KR20070077973A - Apparatus for recognizing a biological face and method therefor - Google Patents

Abstract

A device and a method for recognizing a face are provided to determine whether a face image inputted from an imaging device is a biometric face or a photo, and discriminate the biometric face even if the face image is not moved. A face detector(100) detects a face image from an inputted image by using an AdaBoost(Adaptive Boosting) algorithm. A biometric determiner(120) determines whether the face of the inputted image is the biometric face based on change of a light reflection pattern in a face area according to the change of a lighting position. An eye detector(110) detects an eye area from the detected face image by using a multi-block matching model, which applies a normalized template matching model. The biometric determiner determines whether the face of the inputted image is the biometric face based on the change of a light reflection pattern in the face area and the change of the light reflecting pattern in the eye area.

Description

Apparatus for recognizing a biological face and method therefor}

1 is a view showing the configuration of an embodiment of a biometric face recognition apparatus according to the present invention;

FIG. 2 is a diagram illustrating a detailed configuration of a biometric determination unit of the biometric face recognition apparatus of FIG. 1 according to the present invention; FIG.

3 is a view showing a change in the input face image according to the position of the illumination;

4A and 4B are views illustrating a change in a light reflection pattern of a face according to a change in position of illumination;

5 is a diagram illustrating a light reflection pattern of a real human eye, and

6 is a flowchart illustrating a flow of an embodiment of a biometric face recognition method according to the present invention.

The present invention relates to a biometric face recognition apparatus and a method thereof, and more particularly, to an apparatus and method for determining whether a face shown in an input image is a real human face or a photograph.

Conventional face recognition methods include a feature-based method using a geometrical information of a face and an appearance-based method using a model trained by a training image set. Representative face recognition technologies include PCA (Principle Component Analysis: Eigenfaces), ICA (Independent Component Analysis), and LDA (Linear Discriminant Analysis) as effective projection methods to effectively reduce data in high-dimensional space to low-dimensional data. .

Because PCA uses vectors that maximize the spread of all data as the basis vector, it is suitable for representing data and minimizes errors in restoring data from projected low-dimensional space back to high-dimensional space. PCA represents data using orthonormal eigenvectors obtained as eigenvalues of the data covariance matrix as basis vectors. ICA, on the other hand, calculates the independent components of the data to represent the data and finds the subspace that maximizes the statistical independence between the projected data. LDA is a method that uses linear transformation to minimize intra-class variance and maximum inter-class variance for face vectors in order to obtain information suitable for recognition purposes in a reduced vector space obtained by applying PCA. Such conventional face recognition technology has a problem in that it is impossible to determine whether it is a face of a real person or a face of a photograph.

In a conventional face recognition system, a method of distinguishing a photograph from a real human face may include detecting a movement of a face component in an input image or comparing the background image of an input image with a background image of a reference image. However, since the prior art is based on the information of the motion and the background, it is difficult to determine whether the input image is a real person's face or a face of the picture when the person or face picture is stopped without moving.

The present invention provides a biometric face recognition apparatus and a method for distinguishing whether a face image input through an image input device is a real human face or a photo, and in particular, even if there is no movement. To provide.

Another object of the present invention is to provide a computer-readable recording medium having recorded thereon a program for executing the biometric face recognition method on a computer.

In accordance with another aspect of the present invention, there is provided a biometric face recognition apparatus, including: a face detector configured to detect a face region from an input image; And a living body determination unit configured to determine whether a face shown in the input image is a face of a real person based on a change in a reflection pattern of light in the face area according to a change in position of illumination.

In order to achieve the above technical problem, an embodiment of a biological face recognition method according to the present invention includes: (a) detecting a face region from an input image; And (b) determining whether a face shown in the input image is a face of a real person based on a change in a reflection pattern of light in the face region according to a change in position of illumination. Y

As a result, it is possible to distinguish whether a face shown in the input image is a real human face or a face printed on paper.

Hereinafter, with reference to the accompanying drawings will be described in detail a biometric face recognition apparatus and method according to the present invention.

