JP5489340B2 - Face masking apparatus and method - Google Patents

Description

  The present invention relates to a face masking apparatus and method suitable for use in a surveillance camera system.

  In the surveillance camera system, the face area in the image is made invisible for privacy protection. A masking technique is generally used to make a face invisible. As conventional masking techniques, for example, those disclosed in Patent Documents 1 to 4 are known. Among them, Patent Documents 1 and 2 disclose a method for specifying a human face area by detecting a skin color area from a color image. Patent Documents 3 and 4 disclose a method for masking a face by performing face detection or pattern matching. Further, as a technique for detecting a face in Patent Document 2, there is a face detection method using a discriminator using the Adaboost learning method shown in Non-Patent Documents 1 to 3.

Since the position and size of the face in the image vary, it is common to perform the following processing to actually detect the face in the image.
In order to match the size of the image window and the actual face, the image is scaled stepwise and face detection is performed at each location in the scaled image. Further, the final face area is specified by integrating areas (face candidate areas) determined to have a face in each scaled image (face candidate area processing).

  FIG. 11 is a diagram for explaining a conventional face candidate area integration process. In the figure, an image 1101 is an image to be processed for face detection. A face candidate area 1102 in the image 1101 is an area determined to have a face by face detection processing. Further, the face area 1103 is an area that is integrated by the face candidate area integration process and detected as a final face area. When the image 1101 is scaled step by step and face detection is performed at each location in the scaled image, one or more face candidate areas 1102 are detected around the actual face area with approximately the same size as the actual face area size. The In the face candidate area integration process, face candidate areas 1102 whose center positions are close to each other are determined as the same group. Then, the average position and average size of the face candidate areas 1102 of the same group are obtained, and these are set as the position and size of the face area 1103 detected as the final face area.

Japanese Patent Laid-Open No. 11-146405 JP 2008-2527779 A JP 2008-160354 A JP 2001-086407 A

Bernhard Froba, Andreas Ernst, `` Face Detection with the Modified Census Transform '', Proceedings for Sixth IEEE International Conference on Automatic Face and Gesture Recognition (AFGR), May 2004, p.91-96 Yoav Freund, Robert E. Schapire, `` A decision-theoretic generalization of on-line learning and anapplication to boosting '', Computational Learing Theory Eurocolt'95, Springer Verlag, 1995, p.23-37 Paul Viola, Michael Jones, `` Rapid Object Detection Using a Boosted Cascade of Simple Features '', IEEE Computer Society Conference on Computer Virsion and Pattern Recognition (CVPR), December 2001, ISSN: 1063-6919, Vol.1, p. 511-518

However, the conventional techniques disclosed in the above-mentioned prior art documents have the following problems.
(1) In the techniques disclosed in Patent Documents 1 and 2, excessive masking is performed including parts other than the face (for example, arms and hands). That is, since the skin color area is not necessarily the face area as it is, the arms and hands other than the face are masked.

(2) With the techniques disclosed in Patent Documents 3 and 4, when the entire face is not exposed, such as when a part of the face is hidden, the face cannot be detected and the face is not masked. That is, as shown in FIG. 12, when a part of the face of the person 1203 is concealed by the person 1202 in the image 1201, as shown in FIG. 13, detected feature amount blocks 1301 to 1303 for the person 1203. Since part of the feature amount blocks 1302 and 1303 are covered with the image information of the person 1202, the face detection score value decreases, and as a result, it is determined that there is no face around the face area of the person 1203. This problem arises.

  In addition, when a plurality of faces are close to each other, the face area may be detected deviating from the actual face position, and there is a problem that only a part of the face is masked. As shown in FIG. 14, when the faces of the person 1402 and the person 1403 are close to each other in the image 1401, the face candidate areas 1404 for the faces of each other are close to each other, so that they are determined as the same group. As a result of the integration processing, the position of the face candidate area 1404 detected as the final face area is an intermediate position between the person 1402 and the person 1403, and the size of the face candidate area 1404 is the average of the face sizes of the person 1402 and the person 1403. Therefore, this problem occurs. The above problems occur with high probability in crowds where there are many people.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a face masking apparatus and method that can perform appropriate masking even in crowds where there are many people.

