face 22 in FIG. 2 is a real face, the processor 13 may allow the use of the face image 21 to perform face verification and/or face image registration. Conversely, if it is determined that the target human face is a human face in the photo (ie, an unreal human face), the processor 13 may stop continuing to perform subsequent operations related to face verification or facial image registration. In this way, it is possible to reduce erroneous actions performed because the human face in the photo is misjudged as a real human face.

Fig. 6 is a flowchart of a method for recognizing a real face according to an embodiment of the present invention. Referring to FIG. 6, in step S601, a facial image of a target person's face is obtained. In step S602, the depth information of the target area in the face image is obtained. In step S603, the depth information is analyzed to obtain at least one characteristic value related to the quadratic curve. The quadratic curve reflects the depth distribution state of the target area. In step S604, it is determined whether the at least one characteristic value meets a preset condition. If the at least one feature value meets the preset condition, in step S605, it is determined that the target face is a face in the photo. However, if none of the at least one feature value meets the preset condition, in step S606, it is determined that the target face is a real face.

However, each step in FIG. 6 has been described in detail as above, and will not be repeated here. It is worth noting that each step in FIG. 6 can be implemented as multiple program codes or circuits, and the present invention is not limited. In addition, the method in FIG. 6 can be used in conjunction with the above exemplary embodiments, or can be used alone, and the present invention is not limited.

In summary, the embodiments of the present invention can effectively filter the faces in the photos in front of the camera and/or recognize the real faces in front of the camera, thereby reducing the misjudgment of the faces in the photos as real faces. The misoperation of the implementation.

10: Electronic device 11: Depth camera 12: Storage device 13: Processor 101: Deep learning model 21: Face image 22: Target face 201~205: Reference point 301~306: line segment 401~406: Curve 501: Conic S601~S606: Steps

FIG. 1 is a schematic diagram of an electronic device according to an embodiment of the invention. FIG. 2 is a schematic diagram of a human face image according to an embodiment of the invention. FIG. 3 is a schematic diagram of a target area drawn according to an embodiment of the invention. FIG. 4 is a schematic diagram of a curve reflecting the depth distribution state drawn according to an embodiment of the present invention. FIG. 5 is a schematic diagram of a quadratic curve drawn according to an embodiment of the present invention. Fig. 6 is a flowchart of a method for recognizing a real face according to an embodiment of the present invention.

S601~S606: Steps

Claims (8)

A method for recognizing a real human face includes: obtaining a facial image of a target human face; obtaining depth information of a target region in the facial image; analyzing the depth information to obtain at least one related to a quadratic curve Feature value, wherein the quadratic curve reflects a depth distribution state of the target area; determine whether the at least one feature value meets a preset condition; if the at least one feature value meets the preset condition, determine that the target face is a photo And if the at least one feature value does not meet the preset condition, determining that the target face is a real face, wherein the at least one feature value includes a first feature value and a second feature value, and the analysis The step of obtaining the at least one characteristic value related to the quadratic curve by the depth information includes: ^{describing the quadratic curve by the equation y=a(xb) 2} +c; obtaining the second curve according to the parameter a in the equation A characteristic value; and obtaining the second characteristic value according to the parameter b in the equation. The real face recognition method according to claim 1, wherein the target area includes at least one first area and at least one second area, and the at least one first area includes the position of the nose of the target face , And the at least one second area does not include the position of the nose of the target face. According to the method for recognizing a real face as described in item 1 of the scope of patent application, the first feature value reflects an opening direction of the conic and a degree of curvature of the conic, and the second feature value reflects the The position of an extreme value of the quadratic curve in the quadratic curve. The real face recognition method described in item 1 of the scope of the patent application further includes: analyzing the face image by a deep learning model to obtain the position of at least one facial organ of the target face; and according to the at least The position of a facial organ determines the target area. A real face recognition device includes: a depth camera; and a processor coupled to the depth camera, wherein the processor is used to obtain a facial image of a target face by the depth camera, the processor It is further used to obtain depth information of a target area in the face image by the depth camera, and the processor is further used to analyze the depth information to obtain at least one characteristic value related to a quadratic curve, wherein the quadratic The curve reflects a depth distribution state of the target area. The processor is further used to determine whether the at least one characteristic value meets a predetermined condition. If the at least one characteristic value meets the predetermined condition, the processor is further used to determine the The target face is a face in the photo, and if the at least one feature value does not meet the preset condition, the processor is further used to determine that the target face is a real face, wherein the at least one feature value includes a first Eigenvalue and a second eigenvalue, wherein the operation of the processor to analyze the depth information to obtain the at least one eigenvalue related to the quadratic curve includes: describing the two ^{eigenvalues by the equation y=a(xb) 2 +c} The secondary curve; the first characteristic value is obtained according to the parameter a in the equation; and the second characteristic value is obtained according to the parameter b in the equation. The real face recognition device according to item 5 of the scope of patent application, wherein the target area includes at least one first area and at least one second area, and the at least one first area includes the position of the nose of the target face , And the at least one second area does not include the position of the nose of the target face. For the real face recognition device described in item 5 of the scope of patent application, the first feature value reflects an opening direction of the conic and a degree of curvature of the conic, and the second feature value reflects the The position of an extreme value of the quadratic curve in the quadratic curve. For the real face recognition device described in item 5 of the scope of patent application, the processor is further used to analyze the face image by a deep learning model to obtain the position of at least one facial organ of the target face, And the processor is further used for determining the target area according to the position of the at least one facial organ.

Images