JP5033014B2 - Face recognition device - Google Patents

Description

  The present invention relates to a face authentication apparatus that authenticates a person using a pre-registered template and a face image captured for authentication, and more particularly to a face authentication apparatus that uses a light modulation image as a face image for authentication. It is.

The present applicant has already proposed a face authentication apparatus that can ensure high authentication accuracy even under outdoor sunlight by acquiring a stable grayscale image that is not affected by ambient light (for example, sunlight). (For example, refer to Patent Document 1). In this conventional apparatus, a person's face is irradiated with modulated light obtained by high-speed modulation of infrared light emitted from a light emitting diode, and the reflected light component of the modulated light and the reflected light component of ambient light received by an image pickup device made up of a CCD are obtained. A face image (light modulated image) with only modulated light is obtained by distinction. Further, in the conventional apparatus, a feature amount of a person's face is extracted from the light modulation image, and the feature amount is compared with a template registered in advance, thereby authenticating the person's face corresponding to the template.
JP 2006-155422 A

  However, as described above, the reflected light component of the modulated light emitted from the light emitting diode includes, for example, the reflected light component that is specularly reflected by the spectacles, and thus the facial organ ( In particular, it is very difficult to extract the feature amount of the eye), and as a result, there is a possibility that the authentication accuracy of the person wearing the glasses is significantly lowered.

  The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a face authentication apparatus that can suppress a decrease in authentication accuracy even when a reflected light component that is specularly reflected by glasses or the like is present. is there.

In order to achieve the above object, the invention according to claim 1 outputs modulated light irradiation means for irradiating the target space with modulated light, an image sensor for receiving light incident from the target space, and each pixel of the image sensor. A light modulation image generation unit that generates a light modulation image using only the reflected light component of the modulated light as a pixel value by processing the processed signal, and a target space based on the light modulation image generated by the light modulation image generation unit A feature value of a person's face existing in the image, and comparing the feature value with a pre-registered template to authenticate the person's face corresponding to the template; Specular reflection detection means for detecting a specular reflection portion that is equal to or greater than the threshold value, the modulated light irradiating means has a variable pattern of modulated light that irradiates the target space, and the authentication means is the same person. Different from each other Of a plurality of light modulation image captured by the pattern of the dimmer, with respect to the light modulation image size of the specular reflection portion to be detected by the specular reflection detecting means becomes minimum, it is detected by the regular reflection detecting means The authentication processing is performed so as to suppress the influence of the regular reflection portion.

According to the first aspect of the present invention, the regular reflection detection means detects the regular reflection portion where the pixel value is equal to or greater than the predetermined threshold in the light modulation image, and the authentication means authenticates so as to suppress the influence of the regular reflection portion. Since the process is performed, it is possible to suppress a decrease in authentication accuracy even when there is a reflected light component that is specularly reflected on glasses or the like. Furthermore, it is possible to easily and accurately extract a human face feature amount by performing an authentication process on a light modulation image in which the size of the regular reflection portion detected by the regular reflection detection means is minimized, As a result, authentication accuracy can be improved.

In order to achieve the above object , a second aspect of the invention provides modulated light irradiation means for irradiating a target space with modulated light, an image sensor for receiving light incident from the target space, and an output from each pixel of the image sensor. A light modulation image generation unit that generates a light modulation image using only the reflected light component of the modulated light as a pixel value by processing the processed signal, and a target space based on the light modulation image generated by the light modulation image generation unit A feature value of a person's face existing in the image, and comparing the feature value with a pre-registered template to authenticate the person's face corresponding to the template; Specular reflection detection means for detecting a specular reflection portion that is equal to or greater than the threshold value, the modulated light irradiating means has a variable pattern of modulated light that irradiates the target space, and the authentication means is the same person. Different from each other Of a plurality of light modulation image captured by the pattern of the dimmer, the pixel value of the region including the face of a person with respect to the largest becomes the light modulation image, the specular reflection portion to be detected by the regular reflection detecting means The authentication process is performed so as to suppress the influence .