The face masking apparatus according to the present invention includes an image window extracting unit that extracts an image window from an input image, and a score calculating unit that calculates a score value indicating the likelihood of the face in the image window extracted by the image window extracting unit. The score value calculated by the score calculation means is compared with a predetermined first threshold, and a face detection area setting means for setting a face detection area according to the comparison result, and at least two face detection areas are set. A face group area setting for comparing a distance between the face detection areas set by the face detection area setting means with a predetermined second threshold and setting a face group area including the face detection area according to a comparison result and means, and when masking the face detection area in response to the score value of the face detection region, to determine the necessity of masking a different threshold and the face group region and the region outside .

  According to the above configuration, since the face area is specified based on the score value indicating the likelihood of the face, it is possible to suppress excessive masking of the portion below the face. In addition, since a range including a region (one or a plurality of cases) where the score value is equal to or greater than a predetermined second threshold is set as the face group region range, it occurs with high probability due to crowds, It becomes possible to suppress masking mistakes caused by concealment of part of a person's face and proximity of faces. In this way, appropriate masking can be performed even in crowds where there are many people.

  In the above configuration, a masking unit for masking the face group region set by the face group region setting unit is provided.

  According to the above configuration, only the face group region can be masked.

  The said structure WHEREIN: The said masking means determines the necessity of masking of the said face detection area in the said face group area | region according to the said score value.

  According to the above configuration, for example, an area where the score value is high and the presence of a face is high is masked, and masking is not performed where it is not so much. Thereby, excessive masking can be prevented.

  The said structure WHEREIN: The said masking means determines the degree of masking of the said face detection area in the said face group area | region according to the said score value.

  According to the above-described configuration, for example, an area where the score value is high and the presence of the face is high makes the masking invisible degree high, and the area where the score value is low is made low. Thereby, excessive masking can be prevented.

In the above structure, with a second masking means for determining the necessity of masking different thresholds as pre Kikao group region and the region outside.

  According to the above configuration, a face that is partly concealed in the face group area can be detected and masked as a face area, and a non-face area is not masked outside the face group area. Masking can be performed.

According to the face masking method of the present invention, an image window extracting step for extracting at least two image windows from an input image, and a score for calculating a score value indicating the likelihood of the face in the image window extracted in the image window extracting step. Setting in the calculation step, a face detection region setting step for comparing the score value calculated in the score calculation step with a predetermined first threshold value and setting a face detection region according to the comparison result, and the face detection region setting step It has been the distance between the face detection area in accordance with the comparison result of the comparison with a predetermined second threshold value and a face group region setting step of setting a face group area including the face detection area, the face detection area When masking the face detection area according to the score value, it is determined whether masking is necessary or not based on different threshold values for the inside and outside of the face group area .

  According to the above method, since the face area is specified based on the score value indicating the likelihood of the face, it is possible to suppress excessive masking of the portion below the face. In addition, since a range including a region (one or a plurality of cases) where the score value is equal to or greater than a predetermined second threshold is set as the face group region range, it occurs with high probability due to crowds, It becomes possible to suppress masking mistakes caused by concealment of part of a person's face and proximity of faces. In this way, appropriate masking can be performed even in crowds where there are many people.

  According to the present invention, it is possible to perform appropriate masking even in crowds where many people exist.

1 is a block diagram showing a schematic configuration of a face masking apparatus according to Embodiment 1 of the present invention. The figure which shows an example of the image window extracted by the image window extraction part of the face masking apparatus of FIG. The block diagram which shows schematic structure of the score calculation part of the face masking apparatus of FIG. The figure which shows an example of the learning data used with the face masking apparatus of FIG. The figure which shows an example of the image which the face masking apparatus of FIG. 1 inputs The figure for demonstrating the score value distribution with respect to the input image in the face masking apparatus of FIG. 1 is a flowchart showing a face masking procedure in the face masking apparatus of FIG. The block diagram which shows schematic structure of the face masking apparatus which concerns on Embodiment 2 of this invention. The figure which shows the structure of the surveillance camera system using the face masking apparatus shown in FIG. 1 or FIG. The figure which shows the user interface used for the computer of the surveillance camera system of FIG. The figure for demonstrating the face candidate area | region integration process in a prior art The figure which shows an example of the image where a part of face is concealed The figure for demonstrating the state of the feature-value block in the image window in the example of an image where a part of face is concealed The figure for demonstrating the face candidate area | region integration process result in the example of an image with the face in the prior art, and the image example

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments for carrying out the invention will be described in detail with reference to the drawings.

