CN201084168Y - A face-identifying device - Google Patents

Abstract

The utility model relates to a face recognition device, which comprises an ultrasonic sounder for emitting ultrasonic frequency sweep signals; an ultrasonic receiver for receiving echo signals; a control processor connected with the input end of the ultrasonic generator, the The output end of the ultrasonic receiver is connected to the input end of the recognition result output device, and is interconnected with the memory storing the ultrasonic face information database; the memory is used to store the ultrasonic face information database, and is accessed and controlled by the control processor . The utility model has the advantages of: high spatial resolution can be obtained, rich facial information can be extracted; the influence of background on face recognition can be reduced; 3D models can be distinguished from human faces; Deception; the data volume is reduced, which can improve the recognition speed; it has a high recognition rate; the required ultrasonic face database has the characteristics of small data volume, which is convenient for establishing a large-scale ultrasonic face database.

Description

A face recognition device

technical field

The utility model relates to the technical field of pattern recognition, in particular, the utility model particularly relates to a face recognition device based on pattern recognition.

Background technique

The face recognition device is a device composed of electronic and mechanical components that use the feature information of the face for identity verification. At present, the commonly used face recognition technology is based on the facial features of the face. For the input face image or video stream, it is first judged whether there is a face. If it exists, the position, size and main faces of each face are further given. The location information of the organ. And based on this information, further extract the identity features contained in each face, and compare them with known faces, so as to identify the identity of each face.

However, there are many factors that affect the performance of commonly used face recognition technology, such as: background and hair; translation, scaling, and rotation of the face in the image plane; deflection and pitch of the face outside the image plane; light source position and Changes in intensity; changes in age; changes in expression; influence of attachments (glasses, beard); changes in camera. In addition, attempts to deceive the face recognition system engine through photos, videos, 3D models, etc. have always been the weakness of the face recognition system.

With the development of science and technology, some new methods have made up for the above problems to a certain extent. For example, the technology based on near-infrared can make up for the changes in the position and intensity of the light source to a certain extent. However, this system requires a special infrared sensor. , thereby increasing the cost. This method collects near-infrared face pictures or video streams, and there are also other problems faced by common face recognition technologies.

In order to overcome the influence of posture changes, position changes, and expression changes, a 3D face recognition system is proposed. This method can relieve the pressure caused by the above problems to a certain extent, but recovers 3D shape information from images with a single unknown light source condition. And surface reflectance is a classic visual problem, which is essentially a morbid problem. The establishment of a 3D face model requires the use of special devices, such as laser scanners, etc., and the high cost limits the promotion of this method to a large extent. .

To sum up, there are many difficulties in current face recognition devices, so people expect a new face recognition method or means to solve the current problems and inject vitality into face recognition.

Contents of the invention

The purpose of this utility model is to overcome the deficiencies of the prior art, combine the ultrasonic technology with the pattern recognition technology, propose a high correct rate, independent from the background, not be deceived by photos or videos, and can better distinguish 3D models and Ultrasonic face recognition device based on pattern recognition with small amount of data.

In order to achieve the purpose of the above invention, the face recognition device provided by the utility model includes:

An ultrasonic face recognition device, comprising:

Ultrasonic sounder, used for transmitting ultrasonic frequency sweep signal;

Ultrasonic receiver for receiving echo signals;

The control processor is connected to the input end of the ultrasonic generator, the output end of the ultrasonic receiver and the input end of the recognition result output device, and is interconnected with the memory storing the ultrasonic face information database;

The memory is used to store the ultrasonic face information database, and is accessed and controlled by the control processor.

In the above technical solution, the control processor includes an A/D converter connected to the output terminal of the ultrasonic receiver for digital sampling of the echo signal.

In the above technical solution, the control processor includes: a frequency sweep signal generating unit, an ultrasonic face information database control unit, and an echo processing unit; the frequency sweep signal generating unit generates a frequency sweep signal to drive the ultrasonic sounder; The ultrasonic face database control unit includes operations of reading, writing and modifying the memory storing the database; the echo processing unit processes echo signals.

In the above technical solution, the echo processing unit includes a frequency domain filter circuit, a time domain intercept circuit, a signal demodulation circuit, a feature extraction circuit and a sample training classifier connected in sequence.

