KR20120077422A - Apparatus and method for determining symmetry of face - Google Patents

Abstract

According to an aspect of the present invention, a method for calculating a left-right symmetry of a face by detecting a face from an image including a face includes reducing a region of interest (ROI) image including a face area. Generating a first reduced image; Calculating a symmetry degree between a left region and a right region on the basis of a preset symmetry axis for the face region of the first reduced image; Recalculating the symmetry with respect to the face region of the second reduced image in which the first reduced image is rotated or moved by a predetermined value based on the symmetry axis; And determining the left and right symmetry of the face by updating the previously calculated symmetry to the recalculated symmetry if the recalculated symmetry is greater than the previously calculated symmetry.

Description

Apparatus and Method for Determining Symmetry of Face}

The present invention relates to face recognition, and more particularly, to a method and an apparatus capable of calculating the symmetry of a face.

Cameras (PC cameras, mobile phone cameras, CCD cameras, etc.) and the development of image processing technology in accordance with the radical development of the detection of the face from the image input from the camera, the program to be used for games or security systems using the detected face Are being developed.

In particular, as a mobile terminal equipped with a high-performance camera such as a smartphone or a tablet PC has recently appeared, images can be taken using the portable terminal, and various functions can be performed using a face detected from the captured image. Some applications are being developed.

Meanwhile, as interest in beauty has increased recently, interest in the symmetry of the face (india) has increased, but the existing programs and applications described above were mainly focused on face recognition. The symmetry of the face could not be determined using the face.

Therefore, conventionally, the position and size of major parts of the face, such as eyes, nose, mouth, or eyebrows, are measured by placing the curved special person, which is scaled to determine the symmetry of the face, on the actual face to be judged. By doing so, the left and right symmetry of the face was inevitably calculated.

As described above, in determining the left and right symmetry of the face, since the right and left symmetry of the face may vary according to the physical contact position of the special person and the visual measurement error, the accuracy of the calculated left and right symmetry is guaranteed. The problem is that you can't.

Disclosure of Invention The present invention has been made in view of the above-described problems, and it is an object of the present invention to provide a method and apparatus for calculating a left-right symmetry of a face that can detect a face from an image including the face and calculate a left-right symmetry of the face. .

Another object of the present invention is to provide a method and apparatus for calculating a left and right symmetry of a face capable of calculating the left and right symmetry of a face included in an image that is not aligned based on the symmetry axis.

According to an aspect of the present invention, there is provided a method of calculating a left and right symmetry of a face, the method including: generating a first reduced image by reducing a region of interest (ROI) image including a face region; Calculating a symmetry degree between a left region and a right region on the basis of a preset symmetry axis for the face region of the first reduced image; Recalculating the symmetry with respect to the face region of the second reduced image in which the first reduced image is rotated or moved by a predetermined value based on the symmetry axis; And determining the left and right symmetry of the face by updating the previously calculated symmetry to the recalculated symmetry if the recalculated symmetry is greater than the previously calculated symmetry.

According to another aspect of the present invention, there is provided a method of calculating a symmetry of a face, the method including: generating a first reduced image by reducing an ROI image including a face region; And calculating a symmetry degree between a left region and a right region based on a symmetry axis preset for a face region of the second reduced image in which the first reduced image is rotated or moved by a predetermined value, and calculates the symmetry degree based on the left and right sides of the face. Determining the degree of symmetry.

According to still another aspect of the present invention, there is provided an apparatus for calculating left and right symmetry of a face, including: an image preprocessor configured to preprocess a still image; A ROI image generator configured to generate a ROI image including a face region from the preprocessed still image; An image converter for generating a first reduced image by reducing the ROI image by using a Gaussian pyramid resolution reduction method and generating a second reduced image by rotating or moving the ROI image by a predetermined value; And calculating a symmetry degree between a left region and a right region based on a predetermined symmetry axis with respect to the face region of the first reduced image, and then regenerating the symmetry degree based on the symmetry axis with respect to the face region of the second reduced image. If the recalculated symmetry is greater than the previously calculated symmetry, characterized in that it comprises a symmetry calculator for updating the previously calculated symmetry to the recalculated symmetry.

According to the present invention, since the left and right symmetry of the face is calculated using the image processing technique in the face detected in the still image, the accuracy of the symmetry can be improved.