According to the invention of claim 2, the regular reflection detection means detects the regular reflection portion where the pixel value is equal to or greater than the predetermined threshold in the light modulation image, and the authentication means authenticates so as to suppress the influence of the regular reflection portion. Since the process is performed, it is possible to suppress a decrease in authentication accuracy even when there is a reflected light component that is specularly reflected on glasses or the like. Further, the authentication accuracy is improved by performing the authentication process by selecting the light modulation image having the highest signal-to-noise ratio from the plurality of light modulation images.

In order to achieve the above object , a third aspect of the present invention provides a modulated light irradiating unit that irradiates a target space with modulated light, an image sensor that receives light incident from the target space, and an output from each pixel of the image sensor. A light modulation image generation unit that generates a light modulation image using only the reflected light component of the modulated light as a pixel value by processing the processed signal, and a target space based on the light modulation image generated by the light modulation image generation unit A feature value of a person's face existing in the image, and comparing the feature value with a pre-registered template to authenticate the person's face corresponding to the template; Specular reflection detection means for detecting a specular reflection portion that is equal to or greater than the threshold value, the modulated light irradiating means has a variable pattern of modulated light that irradiates the target space, and the authentication means is the same person. Different from each other When the average value of the pixel values of the image that does not include the reflected light component of the modulated light output from the imaging device is less than or equal to a predetermined value among the plurality of light modulation images captured with the dimming pattern, regular reflection The light modulation image that minimizes the size of the specular reflection portion detected by the detection means is selected , and authentication processing is performed so as to suppress the influence of the specular reflection portion detected by the specular reflection detection means . If the average pixel value of the image that does not contain the reflected light component exceeds the specified value, select the light modulation image that has the largest pixel value in the area that includes the human face, and detect it with the regular reflection detection means. The authentication process is performed so as to suppress the influence of the regular reflection portion .

According to the invention of claim 3, the regular reflection detection means detects the regular reflection portion where the pixel value is equal to or greater than the predetermined threshold in the light modulation image, and the authentication means authenticates so as to suppress the influence of the regular reflection portion. Since the process is performed, it is possible to suppress a decrease in authentication accuracy even when there is a reflected light component that is specularly reflected on glasses or the like. Further, authentication processing is performed on the light modulation image in which the size of the regular reflection portion detected by the regular reflection detection means is the smallest, or the signal-to-noise ratio is the highest among a plurality of light modulation images The authentication accuracy can be improved according to the characteristics of the light modulation image by selecting the light modulation image considered to be the authentication process.

According to a fourth aspect of the present invention, in the invention according to any one of the first to third aspects, the modulated light irradiating means has a variable pattern of the modulated light irradiated to the target space, and the regular reflection detecting means is the same. It is characterized in that a difference image of a plurality of light modulation images obtained by capturing different modulated light patterns of a person is obtained, and a regular reflection portion is extracted from the difference image .

According to the invention of claim 4 , the specular reflection portion can be detected with high accuracy.

  According to the present invention, it is possible to suppress a decrease in authentication accuracy even when there is a reflected light component that is regularly reflected by glasses or the like.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 4 shows an example of the system configuration of a security system for an apartment house including the face authentication device A of the present embodiment. This apartment house security system is installed in an apartment house such as a condominium. In addition to the face authentication device A, the lobby intercom B, the alarm monitoring panel C, the electric lock control device D, the elevator control panel E, It includes a delivery box control panel F, a system control panel G, a numeric keypad device H, a housing information panel (not shown) installed in each dwelling unit, and the like.