In the above technical solution, the sample training classifier is a pattern classifier or a neural network classifier.

In the above technical solution, the pattern classifier may adopt a Bias, GMM, HMM or SVM pattern classifier.

In the above technical solution, the neural network classifier may be replaced by neural network classifiers such as BP, RBF, and SOM networks.

In the above technical solution, there are at least two ultrasonic sounders, and the at least two ultrasonic sounders are installed in different positions, so that the face recognition device can emit ultrasonic waves to the target to be recognized from different directions.

Compared with the prior art, the utility model has the advantages of:

A. The utility model adopts a wide-band linear sweep signal, and through signal compression, a very high spatial resolution can be obtained, and rich facial information can be extracted;

B. Utilizing the fact that the distance between the surrounding reflector and the microphone is generally different from the distance between the face and the microphone, the background echo can be easily separated from the face echo by using digital signal processing, thereby reducing the background Impact on face recognition;

C. Using the different characteristics of the reflection effect of ultrasound on different materials (because the material of the 3D model is different from that of the face, the frequency components of the echo are different), the 3D model can be distinguished from the face;

D. Utilizing the characteristic that the ultrasonic echo is composed of echoes reflected from different parts of the face, it can overcome the deception of photos and videos, thus overcoming a major problem in the current face recognition technology;

E. After the ultrasonic echo is collected, one-dimensional data is obtained through a series of simple processing, and the effective part is selected as the final recognition feature. Compared with the two-dimensional image data in traditional face recognition, it has the obvious advantages of small data volume and easy feature extraction, so it is possible to improve the recognition speed (the data volume of a normal 120*120 two-dimensional image It is 14400bit; if single transmission and single reception is adopted in this patent, the amount of data is about 160bit; if single transmission and double reception are adopted, it is about 240bit;

F. The utility model adopts a plurality of ultrasonic sounders and a plurality of ultrasonic receivers to transmit and receive ultrasonic waves from different angles, which can solve the problem of the influence of changes in facial expressions and postures on face recognition to a certain extent, and has the advantages of High recognition rate;

G. The ultrasonic face database established by the utility model has the characteristics of small amount of data (such as adopting a single-shot and single-receive method, collecting data of 100 different postures, about 16k bit/person), which facilitates the establishment of a large-scale ultrasonic face database .

Description of drawings

Fig. 1 represents the structural block diagram of the ultrasonic face recognition device based on pattern recognition;

FIG. 2 shows a schematic structural diagram of an echo processing unit;

Fig. 3 shows the ultrasonic sounder of an embodiment of the present invention and the schematic diagram of sensor position distribution;

Fig. 4 shows a schematic diagram of the echo signal filtered and intercepted in an embodiment of the present invention;

Fig. 5 shows the echo intensity and the distance relationship figure obtained through signal compression in an embodiment of the utility model;

FIG. 6 shows a partial schematic diagram of the relationship between echo intensity and distance obtained through signal compression in an embodiment of the present invention.

Detailed ways

The basic idea of the utility model is to use ultrasonic waves to recognize human faces. This is because ultrasonic waves have the ability to identify 3D objects, and at the same time, ultrasonic waves have different reflection effects on different materials. The human face has convex and concave surfaces, similar to 3D, and the skin of the human face has both commonality and heterosexuality. The commonality lies in the skin of all human beings, and the heterosexuality lies in the difference in the fineness and elasticity of the skin between different people. Therefore, the 3D-like nature of human faces and the difference in skin between human faces can be used to distinguish different human faces. At the same time, the commonality of the face can also be used to distinguish the face from the 3D model.

During the propagation of the ultrasonic wave, it encounters the reflection of the human face to generate echoes. The echo carries the feature information of the human face, where the intensity of the echo reflects the size of the reflecting surface, and the time delay of the echo relative to the incident wave reflects the distance between the reflecting surface and the sound source. At the same time, the echo also contains the information of human face skin.

The difference between the ultrasonic face recognition method and the ordinary ultrasonic material recognition method is: in the ultrasonic material recognition, the reflected echo information of a reflective surface (material surface) is mainly used to distinguish the material according to the degree of absorption of different frequency components by the material; In face recognition, the reflected echo information of multiple reflective surfaces (different positions of the face) is fully utilized, and the strength of each echo, the sequence relationship between echoes and the degree of absorption of different frequency components by the face are used to distinguish faces. .