In addition, the present invention calculates the symmetry of the face while rotating or moving the image including the face, so that the accurate symmetry can be calculated even if the image to be discriminated is not aligned with respect to the symmetry axis. .

1 is a view schematically showing the configuration of the device for calculating the left and right symmetry of the face according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating an unsharp masking process performed by the unsharp masking processor shown in FIG. 1. FIG.
3 shows an example of a mask used in an unsharp masking process;
4 is a diagram showing a face region detected from a still image.
5 illustrates an ROI image including a face region.
6 schematically illustrates the concept of Gaussian pyramid resolution reduction.
7 is a diagram schematically illustrating a method of calculating symmetry in a reduced image including a face region.
8 is a flowchart illustrating a method for calculating left and right symmetry of a face according to an embodiment of the present invention.
9 is a flowchart showing an image preprocessing process illustrated in FIG. 8.

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

1 is a view schematically showing the configuration of the device for calculating the left and right symmetry of the face according to an embodiment of the present invention.

As shown in FIG. 1, the apparatus 100 for calculating left and right symmetry of a face according to an embodiment of the present invention may include an image preprocessor 110, an ROI image generator 120, an image converter 130, And a symmetry calculator 140.

The image preprocessor 110 performs preprocessing to reduce the effects of image illumination, image noise, and the like on the still image. As shown in FIG. 114, and an unsharp masking processing unit 116.

First, the image resizing unit 112 adjusts the resolution of a still image using an interpolation algorithm (Interpolation). In one embodiment, the projection resizing unit 112 may resize the still image to a resolution of 640 * 640 size.

The image resizing unit 112 may directly acquire an image, which is a resizing target, from a camera (not shown).

Next, the histogram equalizer 114 equalizes the histogram of the resized still image in order to improve the contrast of the resized still image.

Specifically, the histogram equalizer 114 calculates a maximum value and a minimum value among pixel values of the resized still image, and sets the maximum value and the minimum value such that the pixel value of each pixel of the resized still image is within a specific value. By performing nonlinear mapping, the histogram of the resized still image is equalized. This improves the contrast of the resized still image.

Next, the unsharp masking processing unit 116 performs unsharp masking on the histogram equalized still image in order to obtain an image with emphasis on the contour while preserving the histogram equalized still image.

Specifically, the unsharp masking processing unit 116 duplicates the histogram equalized still images into three (hereinafter, referred to as 'first to third still images') as shown in FIG. The still image is smoothed using a 3 * 3 Laplacian convolution mask as shown in FIG. 3. Here, smoothing means an image processing technique used for blurring and noise removal.

Next, the unsharp masking processing unit 116 calculates a difference image between the smoothed second still image and the third still image, and then generates a still image with an outline highlighted by summing the difference image and the first image.

Referring back to FIG. 1, the ROI image generator 120 generates a region of interest (ROI) image including a face region from a still image preprocessed by the image preprocessor 110. As illustrated in FIG. 1, the ROI image generator 120 includes a face region detector 122 and a face region separator 124.

First, the face region detector 122 detects a face region from the still image preprocessed by the image pretreatment 110. In one embodiment, the face region detector 122 detects a face region from a preprocessed still image using an open source computer version library such as OpenCV. In this case, the face area detector 122 may express the detected face area as a quadrangle 300 as shown in FIG. 4 on a still image.

Next, the face region separator 124 separates the face region from the still image using a mask having a predetermined size to generate an ROI image including the face region.

In detail, the face region separator 124 separates the image having the size of the rectangle 300 from the still image, and then, in the rectangle 500 smaller than the rectangle 300 by a predetermined size, as shown in FIG. 5. By changing the pixel values of the pixels included in the outer region of the inscribed oval mask 510 to a constant value (for example, all zeros), ROI images in which regions except for the face region 530 and the face region 530 are divided ( 520).

Referring back to FIG. 1, the image converting unit 130 generates a first reduced image obtained by reducing the ROI image 520 or rotates the ROI image 520 by a predetermined value and then reduces the second image. Create a reduced image. According to an embodiment, the image converter may generate different second reduced images by rotating or moving the ROI image 520 while sequentially changing a predetermined value.

In this case, the image converter 130 may reduce the ROI image 520 by using a Gaussian pyramid resolution reduction method capable of gradually reducing the image as illustrated in FIG. 6. To this end, the image converter 130 includes an image moving unit 132, an image reduction unit 134, and an image restoration unit 136.

