KR101498142B1 - Facial components detection system using haar-like feature and bezier curve - Google Patents

Abstract

The present invention relates to a system for detecting facial components using Haar-like features and Bezier curves and, more specifically, to a system for detection facial components using Haar-like features and Bezier curves which detects a particular area of a face area in a selected image using a Haar-like feature technique, detects a facial component area by binarizing the detected particular area, and detecting facial components from the detected facial component area by applying a Bezier curve. The system comprises: a particular area extraction module for extracting a particular area in a face area from a selected image using a Haar-like technique; a binarization module for generating a facial component area by binarizing the colors of the extracted particular area; a noise removal filtering module for removing noise in the generated facial component area; a facial component detection module for detecting facial components by applying a Bezier curve to the generated face component area; and a facial component display module for displaying facial components detected by the facial component detection module.

Description

FIELD OF THE INVENTION [0001] The present invention relates to a facial component detection system using a feature of a hareq and a Bezier curve,

The present invention relates to a system for detecting facial components using a Haar-like Feature and a Bezier Curve.

More specifically, a specific region of a facial region is extracted from a selected image using a HARQ feature technique, a facial component region is generated by binarizing the extracted specific region, and a Bezier Curve ) And a face component detection system using a Bezier curve.

In image processing and interpretation, human faces are considered to be important elements for visual identification and identification. Since the early 1990s, face recognition and facial expression analysis have been extensively developed. Recently, MPEG-7 face descriptors have been proposed for face searching and identification in the flow of images.

In contrast to the conventional face recognition algorithms, the above-described techniques have the main function of the face descriptor to retrieve the same face images as the query image as quickly and effectively as possible.

The face recognition technology is a technology that allows a computer to recognize one or more faces present in a still image or a moving image as data that can be discriminated. The face recognition technology can be applied to a security system or the like.

In this case, when the face recognition technology is applied to a security system or the like, access can be controlled by confirming the identity using a database.

Unlike fingerprint recognition, which is another biometric technology, such a face recognition technology does not have to directly touch a part of the body of the user with the recognition device, so that the method of acquiring biometric information is less enforceable. However, the face has a disadvantage in that its recognition rate is lower than that of other biometric systems because it is very sensitive to changes in illumination and pose (posture) and is very sensitive to the surrounding environment.

Face recognition methods include knowledge-based methods, feature-based methods, template matching-based methods, and outline-based methods.

The knowledge-based method is based on the assumption that each face component has a certain positional relationship. However, it is difficult to detect various changes such as the inclination and angle of the face.

Feature-based methods are based on size, shape, correlation, color and texture information of face components. This method has the advantages of fast processing time but it is sensitive to noise such as lighting, pose, and complex background.

Template matching based method is a method of creating standard templates of all faces and comparing them with input images. This method has the advantage of being less affected by lighting or background, but it requires considerable effort in template production.

On the other hand, as one of the techniques for face recognition, a face recognition apparatus and method using a Harcock Filler / linear discriminant analysis is disclosed in Japanese Patent Application Laid-Open No. 10-2005-0112219.

The above technique detects a human face on a device recognizing a human face using a Haar-like feature and a linear discriminant analysis (LDA) and generates facial image information to be analyzed from facial image information Extracts partial image information, extracts hiking feature information from partial image information, and extracts final feature information of a face through linear discriminant analysis based on the extracted hiking feature information.

However, the above technique is for detecting a face region, and it is not suitable to detect a specific region in the face region.

Thus, as a technique for detecting a specific area in a face, a method of detecting an eye condition is disclosed in Patent Document 10-1032726.

The above technique also detects the state of the eye region in the face region by using the hakeque technique. By detecting the eye opening and eye closing by using the ellipse which best matches the boundary points, And the eye condition is detected.

That is, it may be easy to detect the eye condition, but it is also inappropriate to detect the eye region.

Published Patent Application No. 10-2005-0112219 (November 30, 2005) Patent Registration No. 10-1032726 (2011. 05. 06.)

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and it is an object of the present invention to provide a face component detection system using a Beaky curve and a hacking feature capable of detecting a specific component of a face quickly and accurately There is.

In order to solve the above problems, the present invention extracts a specific region of a face region from a selected image using Haar-like feature, A feature of the extracted face region by binarizing the hue of a specific region and a feature of detecting the face component by applying a Bezier Curve in the generated face component region and a face component using the Bezier curve A detection system is provided to solve the technical problem.

According to the present invention, there is an effect of detecting a specific component in the face of various races, not limited to Western and Oriental people.

In addition, the time required to detect a specific component on the face is shortened, and the effect can be applied to a drowsiness detecting device, a specific face expression technique of the robot, etc. using the specific component detected.