(Embodiment 1)
FIG. 1 is a block diagram showing a schematic configuration of a face masking apparatus according to Embodiment 1 of the present invention. In the figure, a face masking apparatus 100 according to the present embodiment includes an image input unit 101, a scaled image generation unit 102, an image window extraction unit (image window extraction unit) 103, and a score calculation unit (score calculation unit) 104. And a comparator (face detection area setting means) 105, a face group area determination section (face group area setting means) 106, and a masking section (masking means) 107.

  The image input unit 101 inputs and holds an image obtained by photographing with a camera (not shown). The scaled image generation unit 102 scales and holds the image input by the image input unit 101 using a method such as general sub-sampling, area interpolation, or affine transformation. The image window extraction unit 103 extracts at least two image windows from the input image scaled by the scaling image generation unit 102. The image window extracted by the image window extraction unit 103 is an area to be subjected to face detection processing in the image input by the image input unit 101. FIG. 2 is a diagram illustrating an example of an image window extracted by the image window extraction unit 103.

  The score calculation unit 104 calculates a score value indicating the likelihood of the face in the image window extracted by the image window extraction unit 103. FIG. 3 is a block diagram illustrating a schematic configuration of the score calculation unit 104. In the figure, the score calculation unit 104 includes three discriminators 311 to 313, and for example, calculates score values of the feature amount blocks 211 to 213 in the image window 201 shown in FIG.

  Each feature amount block 211 to 213 in the image window 201 is binarized. This binarization compares the luminance value of the pixel in the block with the average luminance value in the block, and labels “1” if the luminance value of the pixel is higher than the average value, otherwise labeled “0”. It is realized by doing. A score value is determined by the AdaBoost learning method for each of the feature amount blocks 211 to 213 that have been binarized. FIG. 4 shows an example of learning data generated by the AdaBoost learning method.

  The discriminator 311 includes learning data 311-1 and an adder 311-2. The discriminator 312 includes learning data 312-1 and an adder 312-2. The discriminator 313 includes learning data 313-1 and an adder 313-2. A score value of “0” and the feature amount block 211 are input to the discriminator 311, the score value obtained by the discriminator 311 and the feature amount block 212 are input to the discriminator 311, and the discriminator 313 The score value obtained in 312 and the feature amount block 213 are input.

  In the discriminator 311, the learning data 311-1 converts the binarized pattern of the feature block 211 into a score value, and the adder 311-2 sets a score value of “0” to the score value from the learning data 311-1. Add and output. In the discriminator 312, the learning data 312-1 converts the binarized pattern of the feature amount block 212 into a score value, and the adder 312-2 converts the score value from the learning data 312-1 into the score value from the discriminator 311. Are added and output. In the discriminator 313, the learning data 313-1 converts the binarized pattern of the feature amount block 213 into a score, and the adder 313-2 converts the score value from the discriminator 312 into the score value from the learning data 313-1. Add and output. A final score value is output from the discriminator 313.

  The learning data of each of the feature amount blocks 211 to 213 has such a learning value that the score value is large for the binarized pattern for the face image and conversely the score value is small for the binarized pattern for the non-face image. Data is generated by the Adaboost learning method. By generating learning data in this way, when a face exists in the image window 201, the final score value that is the sum of the score values calculated by the discriminators 311 to 313 becomes a large value.

  Returning to FIG. 1, the comparator (face detection area setting means) 105 compares the score value calculated by the score calculation unit 104 with a score threshold A (predetermined first threshold), and determines the face according to the comparison result. Set the detection area. In this case, all face detection areas exceeding the score threshold A are set. A face group area determination unit (face group area setting means) 106 compares the distance between these in at least two face detection areas set by the comparator 105 with a distance threshold A (predetermined second threshold), A face group area including a face detection area is set according to the comparison result. The face group area determination unit 106 classifies face detection areas whose distance is within the distance threshold A and whose score value is equal to or greater than the score threshold A into the same group, and further includes each face detection area of the same group. The area to be set is set as the position and size of the face group area range. The masking unit 107 masks the face group region range output from the face group region determination unit 106 on the image held by the image input unit 101 to generate a face masking image.