First, the image moving unit 132 rotates or moves the ROI image 520 by a predetermined value so that the symmetry calculator 140 can calculate the symmetry with respect to the ROI image 520 at various positions. Let's do it. In this case, the predetermined value may be randomly set or determined according to the symmetry calculated by the symmetry calculator 140.

For example, a first symmetry degree calculated for the ROI image at the first position and a second symmetry degree calculated for the ROI image at the second position in which the ROI image is rotated by the first angle to the left with respect to the first position. As a result, when the second symmetry is larger, the symmetry is increased as the ROI image is rotated to the left.

Accordingly, the image moving unit 132 rotates the ROI image at a third position again by rotating the ROI image by a second angle obtained by adding up the first angle and a value corresponding to half of the first angle based on the first position. Can be generated.

Next, the image reduction unit 134 generates a first reduced image by subsampling the ROI image 520 using a Gaussian convolution mask and subsampling at 1 / n. In addition, the image reduction unit 134 rotates or moves the ROI image 520 by a predetermined value and then generates a second reduced image by performing low pass filtering and 1 / n subsampling using a Gaussian convolution mask. .

In this case, the image reduction unit 134 may store the low pass filtered image in a memory (not shown) for image restoration work for moving and rotating the image.

As described above, in the present invention, reducing the ROI image 520 by using the Gaussian pyramid resolution reduction method reduces an operation process when calculating the symmetry by the symmetry calculator 140, and includes the image included in the ROI image 520. This is to minimize noise reflected in the symmetry calculation.

Next, when the symmetry is calculated by the symmetry calculator 140, the image reconstructor 136 performs high pass filtering on the ROI image 520 through the Laplacian convolution mask and the image reduction unit 134. The ROI image is reconstructed by synthesizing the low pass filtered image.

To this end, the image reconstructor 136 performs high pass filtering on the ROI image through the Laplacian convolution mask and generates the high pass filtered image when the first and second reduced images are generated by the image reduction unit 134. Can be stored in memory.

In the present invention, after the symmetry calculation by the symmetry calculator 140, reconstructing the ROI image 520 from the first reduced image or the second reduced image is performed like the first reduced image and the second reduced image. Rotating or moving an image with a low resolution may damage the image.

In the above-described embodiment, the image converter 130 is described as including the image restorer 136 to restore the reduced image. However, in the modified embodiment, the ROI image 520 is recorded in the memory and the ROI is modified. The ROI image recorded in the memory may be directly rotated or moved without restoring the ROI image to rotate or move the image 520. Therefore, in this case, since the image restoration unit 136 is not necessary, the image restoration unit 136 may be selectively included.

Next, the symmetry calculator 140 calculates the symmetry between the left and right regions based on a predetermined symmetry axis with respect to the face region of the first reduced image generated by the image converter 130, and records the symmetry between the left and right regions. do.

Subsequently, the symmetry calculator 140 calculates the symmetry of the face area of the second reduced image rotated or moved by a predetermined value from the first reduced image, and calculates the face area of the first reduced image. If the calculated symmetry for the face area of the second reduced image is greater than the calculated symmetry, the previous symmetry recorded in the memory is updated to the symmetry of the face area of the second reduced image.

The symmetry degree calculator 140 subsequently calculates a symmetry degree sequentially for each of the second different reduced images generated by the image converter 130, and repeats the above-described process by a predetermined number of times, thereby updating the symmetry degree. Determine the left and right symmetry of the face.

To this end, the symmetry calculator 140 may include a format converter 142 and a calculator 144.

First, as illustrated in FIG. 7, the format converter 142 may be disposed in the left region 710 and the right region 720 of the face region 530 based on a predetermined symmetry axis 700 in the ROI image 520. The RGB (Red, Green, Blue) values of the existing pixels are converted into HSV (Hue, Saturation, Value) values.

The reason why the format converter 142 converts the RGB values of the pixels into the HSV values is that it is more accurate to calculate the symmetry using the HSV values than the RGB values for the human skin color.

Next, as illustrated in FIG. 7, the operation unit 144 may have a difference in H between pixels existing at positions corresponding to each other in the left region 710 and the right region 720 based on the symmetry axis 700. The symmetry is calculated by arithmetically averaging the sum of the values, the sum of the differences for S, and the difference for V. When each of these is expressed as an equation, the following Equations 1 to 4 are given.