  The lobby intercom B is equipped with a camera for imaging visitors, a telephone device for making calls with the alarm monitoring panel C and the housing information panel, and a numeric keypad for the visitors to enter the dwelling unit number of the visited unit. It is installed outside the common entrance J of the apartment house. When a visitor operates the numeric keypad of the lobby intercom B and inputs the dwelling unit number of the visited residence unit, the lobby intercom B passes through the alarm monitoring panel C to the housing information panel installed in the dwelling unit of that dwelling unit number. Thus, a call signal and a visitor image captured by the camera of the lobby intercom B are transmitted. And in the house information board which received the calling signal, the image received from the lobby intercom B is displayed on the display and the calling sound is emitted from the speaker. When the resident of the dwelling unit presses the response button on the housing information board, a communication path is formed between the housing information board and the lobby intercom B, and the visitor and the resident use the lobby intercom B and the housing information board, respectively. You can make a call. Since this type of lobby interphone B, alarm monitoring panel C, and house information panel are well known in the art, detailed illustration and description thereof will be omitted.

  The electric lock control panel D controls locking / unlocking of an electric lock (not shown) provided at the common entrance J. The elevator control panel E controls the operation of an elevator (not shown) installed in the apartment house. The delivery box control panel F controls locking / unlocking of delivery boxes installed in the apartment house. The system control panel G causes the electric lock control panel D to unlock the electric lock or causes the elevator control panel E to operate the elevator according to the control command transmitted from the house information panel and the authentication result of the face authentication device A. Or let the delivery box control panel F unlock the delivery box. That is, when a resident of any dwelling unit operates the numeric keypad H and inputs an ID number assigned to the resident, the face authentication apparatus A uses a template corresponding to the ID number received from the numeric keypad H. The face authentication of the resident is performed, and if authentication is possible, authenticable information is transmitted to the system control panel G. When the system control panel G receives the authentication enable information from the face authentication device A, the system control panel G causes the electric lock control panel D to unlock the electric lock, causes the elevator control panel E to operate the elevator, or causes the delivery box control panel F to Unlock the delivery box. On the other hand, when the authentication by the face authentication device A is impossible, the system control panel G receives the information indicating that the authentication is impossible from the face authentication device A, so that the electric lock control panel D does not unlock the electric lock, and the elevator control panel E does not operate the elevator, and the delivery box control panel F does not unlock the delivery box. Since the electric lock control device D, the elevator control panel E, and the delivery box control panel F are well known in the art, detailed illustration and description thereof are omitted.

  FIG. 1A shows a block diagram of the face authentication apparatus A of the present embodiment. The face authentication apparatus A according to the present embodiment includes a light emitting unit 1, a light emission control unit 2, an imaging unit 3, an arithmetic processing unit 4, a template matching processing unit 5, and a memory unit 6. As shown in FIG. 1B, the light emitting unit 1 includes a plurality of light emitting diodes 1a arranged vertically and horizontally on the same plane, and a plurality of the light emitting diodes 1a individually or in groups of several. And a drive circuit (not shown) for lighting. The light emitting diode 1a emits infrared light. The light emission control unit 2 controls the drive circuit to drive the light emitting diode 1a at a very high frequency (for example, 10 megahertz), thereby irradiating the target space where the person is present with the modulated light. That is, in the present embodiment, the light emitting unit 1 and the light emission control unit 2 correspond to modulated light irradiation means.

  The imaging unit 3 is synchronized with the light emission period of the light emitting unit 1 based on an imaging device having sensitivity to infrared light (for example, a CCD <charge coupled device>) and a timing signal output from the light emission control unit 2. It has a signal processing circuit that individually outputs pixel signals when light is emitted and pixel signals when light is not emitted, and is arranged between the lens that collects infrared light on the light receiving surface of the image sensor, or between the lens and image sensor It also has a diaphragm to be used. The arithmetic processing unit 4 converts the pixel signal at the time of light emission and the pixel signal at the time of non-light emission output from the signal processing circuit into digital signals, and then generates a light modulation image having a difference between the two as a pixel value. Image generation processing, region extraction processing for extracting a region having a specific characteristic (a person's face region) from the light modulation image, and a human face feature amount (for example, eyes, nose, mouth, etc.) from the face region For example, a general-purpose microcomputer performs the light modulation image generation process, the region extraction process, and the feature quantity extraction process. It is realized by installing a program to make it. However, the light modulation image generation processing and the region extraction processing are already well-known techniques as described in Patent Document 1, and thus detailed description thereof is omitted.