1 is a diagram illustrating a configuration of an embodiment of a biometric face recognition apparatus according to the present invention.

Referring to FIG. 1, the biometric face recognition apparatus includes a face detector 100, an eye detector 110, a biometric determiner 120, and a face recognizer 130.

The face detector 100 detects a face region in a continuous input image input in real time through a CCTV, CCD, image camera, or infrared camera including lighting. Although there are various conventional methods for detecting a face region in an input image, the face detector 100 according to an embodiment of the present invention may use a cascaded face classifier using an AdaBoost (Adaptive Boost Learning Filter) algorithm. ) To detect the face area.

Since the conventional face detection method uses motion, a person must move or a color camera must be used when the skin color extraction method is used in color information, and is sensitive to light change, race, and skin color. The method developed to solve this problem is Cascaded Face Classifier using AdaBoost.

The Cascaded Face Classifier algorithm using AdaBoost requires a complex decision boundary for class classification when the change in the class is very severe. The AdaBoost algorithm is a classifier learning algorithm suitable for such a situation. The AdaBoost algorithm generates a strong classifier through a combination of several weak classifiers and is suitable for classifier learning for face detection.

Since the Cascaded Face Classifier algorithm using AdaBoost does not use motion information and color information, it is possible to perform face detection at high speed without the user's limitation in general black and white camera.

The eye detector 110 finds the position of the eye in the face area detected by the face detector 100. The present invention uses the eye as a reference for normalizing the detected face image. The eye detector 110 uses a multi-block matching (MMF) technique to accurately locate and track the eyes under conditions such as wearing glasses, face rotation, and eyes closed.

MMF is an applied model of the Normalized Template Matching technique. It collects various databases (DBs) of the target patterns to be detected, creates a representative image among the collected databases, and then uses image based templage matching. ) Is how to do it. The eye detector 110 calculates in advance various sizes, lighting, rotations, and the like of an eye image using an MMF, and holds the template as a template, and detects the eye position by performing template matching with the face region using the template.

In detail, the eye detector 110 generates a virtual face image using a plurality of eye candidates selected by using an MMF technique, and selects a candidate having an actual face therein. The classifier used in the candidate selection process uses the K-Mean Clustering method for a large-scale face DB, so that low-resolution high-speed eye detection is possible.

K-Mean Clustering Algorithm is the number of clusters to be calculated as K. It is an iterative algorithm that sets K center vectors from the beginning, clusters them based on these vectors, and resets the center vectors from the clustered results. .

The biological determination unit 120 determines whether the face shown in the input image is a real person's face or a photo by measuring a change in the reflection pattern of the light according to a change in the position of the light in the face area detected by the face detection unit 100. . The detailed configuration of the biological determination unit 120 will be described in detail with reference to FIG. 2.

When a certain light shines on an object, the light is reflected and absorbed according to the surface characteristics of each object. The human face also reflects and absorbs light from the eyes, nose, mouth, forehead, and cheeks, and the eyes reflect the light better than other facial parts. That is, the pattern of light reflected from the face of the real person and the pattern of light reflected from the face photograph printed on the paper are different from each other (see FIGS. 4A and 4B).

When illuminating a photograph of a face printed on paper, the degree of reflection and absorption of light depends on the characteristics of the surface of the paper, but does not differ for each part of the face printed on the paper. That is, in the case of a face photograph printed on paper, the amount of light reflected and absorbed by the eyes, nose, mouth, forehead, and cheeks is almost the same. The amount is different.

If the face shown in the input image is identified as the real human face by the biometric identification unit 120, the face recognition unit 130 identifies whose face is the detected face. Therefore, it is possible to prevent crimes using pictures of others in access control systems and security systems.

In detail, the face recognition unit 130 normalizes the face image detected from the input image, extracts a unique face feature, and compares the face image with a reference image previously stored in a predetermined database.