  FIG. 5 is a diagram illustrating an example of an image input by the image input unit 101. The image 501 includes a person 502, a person 503, and a person 504. A part of the face of the person 503 is hidden by the person 502, and the faces of the person 503 and the person 504 are close to each other. This is a scene often seen in crowds where there are many people. FIG. 6 is a diagram showing the distribution of scores output by the score calculation unit 104 for the example of the image shown in FIG. As shown in the figure, in the score value distribution image 601, the black (darker) color indicates that the score value of that region is higher. The face group area range 602 is output from the face group area determination unit 106.

  Next, the operation of the face masking apparatus 100 of this embodiment will be described. FIG. 7 is a flowchart for explaining the operation of face masking apparatus 100 of the present embodiment. In the figure, first, the image input unit 101 inputs an image output from a camera (not shown) (step S1). Next, the scaled image generation unit 102 scales the image input by the image input unit 101 (step S2). Next, the image window extraction unit 103 extracts an image window from the scaled input image (step S3). The image window extracted by the image window extraction unit 103 is an area to be subjected to face detection processing in the image input by the image input unit 101.

  After extracting the image window, the score calculation unit 104 calculates a score value indicating the likelihood of the face in the image window (step S4). After calculating the score value in the image window, the comparator 105 compares the score value calculated by the score calculation unit 104 with the score threshold A, records an area where the score value is greater than or equal to the score threshold A, and determines the face group area The unit 106 is notified (step S5).

  Further, the comparator 105 determines whether or not the processing from step S3 to step S5 has been completed at all image window positions in the scaled image (step S6). If there is an image window that has not been processed, step S3 is performed. Returning to FIG. On the other hand, when the processing from step S3 to step S5 is completed at all the image window positions in the scaled image, it is determined whether the processing at all the scaling stages has been completed (step S7). If the processing at all the scaling stages has not been completed, the process returns to step S2. If the processing at all the scaling stages has been completed, the face group region determination unit 106 determines that the score output from the score calculation unit 104 is the score. If the distance between one or more areas above the threshold A is within the distance threshold A, they are classified into the same group (step S8). Then, for each group, a region that includes (includes) a region belonging to the group is set as the position and size of the face group region range (step S9).

  For example, when the image 501 shown in FIG. 5 is input to the face masking apparatus 100 of the present embodiment, the score calculation unit 104 outputs the score value shown in the score value distribution image 601 in FIG. Since the areas of the score values are close to each other and are equal to or smaller than the distance threshold A, the face group area determination unit 106 classifies them into the same group. Further, since the face group region determination unit 106 outputs a range that includes (includes) the regions belonging to this group, the face group region determination unit 106 outputs the face group region range 602 in FIG.

  The masking unit 107 performs masking processing on the face group region range output by the face group region determination unit 106 on the image held by the image input unit 101 (step S10). As a result, the face group area range in FIG. 6 output by the face group area determination unit 106 is masked.

  As described above, according to the face masking apparatus 100 of the present embodiment, the comparator 105 and the face group region determination unit 106 include a range in which one or a plurality of regions having a score value equal to or higher than the score threshold A are included (included). In order to determine the face group region range 602, masking mistakes that have occurred in the prior art even when a part of the face of the person 503 is hidden and the faces are close to each other as shown in FIGS. Can be suppressed, and the face area can be masked accurately.

  The masking method performed by the masking unit 107 may be any method as long as the face is invisible. For example, if a method of blurring the face region using a low-frequency pass filter or the like is used, the degree of masking can be controlled by adjusting the low-frequency range that passes.

  Further, the distance threshold A used in the face group region determination unit 106 may be a value obtained by multiplying the average value of the size of the region of the score value A or more input to the face group region determination unit 106 by a coefficient. Since the size of the area greater than or equal to the score value A input to the face group area determination unit 106 is substantially equal to the actual face size of the person, an appropriate distance threshold value linked to the actual face size is set thereby. be able to.

  In addition, the face group region range is a region that includes (includes) a plurality of regions each having a score threshold A or higher and a distance threshold A or lower. In FIG. 6, the face group region range 602 is shown as a rectangular region. Needless to say, it is only necessary to include one region to be included (included), and it is needless to say that the same effect can be obtained even in an elliptical region or a region that is not a white region in the face group region range in FIG.

  Further, the masking degree in the face group area may be adjusted according to the score value in the same area. An area where the score value is high and the presence of the face is high increases the degree of masking invisibility, and the area where the score is not high decreases the degree of invisibility. Thereby, excessive masking can be prevented.