  The template matching processing unit 5 collates the feature amount extracted by the feature amount extraction process of the arithmetic processing unit 4 with the template registered (stored) in the memory unit 6 in advance, so that a person in the target space can be identified. It is determined whether or not the person is registered in advance (in this embodiment, a resident of an apartment house). Then, the processing result by the template matching processing unit 5, that is, the authentication result (authentication enabled or disabled) is transmitted to the system control board G. That is, in the present embodiment, the arithmetic processing unit 4 and the template matching processing unit 5 constitute an authentication unit.

  If the authentication result received from the face authentication device A is authentic, the system control panel G causes the electric lock control panel D to unlock the electric lock, causes the elevator control panel E to operate the elevator, and further delivers the delivery box control panel. Let F unlock the delivery box. On the other hand, if the authentication result received from the face authentication device A cannot be authenticated, the system control panel G does not perform any control on the electric lock control panel D, the elevator control panel E, and the delivery box control panel F. Therefore, it is possible to prevent an unregistered person (a person other than the resident of the apartment house) from entering the apartment house without permission.

  By the way, in the light modulation image in which the difference between the light emission image and the non-light emission image is a pixel value, the influence of light shot noise in the image sensor is relatively large. Optical shot noise involves fluctuations that can be described by a Poisson distribution, and its standard deviation is proportional to the square root of the number of photons. Therefore, the signal-to-noise ratio (S / N ratio) of the light modulation image with respect to the light shot noise is the sum of the pixel value (luminance level) of the light modulation image and the pixel value of the image when emitting light and the pixel value of the image when not emitting light. It can be determined by dividing by the square root.

  Here, as shown in FIG. 2A, when the pixel value (amplitude level of the pixel signal) Y1 of the image at the time of light emission is large and the pixel value Y2 of the image at the time of non-light emission is small, the difference between the two Since the pixel value Y3 (= Y1-Y2) also becomes considerably large and the difference between the noise components N1 and N2 of the light shot noise included in the light emitting image and the non-light emitting image becomes large, the S / N ratio is sufficiently high. Larger value. However, when the pixel value Y1 of the image at the time of light emission is small as shown in FIG. 2B, or the difference between the two is small because the pixel value Y2 of the image at the time of non-light emission is large as shown in FIG. The pixel value Y3 of the light modulation image, which is the difference between the two, is also considerably small, and the difference between the noise components N1 and N2 of the light shot noise included in the light emitting image and the non-light emitting image is also small. The ratio becomes a considerably small value. That is, as shown in FIG. 2D, the S / N ratio becomes higher as the modulated light becomes stronger (the light quantity is larger) than the ambient light. In addition, the higher the S / N ratio, the easier and more accurately the feature amount can be extracted. As a result, the stronger the modulated light (the larger the amount of light) with respect to the ambient light, the better the authentication accuracy.

  On the other hand, when the modulated light is specularly reflected on spectacles worn by a person, the specular reflection component increases as the modulated light becomes stronger (the amount of light increases). As a result, in the light modulation image, as shown in FIG. 3A, a portion including a reflected light component that is regularly reflected by the glasses G worn by the person H (hereinafter referred to as “regular reflection portion”) X11. , X12, X21, and X22 increase in accordance with the amount of modulated light and overlap with the facial organs (especially the eyes and nose), and it may be difficult to extract the feature values of the facial organs.