는 각 픽셀의 H에 대한 차이값들의 합을 나타내고,

는 각 픽셀들의 S에 대한 차이값을 나타내며,

는 각 픽셀의 V에 대한 차이값들의 합을 나타낸다. 또한,

는 좌측 영역에 포함된 i,j번째 픽셀의 H값을 나타내고,

는 좌측 영역에 포함된 i,j번째 픽셀의 S값을 나타내며,

는 좌측 영역에 포함된 i,j번째 픽셀의 H값을 나타낸다. 또한,

는 우측영역에 포함된 i,j번째 픽셀의 H값을 나타내고,

는 우측영역에 포함된 i,j번째 픽셀의 S값을 나타내며,

는 우측영역에 포함된 i,j번째 픽셀의 H값을 나타낸다. 그리고,

은 좌측 영역과 우측 영역간의 대칭도를 나타낸다.In Equations 1 to 4

Represents the sum of the difference values for H of each pixel,

Denotes the difference of S for each pixel,

Denotes the sum of difference values for V of each pixel. Also,

Denotes the H value of the i, j th pixel included in the left region,

Represents the S value of the i, j th pixel included in the left region.

Denotes the H value of the i, j th pixel included in the left region. Also,

Denotes the H value of the i, j th pixel included in the right region,

Denotes the S value of the i, j th pixel included in the right region.

Denotes the H value of the i, j th pixel included in the right region. And,

Represents the degree of symmetry between the left region and the right region.

As described above, the present invention calculates the left and right symmetry of the face while rotating or moving the ROI image including the face region through the image converter 130 and the symmetry calculator 140, thereby determining the ROI image to be discriminated. Accurate degrees of symmetry can be calculated even if they are not aligned with respect to this axis of symmetry.

As described above, the apparatus for calculating the left and right symmetry of the face 100 may be mounted and driven in the form of a program or an application in a portable terminal such as a mobile phone, a smartphone, a tablet PC, a notebook, or the like, on which a camera is mounted.

Hereinafter, a method of calculating the lateral symmetry of the face according to the present invention will be described in detail with reference to FIGS. 8 and 9.

8 is a flowchart showing a method of calculating a face left and right symmetry according to an embodiment of the present invention.

First, a still image is received from a device such as a camera (S800), and the received still image is preprocessed (S810). The pretreatment process of the still image will be described in more detail with reference to FIG. 9.

9 is a flowchart illustrating a preprocessing process of a still image according to an embodiment of the present invention. As shown, first, the resolution of the still image is resized using an interpolation algorithm (S900). In one embodiment, the still image is resized to a resolution of 640 * 640 size.

Next, a maximum value and a minimum value of pixel values of the resized still image are calculated (S910), and nonlinear mapping is performed using the maximum value and the minimum value such that the pixel value of each pixel of the resized still image is within a specific value. By performing the operation, the histogram of the resized still image is equalized (S920). This improves the contrast of the resized still image.

Next, the histogram equalized still image is copied to the first to third still images (S930), and the second still image is smoothed using a 3 * 3 Laplacian convolution mask (S940). Here, smoothing means an image processing technique used for blurring and noise removal.

Next, a difference image between the smoothed second still image and the third still image is calculated (S950), and then, the sum image and the first image are summed to generate a still image with an outline highlighted (S960).

Through this preprocessing, it is possible to obtain still images with enhanced contrast and emphasis on outlines.

Referring back to FIG. 8, a face region is detected from the preprocessed still image (S820). In one embodiment, the facial region may be detected using an open source computer version library such as OpenCV. In this case, the detected face region may be represented by the rectangle 300 as shown in FIG. 3 on the still image.

Next, a face region is separated from the still image using a mask having a predetermined size to generate an ROI image including the face region (S830).

Specifically, as shown in FIG. 5, an elliptical mask in which an image having a size of a square 300 is separated from a still image, and then inscribed into the square 500 smaller than the square 300 by a predetermined size ( By using the 510, a pixel value of pixels included in an external area of the mask 510 is changed to a constant value (for example, all zeros) to generate an ROI image in which a face area and an area except the face area are divided. have.

