KR101904924B1 - Apparatus for virtualizing of facial image - Google Patents

Abstract

The present invention relates to a face image virtualizing device which comprises: a face area detecting unit which detects a face area from one or more image data where an object person is photographed, and allocates the number of photography to each of the one or more image data by corresponding to the order of photographing of the one or more image data; a virtual face feature value generating unit which generates a virtualization random number for each of the one or more image data as much as the number of occurrence corresponding to the number of photography, and converts a face feature value extracted from the face area into a virtual face feature value by using the generated virtualization random number; and a virtual face image data generating unit which searches for a virtual face image in a face image database based on a virtual face feature vector where the virtual face feature value is vectorized, and generates virtual face image data by inserting the searched virtual face image into the face area of the image data. Therefore, the face image virtualizing device can provide virtual face image data capable of identifying only public information such as sex, age, and race by changing a face image located in a face area in image data into a virtual face image corresponding to a preset similarity.

Description

[0001] Apparatus for virtualizing a facial image [

The present invention relates to a face image virtualization apparatus, and more particularly, to a face image virtualization apparatus that inserts a virtual face image corresponding to a degree of similarity set in advance in a face region of image data.

Generally, security and surveillance video systems such as CCTV, DVR, etc. are used for 24 hour surveillance in homes, schools, security areas around playgrounds, vulnerable places where employees can not manage at all times, public places such as banks, department stores, It is possible to prevent the occurrence of accidents, dangerous situations and crimes through the monitoring of the situation managers by establishing equipment such as surveillance cameras in the facilities and transmitting and displaying the scene images in real time to the central control center at the remote site connected to the network. It can collect follow-up measures and arrest measures by gathering various safety inspections and various information. In addition to being able to know the abnormality of many surveillance places with fewer manpower, it also provides key information necessary for the investigation of crimes or finding missing persons .

In such a surveillance system, when an image transmitted by a surveillance camera is reproduced or monitored by a central control center, a so-called privacy invasion problem occurs because sensitive personal information related to privacy is exposed as is, such as a human face.

Accordingly, before the image is transmitted from the normal surveillance camera, the image is masked and transmitted to a specific area preset in the image or a specific object detected through image recognition.

However, when masking the face of a person in the image acquired by the surveillance system for privacy, the information related to the privacy invasion such as the sex, age, and race of the person is also damaged.

Accordingly, there is a problem in that it is impossible to classify the images captured by the surveillance system into items such as sex, age, race, etc., and use the images acquired in the surveillance system for academic purposes or marketing purposes.

The object of the present invention is to provide virtual face image data capable of identifying only the public information such as sex, age, and race by changing the face image located in the face region of the image data to a virtual face image corresponding to the preset similarity degree have.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention which are not mentioned can be understood by the following description and more clearly understood by the embodiments of the present invention. It will also be readily apparent that the objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims.

According to another aspect of the present invention, there is provided a face image virtualization apparatus for detecting a face region from at least one image data of a target person, A facial region detection unit for assigning a photographed face to the imaging region, a virtual random number generator for generating the virtual random number for each of the at least one image data by the generation order corresponding to the imaging sequence, A virtual face feature value generation unit for converting the virtual face feature value into a face feature value, and a virtual face feature vector generating unit for searching a virtual face image in the face image database based on the virtual face feature vector in which the virtual face feature value is vectorized, Insert a virtual face image To generate facial image data and includes a virtual facial image data generated.

Preferably, the virtual facial feature value generator may generate the virtual random number by setting a predetermined seed value as an initial value of the random number generation function, and executing the random number generation function by the generation order in the shooting order.

Preferably, the virtual face feature value generation unit may generate a noise value by adding the deviation of the virtual random number to a predetermined similarity value, and add the noise value to the face feature value to generate the virtual face feature value .

Preferably, the virtual face feature value generator may generate the noise value using the following equation.

≪ Equation &

Here, N (i) is a noise value, T is a preset similarity value, P (i) S is a virtual random number, and R1r is a predetermined virtual random number generation range.