  Further, the necessity of masking of the face detection area in the face group area may be determined according to the score value. Masking is performed on areas where the score value is high and the presence of a face is high, and masking is not performed on areas where the score value is not so high.

(Embodiment 2)
FIG. 8 is a block diagram showing a schematic configuration of the face masking apparatus according to Embodiment 2 of the present invention. In the figure, the face masking apparatus 800 of the present embodiment has the same configuration as the face masking apparatus 100 of the first embodiment described above (image input unit 101, scaling image generation unit 102, image window extraction unit 103, score calculation). In addition to the unit 104, the comparator 105, the face group region determination unit 106, and the masking unit 107), a threshold selection unit 108, a comparator 109, and a face candidate region integration unit 110 are provided.

  The threshold selection unit 108 selects the score threshold B if it is outside the one or more face group region ranges output by the face group region determination unit 106, and selects the score threshold C if it is within the face group region range. The score threshold C is set to a value smaller than the score threshold A used in the comparator 105. The score threshold C is set to a value smaller than the score threshold B.

  The comparator 109 compares the score threshold B or the score threshold C selected by the threshold selection unit 108 with the score value output by the score calculation unit 104. As a result, a face candidate area (an image window whose score is equal to or greater than the score threshold) for which a score equal to or greater than the score threshold C is calculated if it is outside the face group area range, and if it is within the face group area range, Will be output. The face candidate area integration unit 110 groups one or a plurality of face candidate areas (image windows whose scores are equal to or higher than the score threshold) by the distance threshold B, and finally determines the average position and average size of the face candidate areas for each group. As a typical face area.

  The masking unit 107 performs a masking process on the face area output from the face candidate area integration unit 110 on the image held by the image input unit 101. In this case, the face area output by the face candidate area integration unit 110 is also masked. The presence of a face is high in the face group area range, and even if the score threshold is lowered, the possibility of erroneous face detection is low, and a face that is partly concealed can be detected as a face area. it can. The face area detected in this way has a narrower range than the face group area range and includes a concealed face in the area, so that excessive masking can be suppressed.

  Thus, according to face masking apparatus 800 of the present embodiment, if threshold selection unit 108 is within the face group region range, score threshold C that is smaller than score threshold A is selected, and comparator 109 and face are selected. Since the candidate area integration unit 110 integrates face candidate areas whose score is equal to or greater than the score threshold C and within the distance threshold B, a face that is partly concealed can be detected as a face area, and mask leakage Can be avoided. In addition, since the possibility of the presence of a face is low outside the face group area compared to the inside, using a score threshold value B higher than the score threshold value C does not mask a non-face area, and excessive masking is also possible. Can be suppressed.

  Note that the masking unit 107 may perform masking processing on the face group area output by the face group area determination unit 106 in addition to the face area output by the face candidate area integration unit 110 as in the first embodiment. good. In this case, since it is more certain that the face area is an area where the face exists, the degree of masking may be higher than the face group area. The masking degree is a degree of making the face invisible.

  Further, the distance threshold A used in the face group region determination unit 106 is a value obtained by multiplying the size of the region input to the face candidate region integration unit 110 by a factor equal to or greater than the score threshold B (not greater than the score threshold C). Also good. Outside the face group area, the size of the area that is greater than or equal to the score threshold B input to the face candidate area integration unit 110 is almost the same as the actual face size of the person. A simple distance threshold can be set.

  The score threshold A used in the comparator 105 is 1 or less to the score threshold B (not higher than the score threshold C) or the average score value input to the face candidate region integration unit 110 outside the face group region 0 A value obtained by multiplying the above coefficients may be used. This is because an area with a low score value that is not determined as a face area due to a part of the face being concealed is included in the face group area range. Thereby, an appropriate score threshold linked to the score threshold A and the score of the actual face region can be set.

(Embodiment 3)
FIG. 9 is a configuration diagram of a surveillance camera system 900 using the face masking device 100 according to the first embodiment or the face masking device 800 according to the second embodiment. The surveillance camera system 900 includes an imaging device 901, a computer (so-called personal computer) 902, a storage device (semiconductor memory, magnetic disk device, etc.) 903 connected to the computer 902, and a display device 904. . The imaging device 901, the computer 902, and the display device 904 are connected through a transmission system 905. The transmission system 905 is a communication network such as a wireless communication network or a wired communication network such as the Internet.