Below in conjunction with accompanying drawing and specific embodiment the utility model is described in further detail:

Example 1

The utility model utilizes the above-mentioned characteristics of the reflection of the ultrasonic wave on the human face and the inverse relationship between the spatial resolution of the continuous frequency sweep signal and the frequency band bandwidth, and by means of the pattern recognition method, effectively overcomes many problems of the traditional face recognition, and realizes the recognition of the human face. identify. The specific implementation process of this embodiment includes the following steps:

As shown in Figure 1, make an ultrasonic face recognition device based on pattern recognition according to the utility model, comprising: an ultrasonic sounder 1, which transmits a continuous frequency-sweeping ultrasonic signal, and the ultrasonic signal encounters an obstacle reflection to generate an echo signal; Receiver 2, receives echo signal; A/D converter 3, is connected with the output terminal of described ultrasonic receiver 2, carries out digital sampling to echo signal; Control processor 4, and the input of described ultrasonic generator 1 Terminal, the output end of described A/D converter 3 is connected with the input end of a display or a sounder 9; After the display or sounder 9 receives the recognition result of the control processor 4, it passes through the image or voice displayed in a manner. Described control processor 4 is made up of frequency-sweeping signal generation unit, ultrasonic human face information database control unit and echo processing unit; Described frequency-sweeping signal generation unit produces and drives the linear frequency-sweeping signal of described ultrasonic sounder 1; Ultrasonic face database control unit comprises the operation of reading, writing, and modifying the memory 5 storing the ultrasonic face information database; the echo processing unit processes echo signals, as shown in Figure 2, mainly including frequency The domain filter circuit, the time domain intercept circuit, the signal compression circuit, the feature extraction circuit and the pattern recognition circuit are sequentially connected. When the result of the pattern recognition circuit judges that the tested object is a human face and in the ultrasonic human face database, the echo processing unit passes the recognition result to the display or the sounder 9, and the display or the sounder 9 transmits the result given in the form of images or sounds. The pattern recognition circuit uses a Bayesian classifier. The basic information 7 includes name, gender, date of birth and facial image, which is entered through a keyboard and an ordinary camera.

The ultrasonic sounder in this embodiment can be a piezoelectric thin film transducer or an array sounder composed of piezoelectric thin film transducers, or an array sounder composed of piezoelectric ceramic transducers or piezoelectric ceramic transducers, Either an electrostatic transducer or an array sounder composed of electrostatic transducers, or an ultrasonic sounder. As shown in Figure 3, in this embodiment, two ultrasonic sounders and two receivers are used, and they are divided into two groups, which are Group A (including electrostatic transducer 1A, sensor 2A) and Group B ( Including electrostatic transducer 1B, sensor 2B), the distance between each group of devices and the face is 1m, where B group is in front of the face, and A group is 45 degrees above the tip of the nose. Considering the spatial resolution of the linear sweep signal, the intensity of the echo, the real-time processing speed of the echo signal, the frequency response curve of the device and the energy consumption, etc., the transmitting frequency of the ultrasonic transmitter 1 is 25KHz--100KHz, and the length It is a frequency sweep signal of 100ms, and the transmission interval is 150ms. The sampling frequency of the echo signal by the A/D converter 3 is 500KHz.

All parts of the above circuits, unless otherwise specified, are conventional products or conventional circuits well known to those skilled in the art and are connected in a conventional manner.

Due to the need to set up an ultrasonic database and perform pattern recognition, the ultrasonic face recognition method based on pattern recognition of the present embodiment includes the following steps:

1) Collect the basic information of the object to be added to the face recognition database, including name, gender, date of birth and the corresponding frontal face image.