Preferably, the facial image virtualization apparatus according to the present invention sets the seed value to the initial value of the reconstructed random number generation function, and outputs the restored random number to the generation number corresponding to the imaging sequence assigned to the virtual face image data And a face feature value restoring unit for generating a restored random number by executing the function.

Preferably, the facial feature value restoration unit may inversely process the noise value used to generate the virtual facial feature value by adding the deviation of the restored random number to the preset similarity value, and add the inversed noise value The restored face feature value may be generated by subtracting the restored face feature value from the face feature value.

Preferably, the facial feature value restoration unit may invert the noise value used to generate the virtual facial feature value using the following equation.

≪ Equation &

Herein, N '(i) is the inversed noise value, T is the preset similarity value, R2 is the restored random number, and R1r is the predetermined virtual random number generation range.

Preferably, the facial image virtualization apparatus searches for a restored face image in the face image database based on the restored face feature vector having the restored facial feature value vectorized, And a facial image data reconstruction unit for reconstructing the restored image data by inserting the retrieved restored face image into the image data.

According to the present invention, virtual face image data capable of identifying only the public information such as sex, age, and race can be provided by changing the face image located in the face region of the image data to a virtual face image corresponding to the preset similarity degree, It is possible to protect personal information of individual persons photographed in image data while using image data photographed in public places for academic purposes or marketing purposes.

1 is a diagram illustrating a connection configuration between a facial image virtualization apparatus and an external apparatus according to an embodiment of the present invention.
2 is a diagram illustrating a configuration of a facial image virtualization apparatus according to an embodiment of the present invention.
3 is a view showing a position of a camera for transmitting image data to a facial image virtualization apparatus according to an embodiment of the present invention and a moving path of a person to be photographed.
FIG. 4 is a view showing an image of image data photographed by a person to be photographed, which is located at point A in FIG. 3, from a camera;
5 is an enlarged view of the detected face area.
6 is a view showing a face image included in the face image database.
7 to 10 are views showing images of virtual face image data in which a virtual face image generated according to an embodiment of the present invention is inserted.
11 and 12 are views showing images of virtual face image data in which a virtual face image generated according to another embodiment of the present invention is inserted.
13 is a diagram illustrating an image of restored image data in which a restored face image generated according to an embodiment of the present invention is inserted.

The above and other objects, features, and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings, which are not intended to limit the scope of the present invention. In the following description, well-known functions or constructions are not described in detail since they would obscure the invention in unnecessary detail. Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the drawings, the same reference numerals are used to denote the same or similar elements.

FIG. 1 is a diagram illustrating a connection configuration between a facial image virtualization apparatus and an external apparatus according to an embodiment of the present invention. FIG. 2 is a diagram illustrating a configuration of a facial image virtualization apparatus according to an embodiment of the present invention, 3 is a view showing a position of a camera for transmitting image data to a facial image virtualization apparatus according to an embodiment of the present invention and a moving path of a person to be photographed.

1 to 3, a facial image virtualization apparatus 100 according to an embodiment of the present invention includes a facial image virtualization apparatus 100 for acquiring a facial image from one or more cameras (C1, C2, C3) The virtual face image data may be generated by receiving the data, or the generated virtual face image data may be restored to generate restored image data.

Here, one or more cameras C1, C2, and C3 may be one of cameras used in various applications such as a CCTV, a PC, a web camera mounted on a notebook computer, and a camera mounted on a mobile terminal. Needless to say, the present invention is not necessarily limited thereto.

Meanwhile, the facial image virtualization apparatus 100 can transmit the generated virtual facial image data and reconstructed image data to an external image database (DB).

For this, the facial image virtualization apparatus 100 according to an embodiment of the present invention

The face region detection unit 110 may detect a face region from one or more image data photographed by one or more cameras C1, C2, and C3.

More specifically, the face area detection unit 110 can obtain coordinates indicating a face area using one or more pieces of luminance information on the image data in one or more pieces of image data.

The face region detection unit 110 detects a position that gives the highest correlation value as a face region by obtaining a correlation value while moving a template prepared in advance in the input image data, The face region can be detected from the image data by using the method of detecting the face region using the spatial method.