  Therefore, in the present embodiment, the pattern (irradiation direction) of the modulated light that irradiates the target space is switched to a plurality of ways, and pixel values are determined for a plurality of light-modulated images that are generated in each pattern and image the same person. A regular reflection detection process for detecting regular reflection portions X11,... For example, the light modulation image generated in a pattern that blinks all the light emitting diodes 1a included in the light emitting unit 1 is as shown in FIG. 3A, and another pattern that blinks only some of the light emitting diodes 1a. FIG. 3B shows the light modulation image generated in FIG. 3B, and FIG. 3C shows the light modulation image generated in another pattern that blinks only some other light emitting diodes 1a. If it is as shown, the size of the regular reflection portions X11,... Detected by the regular reflection detection process is the smallest in the light modulation image shown in FIG. Then, among the plurality of light modulation images obtained by imaging the same person with different modulation light patterns, the light modulation image in which the size of the regular reflection portions X11,... Detected by the regular reflection detection process is minimized. For the light modulation image shown in FIG. 3B, the arithmetic processing unit 4 executes a feature amount extraction process to extract a feature amount, and a matching process for matching the feature amount with the template is performed as a template. If the matching processing unit 5 is used, it is possible to easily and accurately extract the feature amount of a person's face, and as a result, it is possible to suppress a decrease in authentication accuracy even when there is a reflected light component that is specularly reflected on glasses or the like. Can do.

  Here, in the imaging environment in which the S / N ratio of the light modulation image is lowered due to strong ambient light as described above (see FIG. 2C), the modulation light pattern is generated by switching between a plurality of patterns. Among the plurality of light-modulated images, authentication processing is performed on the light-modulated image having the largest pixel value in the region including the human face, that is, the light-modulated image considered to have the highest S / N ratio ( By performing the feature amount extraction process and the collation process), it is possible to improve the authentication accuracy in the imaging environment as described above.

  It should be noted that the authentication process is performed on the light modulation image in which the size of the regular reflection portion X11,... Detected by the regular reflection detection process is minimum, or the pixel value of the region including the person's face is the largest. Either of the authentication processing for the light modulation image may be selected by the arithmetic processing unit 4. For example, when the average pixel value of the non-light emitting image exceeds a predetermined value, the latter light modulation image is selected with priority on the S / N ratio, and the average pixel value of the non-light emitting image is predetermined. When the value is less than the value, the former light-modulated image may be selected so as to suppress the influence of the regular reflection portions X11,.

  Here, as a method for detecting the regular reflection portion, in addition to the above-described method for comparing the pixel value with the threshold value, a light modulation image generated with another pattern that causes only some of the light emitting diodes 1a to blink (FIG. 3). (B)) and a difference image (pixel values of pixels at the same position) between the light-modulated image (see FIG. 3C) generated with another pattern that causes only some other light-emitting diodes 1a to blink. There is a method for obtaining an image having a pixel value as a difference between the two. That is, in the difference image of two light modulation images captured with different patterns of modulated light, the pixel value of the specular reflection portion (for example, the spectacle portion) becomes extremely large compared to other portions, A portion where the pixel value of the difference image exceeds a predetermined threshold can be extracted as a regular reflection portion. According to this method, the detection accuracy of the regular reflection portion can be increased as compared with the method of comparing the pixel value of the light modulation image with the threshold value.

  Further, if the arithmetic processing unit 4 executes image processing for combining the light modulation image shown in FIG. 3B and the light modulation image shown in FIG. 3C, as shown in FIG. Regular reflection portions X11, X12, X21, and X22 existing in each light modulation image can be reduced (deleted). And if an authentication process is performed with respect to the light modulation image (composite image) which deleted the regular reflection part X11, X12, X21, X22 in this way, an authentication precision can further be improved. Note that image processing for combining the two light modulation images and deleting the regular reflection portions X11, X12, X21, and X22 as described above is well known in the art and will not be described in detail.

1 shows an embodiment of the present invention, (a) is an overall block diagram, and (b) is a schematic configuration diagram of a light emitting unit. (A)-(d) is explanatory drawing of the S / N ratio of the light modulation image in the same as the above. (A)-(d) is explanatory drawing of the regular reflection detection process etc. in the same as the above. It is a system configuration figure of the security system for apartment houses using the same.