2) Measure multiple sets of data for each measurement object, extract features, and establish an ultrasonic face database;

Considering the directivity of the ultrasonic sounder and ensuring the relative stability of the echo to the same object, first set a reference point, which is 1m away from the microphone, and both the ultrasonic sounder and the microphone point to the reference point. At the same time, in order to reduce the influence of the surrounding objects on the echo as much as possible, the device should be placed in an open room, and the distance between the reference point and the ground is 1.14m. During the sample collection process, the test subject's nose tip is close to the reference point and looks forward, and may be stationary or have slight movements during the entire measurement process. Send the pulse signal and record the echo signal. For each measurement object, ultrasonic waves are emitted by the ultrasonic generator 1B in the horizontal direction and 100 echo signals, ie, 100 samples, are recorded by the microphone 2B in the horizontal direction. Then, the ultrasonic generator 1A at an oblique direction of 45 degrees was used to emit ultrasonic waves, and the microphone 2A at an oblique direction of 45 degrees was used to receive and record 100 echo signals. A total of 200 samples were collected for each measurement object. Samples are collected by the above method for each object who wants to be added to the ultrasonic face recognition database. Each sample is processed separately, and the processing process is shown in Figure 2.

First, because the frequency range of the echo signal is 25kHz-100kHz, a 20k-110kHz band-pass filter is used to filter the received echo signal, and then the data of the first 100ms of each echo signal is intercepted, and the intercepted echo signal The wave signal is shown in Figure 4.

The second is signal compression. Because the duration of the frequency sweep signal is 100ms, the reflected echoes at different positions are mixed together, so it is difficult to directly extract the echo information of the face from the intercepted echo signal, so it is necessary to obtain high resolution through signal compression echo distance information. FM signal compression can be realized with a matched filter, or by mixing the transmitted signal with the echo signal and doing Fourier transform. This example chooses the latter method: first mix the transmitted signal with the original signal, and then do fast Fourier transform. Fig. 5 shows the relationship between echo strength and distance obtained through signal compression. Figure 6 is a partial enlargement of Figure 5. According to the known condition that the distance between the face and the microphone is 1m, it can be clearly seen that the echo of the face appears in the area of 1m-1.5m, and has a high SNR.

The next step is to extract the feature vectors. Feature extraction methods include selecting time-domain envelope, frequency-domain energy, coefficients of signals such as discrete cosine transform or wavelet transform, time-domain envelope or frequency-domain energy or cepstral coefficients after signal demodulation, or using modern spectrum estimation The obtained coefficients or wavelet-transformed coefficients and the compressed signal are extracted from the above-mentioned series of features as part or all of the features as feature vectors. In the embodiment, two methods are used to extract features. The first is to extract 15 eigenvalues as feature vectors according to McKerrow (McKerrow, P.J.and Yoong, K.K.Face classification with ultrasonic sensing, Proceedings TOWARDS Autonomous Robotic Systems TAROS-06, September4-6, pp 111-117.) The second method is to directly intercept the echo data of the face from the compressed echo signal, and use this as a feature vector.

The details of the second method are as follows: First, the noise 1 (as shown in Figure 6, composed of crosstalk noise and direct sound) is removed; secondly, the threshold method is used to find the starting point of the face echo; finally, according to the average geometry of the face, Dimensions intercept fixed-length data. (This threshold method is a general threshold method. As can be seen from Figure 6, after the noise 1 is removed, the remaining strong signal is the face echo signal, so the threshold method can be used to find the starting point of the face echo. This In the embodiment, the data of 80 points are intercepted from the starting point as feature vectors)

After the feature vector is extracted, the classifier is selected. In this example, the Bias classifier is selected. In order to verify its performance, 60% of each type of samples in the database are selected to form a training sample set; the remaining 40% are used as a test sample set. Table 1 lists the classification results of the Bias classifier, and compares the two feature extraction methods. The recognition results are shown in Table 1.

Table 1

Feature extraction method   Ultrasonic detection direction Recognition rate based on Bayesian method Extraction method proposed by McKerrow In front of 67.0% 45 degrees obliquely upward 82.4% Straight ahead + 45 degrees obliquely above 96.7% The second feature extraction method In front of 97.5758% 45 degrees obliquely upward 91.2121% Straight ahead + 45 degrees obliquely above 99.3939%

In this form, "directly in front + 45 degrees obliquely above" refers to comprehensively processing the received data of two sets of ultrasonic microphones before obtaining the recognition result.

It can be seen from Table 1 that the second feature extraction method is significantly better than the feature extraction method proposed by McKerrow, and it can be clearly seen that the recognition rate is higher than that of a single microphone by combining the features extracted by multiple microphones. Rate.