Of course, the face region detection unit 110 is not necessarily limited to this, and can be selectively applied to various known methods in detecting a face region.

FIG. 4 is a view showing an image of image data photographed by a person to be photographed, which is located at point A in FIG. 3, from a camera;

4, the face area detection unit 110 detects an area including the face image of the person to be photographed, excluding the background image, from the image data VD1 photographed from the camera C1 as the face area 10 can do.

At this time, the face region detection unit 110 may detect the face region 10 as a circle or an ellipse corresponding to the shape of the face image, and may detect the face image as a rectangle included in the face region 10.

Meanwhile, the face region detection unit 110 may perform a role of assigning a photographing sequence to one or more pieces of image data corresponding to the order in which one or more image data are captured.

To this end, the face area detection unit 110 may assign a photographing time to each image data in accordance with the order in which the face area is detected when the face area is detected from the image data input from one or more cameras (C1, C2, C3) have.

For example, as shown in FIG. 3, as the subject to be photographed moves in the order of A, B, and C, the subject to be photographed is first photographed on the camera C1 and then on the camera C2 and camera C3 And the third time.

Accordingly, the face area detection unit 110 can detect the face area first in the image data photographed from the camera C1, and the second and third faces in the image data photographed from the camera C2 and the image data photographed from the camera C3, Area can be detected.

At this time, the face area detection unit 110 may allocate 1 to 3 photographing times of the respective image data photographed by the cameras C1 to C3 according to the order in which the face areas are detected.

Accordingly, the face area detection unit 110 may not allocate the photographing sequence to the image data in which the face area is not detected. For example, even if the image data photographed by the camera C4 in which the photographed person is not photographed is inputted, the face area detection section 110 does not detect the face area in the image data photographed by the camera C4, It is not possible to allocate them to the image data.

5 is an enlarged view of the detected face area.

5, the facial feature value generation unit 120 may generate facial feature values in the face region detected by the facial region detection unit 110. [

5, the facial feature value generation unit 120 generates facial feature points O (i, j) representing facial features that can identify an individual in the face region 10 detected from the facial region detection unit 110, ), And coordinate information of the extracted facial feature point O can be generated as a facial feature value.

At this time, the facial feature value generation unit 120 may apply various known methods to extract facial feature points O.

For example, the facial feature value generation unit 120 includes a plurality of average models of facial feature point coordinates, compares the average facial feature point coordinates with coordinates corresponding to the luminance value of the facial region, Can be extracted.

Then, the facial feature value generation unit 120 may generate facial feature points O as facial feature values based on the coordinate information of the facial feature points O.

The virtual facial feature value generation unit 130 may convert the facial feature value generated by the facial feature value generation unit 120 to a virtual facial feature value using a virtual random number.

More specifically, the virtual facial feature value generation unit 130 can generate a virtual random number in response to the imaging sequence through the random number generation function.

To this end, the virtual facial feature value generator 130 may generate a virtual random number by setting a predetermined seed value as an initial value of a random number generating function, and executing a random number generating function in the order of the imaging order.

At this time, the virtual facial feature value generating unit 130 may execute a random number generating function so that a virtual random number is generated within a predetermined virtual random number generating range.

In other words, the virtual facial feature value generation unit 130 can generate a virtual random number by executing the random number generation function by the imaging sequence assigned to the image data.

For example, the virtual face feature value generation unit 130 generates a virtual face feature value by using the virtual random number to be applied to the image data VD1 of the camera C1, which is first detected from the face area detection unit 110, A virtual random number can be generated by executing the random number generating function once in correspondence with the photographing serial number 1 of the video data VD1 of the camera C1.

The virtual face feature value generation unit 130 generates a virtual random number to be applied to the image data of the camera C2 in which the face area 10 is detected for the second time from the face area detection unit 110 and the photographing rotation number is assigned as 2, A virtual random number can be generated by executing the random number generating function twice corresponding to the photographing sequential number 2 of the video data of the camera C2.

Here, the random number generation function can be expressed by the following equation (1).