  An image output from the imaging device 901 is transmitted to the computer 902 through the transmission system 905 and masked by the computer 902. The masked image, that is, the masked image is transmitted to the display device 904 through the transmission system 905 and displayed on the display device 904. The storage device 903 is used for holding an unprocessed image and an intermediate result, and recording.

  FIG. 10 is a diagram showing a user interface 1000 used in the computer 902. In the figure, an input image window 1001 is a window for displaying an image output from the imaging device 901. The masking image window 1002 is a window that displays an image after masking processing for the input image. A masking range 1003 is set in the masking image window 1002. The threshold adjustment slider 1004 sets various threshold values (score threshold values A and B, distance threshold values A and B). Various threshold values are set by adjusting the slider position with a mouse drag operation or cursor keys.

  The threshold adjustment option input 1005 is used to input an option for threshold adjustment, and inputs ON / OFF of a threshold option item and a numerical value by a check button or numerical value input. In this case, ON / OFF or numerical value input is performed by mouse or key input. The mask degree graph input 1006 is a graph-type input unit that adjusts the mask degree with respect to the score value. The mask degree with respect to the score value can be adjusted by changing the shape of the graph by clicking and dragging the mouse.

  An input image is displayed in the input image window 1001. Also, the masking image window 1002 displays the result of masking processing at various threshold values and mask degrees that are currently set by a threshold adjustment slider 1004, a threshold adjustment option input 1005, and a mask degree graph input 1006. The user performs fine adjustment so that an optimal masking process can be obtained with the threshold adjustment slider 1004, the threshold adjustment option input 1005, and the mask degree graph input 1006 while viewing the image of the masking process result. In addition, it is confirmed that excessive masking is not performed in comparison with the input image. With the user interface as described above, the burden on the user can be reduced and masking errors can be easily checked and finely adjusted.

  The present invention has an effect that always appropriate masking can be performed, and can be applied to a system that distributes video information such as monitoring, broadcasting, and the Internet that requires privacy protection.

DESCRIPTION OF SYMBOLS 100,800 Face masking apparatus 101 Image input part 102 Scaling image generation part 103 Image window extraction part 104 Score calculation part 105, 109 Comparator 106 Face group area | region determination part 107 Masking part 108 Threshold selection part 110 Face candidate area | region integration part 311- 313 Discriminator 900 Monitoring camera system 901 Imaging device 902 Computer 903 Storage device 904 Display device 905 Transmission system 1000 User interface

Claims (6)

Image window extraction means for extracting an image window from the input image;
Score calculating means for calculating a score value indicating the likelihood of the face in the image window extracted by the image window extracting means;
A face detection area setting means for comparing the score value calculated by the score calculation means with a predetermined first threshold and setting a face detection area according to the comparison result;
When at least two face detection areas are set, the distance between the face detection areas set by the face detection area setting means is compared with a predetermined second threshold, and the face detection areas are set according to the comparison result. Face group region setting means for setting a face group region including;
Equipped with a,
A face masking apparatus , wherein when masking the face detection area according to the score value of the face detection area, it is determined whether masking is necessary or not based on different threshold values in and outside the face group area .   2. The face masking apparatus according to claim 1, further comprising masking means for masking the face group area set by the face group area setting means.   3. The face masking apparatus according to claim 2, wherein the masking unit determines whether or not the face detection area in the face group area needs to be masked according to the score value.   The face masking apparatus according to claim 2, wherein the masking unit determines a masking degree of the face detection area in the face group area according to the score value. Face mask according to claim 1, characterized in that a second masking means for determining the necessity of masking with different thresholds and front Kikao group region and the region outside. An image window extraction step for extracting at least two image windows from the input image;
A score calculating step for calculating a score value indicating the likelihood of the face in the image window extracted in the image window extracting step;
A face detection area setting step of comparing the score value calculated in the score calculation step with a predetermined first threshold and setting a face detection area according to the comparison result;
A face group region setting step of comparing a distance between the face detection regions set in the face detection region setting step with a predetermined second threshold and setting a face group region including the face detection region according to a comparison result;
Equipped with a,
A face masking method , wherein when masking the face detection area according to the score value of the face detection area, it is determined whether masking is necessary or not based on different threshold values inside and outside the face group area .

Images