Next, the ROI image is reduced to generate a first reduced image (S840). In one embodiment, the first reduced image may be generated using a Gaussian pyramid resolution reduction method.

Specifically, the first reduced image may be generated by first performing low pass filtering on the ROI image using a Gaussian convolution mask and subsampling the low pass filtered ROI image at 1 / n.

Next, the degree of symmetry between the left region and the right region is calculated based on the symmetry axis preset for the face region of the first reduced image (S850). In one embodiment, the degree of symmetry is a sum of difference values for H between pixels existing at positions corresponding to each other in the left region and the right region with respect to the axis of symmetry, as described in Equations 1 to 4 above, The sum of the differences for S, and the arithmetic mean of the differences for V are calculated.

To this end, the process of converting the RGB value of each pixel to the HSV value may be further included.

Next, a second reduced image in which the first reduced image is rotated or moved by a predetermined value is generated (S860). In one embodiment, the second reduced image may be generated by rotating or moving the ROI image by a predetermined value to reduce the size. In this case, the ROI image to be rotated or moved may be the ROI image generated in S830 but may be a ROI image reconstructed from the first reduced image.

In detail, the ROI image may be reconstructed by synthesizing the low pass filtering image of the ROI image generated through the Gaussian convolution mask and the high pass filtering image of the ROI image generated through the Laplacian convolution mask in the process of generating the first reduced image. Can be.

As described above, rotating or moving the ROI image without rotating or moving the scaled down image and generating the scaled down image again based on this may cause damage to the image when the lower resolution image is rotated or moved. to be.

Thereafter, the degree of symmetry between the left region and the right region is recalculated with respect to the face region of the second reduced image based on the axis of symmetry (S870). Since the method of calculating the symmetry is the same as calculating the symmetry of the face region of the first reduced image, detailed description thereof will be omitted.

Next, it is determined whether the recalculated symmetry is greater than the previously calculated symmetry (S875). If the recalculated symmetry is larger, the previously calculated symmetry is updated to the recalculated symmetry (S880).

Next, it is determined whether a predetermined number of repetitions has been performed (S890), and when a predetermined number of repetitions has been reached, the final updated symmetry is determined as face symmetry (S895). After changing (S897), the process of S860 to S890 is repeated.

On the other hand, if the symmetry degree recalculated in S875 is not greater than the symmetry degree previously calculated, the process branches to S890 to perform the subsequent process.

The above-described method for calculating the left and right symmetry of the face may be implemented in the form of program instructions that can be executed by various computer means, and recorded on a computer-readable recording medium. In this case, the computer-readable recording medium may include program instructions, data files, data structures, and the like, alone or in combination. On the other hand, the program instructions recorded on the recording medium may be those specially designed and configured for the present invention or may be available to those skilled in the art of computer software.

Those skilled in the art to which the present invention pertains will understand that the above-described present invention can be implemented in other specific forms without changing the technical spirit or essential features.

Therefore, it is to be understood that the embodiments described above are exemplary in all respects and not restrictive. The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: left and right symmetry calculator 110: image pre-processing unit
112: image resizing unit 114: histogram equalization unit
116: unsharp masking processing unit 120: ROI image generating unit
122: face region detector 124: face region separator
130: image conversion unit 132: image moving unit
134: image reduction unit 136: image restoration unit
140: symmetry calculation unit 142: format conversion unit
144: arithmetic unit

Claims (16)