&Quot; (1) "

Here, P is a random number generating function, i is the issuing number of the random number generating function, and S is a preset seed value.

The virtual facial feature value generator 130 may generate the same virtual random number in the same generation time using the random number generating function having the same seed value and virtual random number generating range.

For example, when the virtual facial feature value generator 130 generates the virtual random numbers 30, 15, and 45 in the first to third generation times using the random number generating function, the virtual face feature value generating unit 130 generates the virtual random number The same virtual random numbers 30, 15 and 45 can be sequentially generated by generating the virtual random number through the first to third generation times using the same random number generating function.

Meanwhile, the virtual face feature value generation unit 130 generates a noise value by adding a deviation of a virtual random number to a preset similarity value, and adds a noise value to the face feature value to generate a virtual face feature value Can be performed.

More specifically, the virtual facial feature value generation unit 130 generates a virtual random number by executing a random number generation function corresponding to the imaging sequence assigned to the image data, and generates a virtual random number by multiplying the generated virtual random number by an intermediate value The noise value can be generated by adding the difference value between the generated virtual random numbers to the preset similarity value.

At this time, the virtual face feature value generation unit 130 may generate the noise value using the following equation (2).

&Quot; (2) "

Here, N (i) is a noise value, T is a preset similarity value, P (i) S is a virtual random number, and R1r is a predetermined virtual random number generation range.

For example, when the virtual face feature value generator 130 generates a noise value applied to the image data VD1 of the camera C1 with preset similarity values and predetermined virtual random number generation ranges of 50% and 50, respectively, The virtual facial feature value generation unit 130 can generate a virtual random number by executing a random number generation function by one shot number assigned to the video data VD1 of the camera C1. Then, the virtual facial feature value generation unit 130 generates a noise value of 35 by adding the difference value 25 between the virtual random number 30 and the intermediate value 25, which is the intermediate value of the virtual random number generation range 50, .

As described above, the virtual face feature value generation unit 130 can generate noise values of 50 and 20, respectively, applied to the image data of the camera C2 and the camera C3, which are assigned two times and three times, respectively .

This can be expressed by the following equation (3).

&Quot; (3) "

Here, N (i) is a noise value, T is a preset similarity value, P (i) S is a virtual random number, and R1r is a predetermined virtual random number generation range.

Then, the virtual face feature value generation unit 130 can generate the virtual face feature value by virtualizing the generated noise value to the face feature value generated from the face feature value generation unit 120.

Accordingly, the virtual face feature value generation unit 130 can generate a virtual face feature value including another face feature by modifying the face feature value including the face feature of the face region.

At this time, as the magnitude of the added noise value is larger, a virtual face feature value including a face feature different from the face feature value can be generated.

That is, the virtual face feature value generation unit 130 generates a noise value in inverse proportion to a preset similarity value, thereby generating a noise value having a larger value as a preset similarity value is set to a smaller value, And a virtual face feature value.

In contrast, the virtual facial feature value generator 130 may generate a noise value having a smaller value as the previously set similarity value is set to a larger value, thereby generating a virtual facial feature value including a facial feature similar to the facial feature value.

Accordingly, the facial image virtualization apparatus 100 according to the present invention adjusts a similarity value set in advance and inserts a virtual face image similar to the face image of the image data into the image data, or identifies an individual among the face images of the image data Only the public information (race, age, sex) except for the information contained therein can insert a similar virtual face image into the image data.

The virtual face feature vector generation unit 140 may perform a role of vectorizing the virtual face feature value generated from the virtual face feature value generation unit 130 to generate a virtual face feature vector.

More specifically, the virtual facial feature vector generation unit 140 may extract the virtual facial feature vector including the mxn-dimensional information from the virtual facial feature value, which is a coordinate value configured by a matrix.

The virtual face feature vector generation unit 140 may generate virtual face feature vectors by vectorizing the virtual face feature values using a known vectorization method.

It should be noted that the vectorizing method used by the virtual face feature vector generating unit 140 is not limited.