There are a number of effective features on the face for 200 to 3000 recognitions. In order to extract the features that can be used robustly for lighting and facial expression change among these effective features, face recognition module learning process for large-scale DB is conducted. In general, the feature extraction process (Principle Component Analysis) is used for the feature extraction, which means the conversion from the image space to the feature space, and the applied versions thereof include LDA and ICA.

3 is a diagram illustrating a change of an input face image according to a position of lighting.

Referring to FIG. 3, in the input face image, faces of the same person are differently represented according to lighting positions and brightness. When the similarity is measured by cosine distance for each face image of FIG. 3, the image of (a) is similar to the image of (c) than the image of (b). As described above, since a human face is sensitive to a change in illumination unlike a face image of a photograph, the present invention may determine whether an input face image is a real human face image or a photo based on a change of a light reflection pattern of a face according to a light conversion. .

4A and 4B are views illustrating a change in a light reflection pattern of a face according to a change in position of illumination.

(A) and (b) of FIG. 4a are face images photographing real human faces according to a change in illumination position, and (c) and (d) are photographs of a face-printed paper according to a change in illumination position. It is a video. In the case of (a) and (b) of FIG. 4a, the light reflection pattern of the input face image is different according to the change in the lighting position. However, in the case of (c) and (d), It can be seen that the reflection pattern hardly changes.

FIG. 4B is an image in which each image of FIG. 4A is changed into a binary image of black and white based on a predetermined threshold value so that the change of the reflection pattern of light can be clearly understood. Looking at the binary image of Figure 4b, it can be seen that the reflection pattern of the light is clearly different according to the change in the position of the light in the case of a real human face ((a) and (b)), but in the case of paper almost no change ((C) and (d)).

In particular, when the light is photographed by placing the front face and the side of the face respectively, since the actual human face is a three-dimensional solid shape, the light reflection pattern shows a greater difference depending on the position of the front and side lighting. However, in the case of a two-dimensional plane as shown in the photo, the reflection pattern of the light is almost constant regardless of the change in the position of the light.

FIG. 2 is a diagram illustrating a detailed configuration of the biometric determination unit of the biometric face recognition apparatus of FIG. 1 according to the present invention.

Referring to FIG. 2, the biological determiner 120 includes a motion determiner 200, a pattern measurer 210, and a biological face determiner 220. Hereinafter, a configuration of determining whether a face is a real human face based on a light reflection pattern in the eye among the face areas will be described.

Referring to FIGS. 1 and 2, the motion determiner 200 detects the movement of the face region seen in the input image based on the presence or absence of a change in distance between both eyes detected by the eye detector 110 (FIG. 1). . For example, when a person comes closer to the camera, the distance between both eyes detected by the eye detector 110 becomes larger, and when the person moves away, the distance between both eyes becomes smaller. The motion determiner 200 determines that there is a motion when the distance difference between the eyes in the previous image and the current image of the continuous input image changes more than a predetermined threshold value.

As shown in FIGS. 3, 4A, and 4B, in the case of a real human face, a light reflection pattern may be changed according to a change in position of illumination. The change in position of the illumination occurs when the physical position of the actual illumination changes and when the face shown in the input image moves relative to the illumination.

When the pattern measuring unit 210 detects a movement of the face region visible in the input image, the pattern measuring unit 210 measures a change in the light reflection pattern of the face region according to the movement. In addition, if the movement of the face region detected in the input image is not detected (ie, stopped), the pattern side unit sequentially turns on the current light and the auxiliary light existing at a different position from the current light, The change in the light reflection pattern of the face region photographed by the illumination and the auxiliary illumination is measured. The pattern measuring unit 210 obtains a light reflection pattern of the face region according to the change of illumination by using Equation 1 below.

Here, u _i is an i-th face image and f _i is a difference image of u _i and u _i-1 .

The difference between a real human face and a face printed on paper is more evident in the eyes. In the case of the real human eye, the reflection of light by the light occurs most, but the eye of the face printed on paper only reflects the same degree as other parts.

Therefore, the pattern measuring unit 210 may determine the living body face more accurately by detecting the light reflection pattern of the eye in addition to the face area. 5 is a view illustrating a light reflection pattern of an eye by illumination.