3) After the classifier is selected, the actual detection is carried out; the ultrasonic sounder 1 emits a continuous frequency-sweeping ultrasonic signal, and receives the echo signal with the ultrasonic receiver 2;

4) The control processor 4 performs preprocessing on the echo signal (including frequency domain filtering, time domain interception and signal compression on the echo signal);

5) the control processor 4 extracts a feature vector to the echo signal;

6) Use the pattern classifier to judge whether the echo signal is a human face echo signal (a threshold is set in this embodiment, if it exceeds the threshold, it is considered to be a human face echo signal), if so, then perform 7), otherwise return step 3);

7) Utilize the Bias pattern classifier to carry out identification comparison in the established ultrasonic face information database, if the target to be identified exists in the database, then perform step 8), otherwise return to step 3);

8) Give the recognition result on the display or sounder in the form of image or sound.

9) After completing the relevant processing, return to the above step 3) and start executing sequentially downwards.

Although the Bias pattern classifier is used as a sample training classifier in this embodiment, the classifier can be replaced by other types of pattern classifiers such as GMM, HMM, SVM, etc., or can be classified by neural networks such as BP, RBF, and SOM networks. device replacement. In addition, the utility model can also provide different types of recognition results according to requirements, such as identification (identification), that is, to judge which face a given face ultrasonic echo belongs to; or verification (verification), that is, to judge a given face. Whether the face ultrasound echo belongs to a certain person.

In addition, the identification method in this embodiment is only an example, and the method can be implemented by different hardware or software, and is not limited to the identification device in this embodiment. As the utility model also can adopt ultrasonic face recognition device in a broader sense, as a kind of ultrasonic face recognition device, this device comprises: ultrasonic sounder, is used for transmitting ultrasonic signal; Ultrasonic receiver, is used for receiving echo signal; The extraction module is used to extract the eigenvector of the echo signal; the ultrasonic face information database is used to store the basic information of registered members and the eigenvector of the pre-acquired ultrasonic echo signal; the identity recognition module is used to extract the ultrasonic echo signal The eigenvectors are compared and analyzed to obtain the recognition results.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present utility model rather than limit them. Although the present utility model has been described in detail with reference to the embodiments, those skilled in the art should understand that any modification or equivalent replacement of the technical solution of the present utility model does not depart from the spirit and scope of the technical solution of the present utility model, and all of them should cover In the scope of the claims of the present utility model.

Claims (8)

1. A face recognition device, comprising: Ultrasonic sounder, used for transmitting ultrasonic frequency sweep signal; Ultrasonic receiver for receiving echo signals; The control processor is connected to the input end of the ultrasonic generator, the output end of the ultrasonic receiver and the input end of the recognition result output device, and is interconnected with the memory storing the ultrasonic face information database; The memory is used to store the ultrasonic face information database, and is accessed and controlled by the control processor. 2. ultrasonic face recognition device according to claim 1, is characterized in that, described control processor comprises A/D converter, and this A/D converter is connected with the output end of described ultrasonic receiver, for The echo signal is digitally sampled. 3. The ultrasonic face recognition device according to claim 1 or 2, wherein the control processor comprises: a frequency sweep signal generation unit, an ultrasonic face information database control unit and an echo processing unit; The frequency signal generation unit produces the frequency sweep signal that drives the ultrasonic sounder; the ultrasonic human face database control unit includes the operations of reading, writing and modifying the memory of the storage database; the echo processing unit processes the echo signal . 4. ultrasonic face recognition device according to claim 3, is characterized in that, described echo processing unit comprises frequency domain filter circuit, time domain interception circuit, signal demodulation circuit, feature extraction circuit and sample training circuit connected sequentially Classifier. 5. The ultrasonic face recognition device according to claim 4, wherein the sample training classifier is a pattern classifier or a neural network classifier. 6. The ultrasonic face recognition device according to claim 5, wherein the pattern classifier can adopt Bias, GMM, HMM or SVM pattern classifiers. 7. ultrasonic face recognition device according to claim 5, is characterized in that, described neural network classifier can adopt neural network classifiers such as BP, RBF, SOM network to replace. 8. The ultrasonic face recognition device according to claim 1, wherein the ultrasonic sounder is at least two, and the at least two ultrasonic sounders are installed in different positions, so that the face recognition device can be Ultrasonic waves are emitted in different directions to the target to be identified.

Images