FIG. 6 is a view showing a face image included in the face image database, and FIGS. 7 to 10 are views showing an image of virtual face image data in which a virtual face image generated according to an embodiment of the present invention is inserted to be.

6 to 10, the virtual face image data generation unit 150 generates the face image data (FF) included in the face image database based on the virtual face feature vector generated from the virtual face feature vector generation unit 140, (VF1, VF2, VF3) among the virtual face images (VF1, VF2, VF3).

6, the virtual face image data generation unit 150 compares the generated virtual face feature vector with a face feature vector of a face image (FF) included in the face image database, The facial feature vector having a high degree of similarity corresponding to the face image FF can be searched with the virtual face images VF1, VF2, VF3.

To this end, the virtual facial image data generation unit 150 includes a Principal Component Anal- ysis (PCA), a between-class distribution and a within-class distribution for analyzing the similarity between vectors using a covariance matrix, Linear Discriminant Analysis (LDA) which maximizes the distribution ratio can be used.

Accordingly, the virtual face image data generation unit 150 can search virtual face images VF1, VF2, and VF3 to be inserted into the face region of the image data.

As described above, the virtual face feature value generating unit 130 generates the virtual face feature value by setting the similarity value preset at the time of generating the noise value to 70%, which is a comparatively high value, ) Search for similar virtual face images (VF1, VF2, VF3), although public information (race, age, sex) can not identify personal information identifiable from the original face image within the face region of the image data.

7, the virtual face image data generation unit 150 generates virtual face image data by inserting the retrieved virtual face images VF1, VF2, and VF3 in the face region of the image data can do.

In addition, the virtual face image data generation unit 150 may store the virtual face image data in which the virtual face images VF1, VF2, and VF3 are embedded, while preserving the previously allocated imaging sequence.

Then, the facial feature value restoring unit 160, which will be described later, can restore the virtual face image data by using the captured image held in the virtual face image data.

As described above, the virtual face feature value generation unit 130 generates a virtual random number in accordance with the imaging sequence assigned to each of the image data input from each of the cameras C1 to C3, generates a virtual random number using the generated virtual random number, May be generated for each image data to generate virtual face feature values.

Accordingly, the virtual random number applied to each image data is different, and the virtual face feature value generated by using the noise value, the virtual face feature value generated by using the noise value, and the virtual face feature vector obtained by using the virtual face feature value, And may be different for each data.

Further, the virtual face image data generator 150 searches the virtual face images VF1, VF2, and VF3 based on virtual face feature vectors that are different for each image data, thereby generating virtual face images VF1, VF2, VF3) may also be different for each image data.

For example, the virtual face images applied to the respective image data photographed from the cameras C1, C2, and C3 may be different by VF1, VF2, and VF3, respectively.

7, the virtual face image data generator 150 inserts the virtual face image VF1 into the face area 10 of the image data VD1 photographed from the camera C1 to generate virtual face image data VD1 ' Can be generated.

8, the virtual face image data generation unit 150 inserts the virtual face image VF2 into the face region 10 of the image data VD2 photographed from the camera C2 to generate virtual face image data VD2 ' Can be generated.

10, the virtual face image data generation unit 150 inserts the virtual face image VF3 in the face region 10 of the image data VD3 photographed from the camera C3 to generate virtual face image data VD3 ' Can be generated.

When the virtual face images VF1, VF2, and VF3 inserted in the virtual face image data are compared, the virtual face images VF1, VF2, and VF3 all have the same similarity as the original face image data of the image data, Are not the same face type.

On the other hand, when the virtual face feature value generation unit 130 generates a virtual face feature value by setting the similarity value preset at the noise value generation to a relatively low value of 10%, the virtual face image data generation unit 150 generates A virtual face image VF 'which is not similar to the original face image is inserted into the face area 10 of the image data VD1 photographed from the camera C1, Virtual face image data VD1 ".

Meanwhile, the face area detection unit 110 according to another embodiment of the present invention is partially similar in structure and role to the face area detection unit 110 according to an embodiment of the present invention. However, when detecting a face area on the image data, The same photographing sequence can be allotted to all the image data regardless of the detection order of the area.