)을 구한다.Specifically, the pattern measuring unit 210 measures the reflection pattern of light in the eye area when the position of the eye is determined by the eye detector 110. The pattern measurer 210 obtains a histogram of the eye image to find an area where illumination is most prominent in the eye, and increases the contrast by smoothing the histogram brightness distribution to a wide area. And binary image consisting of black and white using Equation 2 below (

)

는 i 번째 눈 영상내에 속해있는 k 번째 화소이며,

와

는 i번째 눈 영상내에 속해있는 화소들의 평균과 표준편차를 나타낸다.here,

Is the k th pixel belonging to the i th eye image,

Wow

Denotes the mean and standard deviation of the pixels in the i-th eye image.

The binarized eye image obtained in Equation 2 is removed by noise and the light reflection pattern of the eye is extracted by using the expansion and erosion image processing technique of the Morphological Operation of Equation 3 below. The basic idea of the morphology technique is to extract only the desired part of the image by reflecting the information of the target object that knows the geometric shape in advance.

B _e and B _d are the erosion (2x2) and dilation (5x5 circle) operators.

The biological face determiner 220 determines whether the person is a real person by measuring the similarity between the binary image of the eye extracted by the pattern measuring unit and the reference image of the pre-stored eye using the Euclidean distance measure.

The Euclidean distance used by the biometric face grasping unit 220 is a geometric distance between two patterns a and b and is used as a criterion for evaluating similarity.

Here, a (k) is a reference image, b _i (k) is an eye image of the i-th input image, k is the number of pixels. The Euclidean distance is a difference between brightness values of a reference image and an input image. The biometric face determiner 220 converts the input image into a face image of a person if the value D obtained as shown in Equation 5 below is smaller than a threshold value. To judge. The reference image and the threshold may vary from place to place where face recognition is required.

6 is a flowchart illustrating a flow of an embodiment of a biometric face recognition method according to the present invention.

Referring to FIG. 6, an image is received through a CCTV, a camera, or the like including lighting (S600). A face region in the image is detected by using a cascaded face classifier algorithm using an Adaboost on the input image (S605). If the face region is detected, the position of the eye is determined through the MMF technique with respect to the image of the face region (S610).

The location of the identified eye is tracked in real time to determine whether there is a motion in the face region (S615). For example, when the distance between both eyes is far, the face area is far from the camera. On the contrary, when the distance between both eyes is close, the face area is closer to the camera.

When the face region moves, the position of the light reflected on the face region changes relatively even though there is no change in the position of the light, and thus the change of the light reflection pattern in the face region according to the movement of the face region is measured (S625).

If the face area does not move, the physical position of the illumination is changed and then the change of the light reflection pattern of the face area under the illumination of each other position is measured (S630). The change in the position of a light moves one light or alternates both lights located in different places.

Then, the light reflection pattern for the eye area is obtained (S635). In the case of the eye, the reflection of light is better than that of other areas of the face, so the eyes have a distinct difference from the face drawn on paper.

Based on the reflection pattern of the light on the face area and the reflection pattern of the light on the eye area according to the change in the position of the illumination, it is determined whether the face shown in the input image is a real human face (S640). As shown in FIGS. 4A and 4B, the face printed on the paper and the real face show a distinct difference in the light reflection pattern according to the change in the illumination position, and thus it is easy to distinguish whether the face is a real human face.

If it is identified as a real person's face, it recognizes who's face for authentication or the like (S645). This prevents misuse of another person's face in an authentication system using a face.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in descriptive sense only and not for purposes of limitation. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

According to the present invention, the input face image is detected by changing the light reflection pattern of the face and the light reflection pattern of the pupil according to the change in the position of the light, whether the input face image is a real human face image or a photographic face. By discriminating, it is possible to prevent a criminal act that attempts to recognize a face by using a picture of another person in the face recognition system.