Explanation of symbols

A Face recognition device 1 Light emitting part (modulated light irradiation means)
2 Light emission control unit (modulated light irradiation means)
3 imaging unit 4 arithmetic processing unit (light modulation image generation unit, region extraction unit, regular reflection detection unit, authentication unit)
5 Template matching processing unit (authentication means)

Claims (4)

Modulated light irradiation means for irradiating the target space with modulated light, an image sensor that receives light incident from the target space, and a signal output from each pixel of the image sensor to process only the reflected light component of the modulated light A light modulation image generation unit that generates a light modulation image having a pixel value as a pixel value, a feature amount of a human face existing in the target space based on the light modulation image generated by the light modulation image generation unit, and the feature Authentication means for authenticating a person's face corresponding to the template by comparing the amount with a pre-registered template, and regular reflection for detecting a regular reflection portion where the pixel value is equal to or greater than a predetermined threshold in the light modulation image Detecting means,
The modulated light irradiating means has a variable pattern of modulated light that irradiates the target space,
The authentication means generates a light modulation image that minimizes the size of the specular reflection portion detected by the specular reflection detection means from among a plurality of light modulation images obtained by imaging the same person with different modulated light patterns. On the other hand , an authentication process is performed so as to suppress the influence of the regular reflection portion detected by the regular reflection detection means. Modulated light irradiation means for irradiating the target space with modulated light, an image sensor that receives light incident from the target space, and a signal output from each pixel of the image sensor to process only the reflected light component of the modulated light A light modulation image generation unit that generates a light modulation image having a pixel value as a pixel value, a feature amount of a human face existing in the target space based on the light modulation image generated by the light modulation image generation unit, and the feature Authentication means for authenticating a person's face corresponding to the template by comparing the amount with a pre-registered template, and regular reflection for detecting a regular reflection portion where the pixel value is equal to or greater than a predetermined threshold in the light modulation image Detecting means,
The modulated light irradiating means has a variable pattern of modulated light that irradiates the target space,
The authentication means performs regular reflection on the light modulation image in which the pixel value of the region including the face of the person is the largest among the plurality of light modulation images obtained by imaging the same person with different modulation light patterns. face authentication apparatus you and performs an authentication process so as to suppress the influence of the specular reflection portion to be detected by the detection means. Modulated light irradiation means for irradiating the target space with modulated light, an image sensor that receives light incident from the target space, and a signal output from each pixel of the image sensor to process only the reflected light component of the modulated light A light modulation image generation unit that generates a light modulation image having a pixel value as a pixel value, a feature amount of a human face existing in the target space based on the light modulation image generated by the light modulation image generation unit, and the feature Authentication means for authenticating a person's face corresponding to the template by comparing the amount with a pre-registered template, and regular reflection for detecting a regular reflection portion where the pixel value is equal to or greater than a predetermined threshold in the light modulation image Detecting means,
The modulated light irradiating means has a variable pattern of modulated light that irradiates the target space,
The authentication unit is an average value of pixel values of images that do not include the reflected light component of the modulated light output from the imaging element among a plurality of light-modulated images obtained by imaging the same person with different modulated light patterns. Is less than or equal to a predetermined value, a light modulation image that minimizes the size of the regular reflection portion detected by the regular reflection detection unit is selected to suppress the influence of the regular reflection portion detected by the regular reflection detection unit. When the average value of the pixel values of the image that does not include the reflected light component of the modulated light exceeds a predetermined value, the light modulated image in which the pixel value of the region including the person's face is the largest is obtained. select, face authentication apparatus and performing impact authentication process so as to suppress the specular reflection portion to be detected by the regular reflection detection means. The modulated light irradiating means has a variable modulated light pattern that irradiates the target space, and the regular reflection detecting means is a differential image of a plurality of light modulated images obtained by imaging the same person with different modulated light patterns. look, face authentication device according to any one of claims 1-3, characterized by extracting the specular reflection portions of the difference image.

Images