For example, the face region detection unit 110 first detects the face region from the image data photographed from the camera C1, detects the face region from the image data photographed from the cameras C2 and C3 for the second and third times, The same shooting sequence number 1 can be assigned to the data.

Accordingly, the virtual facial feature value generation unit 130 generates the virtual random number applied to all the image data by executing the random number generation function only once, thereby generating the same virtual random number.

Further, as the virtualization random numbers applied to all the image data are the same, the virtual face feature value generated using the virtualized random number, the virtual face feature vector having the virtual face feature value vectorized, and the virtual face Images may be the same for all the image data.

11 and 12 are views showing images of virtual face image data in which a virtual face image generated according to another embodiment of the present invention is inserted.

11 and 12, the face area detection unit 110 according to another embodiment assigns a photographing rotation number to all the image data so that the image data can be applied to the respective image data photographed from the cameras C1, C2, and C3 The virtual face images may be identical to VF1, respectively.

That is, as the virtual face image data VD1 'is generated by inserting the virtual face image VF1 into the face area 10 of the image data VD1 photographed from the camera C1, the virtual face image data generating unit 150 generates the virtual face image data VD1' Virtual face image data VD2 ", VD3 "can be generated by inserting the virtual face image VF1 into the face region 10 of the image data VD2 and VD3 photographed from the virtual face image data VD2 and VD3 respectively.

Next, the process of generating restored image data by restoring the virtual face image inserted in the face area of the virtual face image data will be described.

The face region detection unit 110 may detect the face region on the virtual face image data.

The method of detecting the face area on the virtual face image data by the face area detecting unit 110 is the same as the method of detecting the face area on the above-described image data, and therefore, the repeated description will be omitted.

Next, the facial feature value generation unit 120 may generate a virtual facial feature value in the face region on the virtual camera image data detected by the facial region detection unit 110. FIG.

Here, the virtual face feature value may be a value generated by adding the noise value to the face feature value from the virtual face feature value generation unit 130 described above.

Since the method of generating the virtual face feature value in the face region of the virtual image data is the same as the method of generating the face feature value in the face region of the image data described above, Omit it.

The facial feature value restoring unit 160 may generate a restored random number by using a preset seed value and a predetermined virtual random number generation range.

More specifically, the facial feature value restoring unit 160 receives a predetermined seed value used in the process of generating the virtual image data to be restored, and a predetermined virtual random number generation range at the same time, And generate a restored random number.

At this time, the facial feature value restoring unit 160 may generate a restored random number by executing a random number generating function as much as the photographing number assigned to the virtual image data to be restored.

That is, the facial feature value restoring unit 160 may generate the virtual random number used in the process of generating the virtual image data backward as the restored random number.

Accordingly, the restored random number generated from the facial feature value restoring unit 160 may be the same as the virtual random number used in the process of generating the virtual image data.

At this time, the facial feature value restoring unit 160 may generate restored random numbers using the following equation (4).

&Quot; (4) "

Here, P is a random number generating function, i is the issuing number of the random number generating function, and S is a preset seed value.

On the other hand, the facial feature value restoring unit 160 may reduce the noise value by adding the deviation of the restored random number to the predetermined similarity value used in the process of generating the virtual image data to be restored, And a restored face feature value is generated by adding the value of the restored face feature value.

That is, the facial feature value restoring unit 160 reconstructs the facial feature value, which is the same as the face feature value obtained from the face image of the image data before the virtualization is performed, using the preset similarity value used in the process of generating the virtual image data Can be generated.

At this time, the facial feature value restoring unit 160 may generate a noise value that is inversely calculated using the following equation (5).

Equation (5)

Herein, N '(i) is the inversed noise value, T is the preset similarity value, P' (i) S is the restored random number, and R1r is the predetermined virtual random number generation range.

The reconstructed facial feature vector generator 170 may perform vectorization of reconstructed facial feature values generated from the facial feature reconstructor 160 to generate reconstructed facial feature vectors.