Generating a first reduced image by reducing a region of interest (ROI) image including a face region;
Calculating a symmetry degree between a left region and a right region on the basis of a preset symmetry axis for the face region of the first reduced image;
Recalculating the symmetry with respect to the face region of the second reduced image in which the first reduced image is rotated or moved by a predetermined value based on the symmetry axis; And
Determining the left and right symmetry of the face by updating the previously calculated symmetry to the recalculated symmetry if the recalculated symmetry is greater than the previously calculated symmetry. Symmetry calculation method. The method of claim 1,
And repeating the recalculation step and the determining step a predetermined number of times while changing the predetermined value to determine the last updated symmetry degree as the left and right symmetry degree of the face. The method of claim 1,
In the recalculating step, the second reduced image is generated by rotating or moving the ROI image by the predetermined value to reduce the left and right symmetry. The method of claim 1, wherein the retrieving step includes:
Restoring the ROI image from the first reduced image;
Generating the second reduced image by rotating or moving the reconstructed ROI image by the predetermined value to reduce the reconstructed ROI image; And
And recalculating the symmetry degree with respect to the face area of the second reduced image based on the symmetry axis. The method of claim 4, wherein
In the step of restoring the first reduced image to the ROI image, the ROI generated by the low pass filtering image and the Laplacian convolution mask of the ROI image generated by the Gaussian convolution mask in the process of generating the first reduced image. And a high pass filtering image of the image to restore the ROI image. The method of claim 1,
In the generating of the first reduced image, the ROI image is low pass filtered using a Gaussian convolution mask and the first reduced image is generated by subsampling the low pass filtered ROI image. Method of calculating the left and right symmetry. The method of claim 1,
The degree of symmetry may include a sum of difference values for H, a sum of difference values for S, and a difference value for V between pixels existing at positions corresponding to each other in the left and right regions based on the symmetry axis. A method for calculating the left and right symmetry of a face, which is calculated by performing an arithmetic mean. The method of claim 1,
Preprocessing the still image through resizing, histogram equalization, and unsharp masking of the still image; And
Detecting the face region from the preprocessed image and generating the ROI image including the face region,
In the generating of the ROI image, a method of calculating left and right symmetry of a face, characterized in that the ROI image is divided into regions excluding the face region and the face region using a mask having a predetermined shape. Generating a first reduced image by reducing an ROI image including a face region; And
Based on a symmetry axis preset for the face area of the second reduced image in which the first reduced image is rotated or moved by a predetermined value, a symmetry degree between the left area and the right area is calculated, and the calculated symmetry is based on left and right symmetry of the face. And a road determining step. A computer-readable recording medium having recorded thereon a program for performing the method according to any one of claims 1 to 9. An image preprocessor configured to preprocess still images;
A ROI image generator configured to generate a ROI image including a face region from the preprocessed still image;
An image converter for generating a first reduced image by reducing the ROI image by using a Gaussian pyramid resolution reduction method and generating a second reduced image by rotating or moving the ROI image by a predetermined value; And
The symmetry degree between the left region and the right region is calculated based on a predetermined symmetry axis with respect to the face region of the first reduced image, and then the symmetry degree is recalculated based on the symmetry axis with respect to the face region of the second reduced image. And a symmetry calculator for updating the previously calculated symmetry degree to the recalculated symmetry if the recalculated symmetry is greater than the previously calculated symmetry. The method of claim 11,
The image converting unit sequentially generates different second reduced images while changing the predetermined value,
The symmetry calculator calculates a symmetry degree sequentially for each of the different second reduced images, and updates the symmetry degree that is greater than the previously calculated symmetry degree to determine the last updated symmetry degree as the left and right symmetry degree. A device for calculating left and right symmetry of a face, characterized in that The method of claim 11, wherein the image converter,
An image moving unit which rotates or moves the ROI image by a predetermined value;
An image reduction unit generating the first or second reduced image by subsampling the ROI image or the ROI image rotated or shifted by the predetermined value by using a Gaussian convolution mask to subsample the image; And
And an image reconstructor configured to reconstruct the ROI image by synthesizing the high pass filtered image and the low pass filtered image through the Laplacian convolution mask. The method of claim 11, wherein the symmetry calculation unit,
A format converter for converting RGB values of pixels existing in the left region and the right region into HSV; And
The symmetry by arithmetically averaging the sum of difference values for H, the sum of difference values for S, and the difference values for V between pixels existing at positions corresponding to each other in the left and right regions with respect to the symmetry axis. A device for calculating the left and right symmetry of a face, characterized in that it comprises a calculation unit for calculating a degree. The method of claim 11,
The image preprocessor includes: an image resizing unit configured to adjust the resolution of the still image;
A histogram equalizer which equalizes the histogram of the resized still image to improve contrast of the resized still image; And
And an unsharp masking processing unit for unsharp masking to emphasize the contour by unsharp masking the histogram equalized still image. The method of claim 11, wherein the ROI image generating unit,
A face region detector detecting the face region from the preprocessed image; And
A face region separator configured to separate the face region from the still image by using a mask having a predetermined shape to generate a ROI image including the face region,
And the face region separating unit divides the face region and the region excluding the face region by changing pixel values of pixels included in an area outside the mask to a predetermined value.

Images