Claims (15)

A face detector detecting a face region from an input image; And A biometric face recognition unit which determines whether a face shown in the input image is a face of a real person based on a change of a reflection pattern of light in the face region according to a change in position of illumination; Device. The method of claim 1, And the face detector detects a face region from the input image by using an Adaboost algorithm. The method of claim 1, And an eye detector configured to detect an eye region from the detected face region. And the biometric determination unit determines whether the input image is a face image of a real person based on a light reflection pattern in the face region and a light reflection pattern in the eye region. The method of claim 3, wherein The face detection apparatus of claim 1, wherein the eye detector detects a position of an eye in the face region using a multi-block matching method, which is an applied model of a normalized template matching method. . The method of claim 1, The biometric determination unit converts the image of the face region for each before and after the position change of the illumination into a black and white binarized image based on a predetermined threshold value, and then changes the light reflection pattern based on the difference between the binarized images. And measuring whether the input image is an image of a real person based on the change of the reflection pattern. The method of claim 1, wherein the biological determination unit, A motion determination unit which detects both eye regions in the detected face region and determines movement of the face region based on a change in distance between the two eye regions; If there is no movement of the face region, the change of the reflection pattern of light with respect to the face region of the input image photographed through illumination of different positions is measured. If there is movement of the face region, the light of the face region according to the movement is measured. A pattern measuring unit measuring a change in the reflection pattern; And And a biometric face determiner configured to determine whether the input image is a face image of a real person based on the measured change of the reflection pattern. The method of claim 6, The pattern measuring unit detects eyes in the face region, smooths the histogram brightness distribution of the detected eyes to a wide region, and converts the image of the eyes into black and white binary images based on a predetermined threshold, and then uses a morphology technique. using the morphological operation expansion and erosion image processing techniques to remove noise and measure the light reflection pattern of the eye, The biometric face detecting unit compares the similarity between the light reflection pattern image of the eye and a pre-stored reference image using an euclidean distance measure to determine whether the face shown in the input image is a real human face. Biometric face recognition device. (a) detecting a face region from an input image; And (b) determining whether a face shown in the input image is a face of a real person based on a change in a reflection pattern of light in the face area according to a change in position of illumination; Recognition method. The method of claim 8, The step (a) comprises detecting a face region from the input image using an Adaboost algorithm. The method of claim 8, Step (a) further comprises the step of detecting the eye area in the detected face area, The step (b) includes determining whether the input image is a face image of a real person based on the light reflection pattern in the face region and the light reflection pattern in the eye region. Recognition method. The method of claim 10, Step (a) includes detecting the position of the eye in the face region using a multi-block matching method which is an applied model of a normalized template matching method. Biometric face recognition method. The method of claim 8, In the step (b), the image of the face region for each before and after the position change of the illumination is converted into a black and white binarized image based on a predetermined threshold value, and then the light reflection pattern is generated based on the difference between the binarized images. Measuring a change, and determining whether the input image is an image of a real person based on the change of the reflection pattern. According to claim 8, wherein step (b) is, (b1) detecting both eye regions in the detected face region and determining a movement of the face region based on a change in distance between the two eye regions; (b2) if there is no movement of the face region, measuring a change of a light reflection pattern with respect to a face region of an input image photographed through illumination at different positions; and if there is a movement of the face region, the face region according to the movement. Measuring a change in the light reflection pattern of the light; And (b3) determining whether the input image is a face image of a real person based on the measured change of the reflection pattern. The method of claim 13, In the step (b2), the eye is detected in the face region, the histogram brightness distribution of the detected eye is smoothed into a wide region, and the image of the eye is converted into a black and white binary image based on a predetermined threshold value. Measuring the reflection pattern of light of the eye after removing noise using an expansion and erosion image processing technique of a morphological operation, and The step (b3) is a step of determining whether the face shown in the input image is a real human face by comparing the similarity between the light reflection pattern image of the eye and a pre-stored reference image using an euclidean distance measure. Biometric face recognition method comprising a. A computer-readable recording medium having recorded thereon a program for executing the biometric face recognition method according to any one of claims 8 to 14.

Images