The restored facial feature vector generator 170 also generates restored facial feature vectors by vectorizing the restored facial feature values. The virtual facial feature vector generator 140 performs vectorization of the virtual facial feature values, The same method as the method of generating the vector, so that the repetitive description will be omitted.

13 is a diagram illustrating an image of restored image data in which a restored face image generated according to an embodiment of the present invention is inserted.

13, the facial image data restoring unit 180 restores facial image (FF of FIG. 6) included in the facial image database based on the restored facial feature vector generated from the restored facial feature vector generating unit 170, The restored face image VF4 may be searched.

The facial image data restoring unit 180 may generate the restored face image VF4 based on the restored face feature vector generated by the facial feature vector generating unit 170 according to another embodiment.

The facial image data restoring unit 180 may generate restored image data D1 by inserting the restored face image VF4 into the face region of the virtual face image data VD1 '.

Accordingly, the facial image virtualization apparatus 100 according to the present invention generates virtual face image data VD1 'by inserting a virtual face image into the face region of the image data, It is possible not only to virtualize the face image to include public information such as age, but also to restore the virtualized virtual face image data VD1 'into restored image data D1 of original shape.

Herein, in restoring the reconstructed image data D1, the preset seed value used for generating the virtual face image data VD1 ', the predetermined virtual random number generation range, It is provided only to authorized administrators, thereby improving security.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, But the present invention is not limited thereto.

100: face image virtualization apparatus 110: face region detection unit
120: facial feature value generation unit 130: virtual face feature value generation unit
140: virtual face feature vector generation unit 150: virtual face image data generation unit
160: facial feature value restoring unit 170: restored facial feature vector generating unit
180: facial image data restoration unit

Claims (8)

A face area detecting unit for detecting a face area from at least one image data on which a target person is photographed and allocating a photographed image for each of the at least one image data corresponding to a sequence in which the at least one image data is photographed;
A virtual face feature extracting unit for generating a virtual random number for each of the at least one image data by an occurrence order corresponding to the shooting sequence and converting the extracted facial feature value into the virtual face feature value using the generated virtual random number, Value generating unit and
A virtual face image is searched in the face image database based on the virtual face feature vector in which the virtual face feature value is vectorized and the virtual face image is inserted into the face area of the image data to generate virtual face image data And a virtual face image data generation unit,
Wherein the virtual face feature value generating unit comprises:
Generating a virtual face feature value by adding a deviation of the virtual random number to a preset similarity value, generating a noise value using the following equation, adding the noise value to the face feature value,
≪ Equation &

Here, N (i) is a noise value, T is a preset similarity value, P (i) S is a virtual random number, and R1r is a predetermined virtual random number generation range.
Wherein the noise value is generated in inverse proportion to a preset similarity value, and a larger value is generated as a preset similarity value is set to a smaller value.
The method according to claim 1,
The virtual face feature value generation unit
Sets a seed value that is set in advance as an initial value of a random number generation function and executes the random number generation function by the generation order in the shooting order, thereby generating the virtual random number.
delete delete The method according to claim 1,
A facial feature value restoring unit configured to set the predetermined seed value as an initial value of the restored random number generating function and to execute the restored random number issuing function by an occurrence order corresponding to the photographing sequence assigned to the virtual face image data, A facial image virtualization apparatus comprising:
6. The method of claim 5,
The facial feature value restoring unit
Wherein the face feature value is obtained by adding the deviation of the restored random number to the preset similarity value to inverse the noise value used to generate the virtual face feature value, subtracting the inversed noise value from the virtual face feature value, The facial image virtualization apparatus generates restored restored facial feature value.
The method according to claim 6,
The facial feature value restoring unit
And the inverse of the noise value used to generate the virtual facial feature value using the following equation.

≪ Equation &

Herein, N '(i) is the inversed noise value, T is the preset similarity value, R2 is the restored random number, and R1r is the predetermined virtual random number generation range.
The method according to claim 6,
A restored face image is retrieved from the face image database based on the restored face feature vector having the restored face feature value vectorized and the retrieved restored face image is inserted into the face area of the virtual face image data, And a face image data restoring unit for generating restored image data.

Images