1 is a view showing eight square features of the Harcake feature used in the present invention.
Figure 2 shows an integral image applied to the Harcourt feature used in the present invention.
FIG. 3 is a diagram schematically illustrating a configuration of a face component detection system using a Hahake feature and a Bezier curve according to the present invention.
FIG. 4 is a diagram illustrating a process for detecting face components in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.
FIG. 5 is a photograph of an embodiment in which a specific region is detected in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.
FIG. 6 is a diagram illustrating a transformed image using a YCbCr transformer in the binarization method in the face component detection system using the Halake feature and the Bezier curve according to the present invention.
FIG. 7 shows an image transformed by using a gray scale transforming unit in a binarizing method in a face component detecting system using a Beaky curve and a Harack feature according to the present invention.
FIG. 8 is a diagram illustrating an image obtained by binarizing an image by grayscale conversion in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.
9 is a diagram illustrating an example of detecting face components in a specific region in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.
FIG. 10 is a diagram showing a result (A) of detecting a face component without applying the harking feature, and a face component detection result (B) using the harking feature according to the present invention and the face component detection system using a Bezier curve .

The terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary terms and the inventor may properly define the concept of the term in order to best describe its invention It should be construed as meaning and concept consistent with the technical idea of the present invention.

Therefore, the embodiments described in the present specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and are not intended to represent all of the technical ideas of the present invention. Therefore, various equivalents It should be understood that water and variations may be present.

Before describing the present invention with reference to the accompanying drawings, it should be noted that the present invention is not described or specifically described with respect to a known configuration that can be easily added by a person skilled in the art, Let the sound be revealed.

Before describing the contents of the present invention in detail, the technology used in the present invention will be described first, and then the constitution of the present invention using this technology will be described.

First, the Haar-like Feature technique is described.

The Haar-like feature using the integral image is one of the face component detection methods that Viola and Jones proposed, which is robust to the environmental factors and can be operated quickly.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 of the accompanying drawings shows eight rectangular features of the Harcake feature used in the present invention, wherein (a), (b), (c) and (d) (E), (f), (g) and (h) are used for detecting a line of an image.

These square features calculate the feature values by calculating the difference between the pixel sum of the black area and the pixel sum of the white area. At this time, the calculation speed can be improved rapidly by applying the integral image. [Equation 1] below is a formula for obtaining an integral image.

[Formula 1]

Here, ii (x, y) is an integral image and i (x ', y') is an original image.

The integral image has the sum of all pixels from the origin i (0, 0) to i (x, y) of the original image.

Figure 2 shows an integral image applied to the Harcourt feature used in the present invention.

The pixel sum of each of the regions A, B, C, and D in FIG. 2 can be obtained by using the above-described [Expression 1]. That is, the number 4 in FIG. 2 of the accompanying drawings has the sum of the input image areas A + B + C + D. The number 2 represents the sum of the A + B regions, the number 3 represents the sum of the A + C regions, and the number 1 represents the sum of the A regions. Therefore, by calculating 4 + 1- (2 + 3) in the numeric area, the area of square D is calculated.

In this case, by applying the square feature of FIG. 1 to the input image, the pixel sum of the black region and the white region is obtained through Equation 1, and the difference between the two sums is calculated to obtain the feature value of the component.

Next, the binarization technique will be described.

One of the binarization techniques is to transform the RGB color model into a YCbCr color model and then binarize it. This technique can reduce the influence of illumination by converting each RGB color image into a YCbCr color model, since the Y value, which is the brightness part, can be excluded.

The YCbCr conversion equation is shown in [Equation 2] below.

[Formula 2]

Y = 0.299R + 0.587G + 0.114B

Cb = 128 + (- 0.169R + 0.331G + 0.5B)

Cr = 128 + (0.5R-0.149G-0.081B)

Here, Y is luminance (amount of light), Cb and Cr are color difference values, R is red, G is green, and B is blue.

Using the skin color reference range proposed by Douglas Chai as a distribution map of the Cb value and the Cr value after the conversion into the YCbCr color model in a predetermined skin color range, It is possible to binarize the portion determined to be white by adjusting the value.

In addition to the conversion to the YCbCr color model, the binarization method is a binarization method of converting the original color image into grayscale and comparing the gray scale value with the threshold value. [Equation 3] below is an expression for converting the RGB color to grayscale.

[Formula 3]

Y = 0.299R + 0.587G + 0.114B

R = G = B = Y

Here, Y is the luminance (amount of light), R is the red value, G is the green value, and B is the blue value.

In the case of performing specific image processing on the RGB color model, since the processing for each of the R, G, and B channels must be performed, the calculation time is increased about three times as compared with the gray scale image. Therefore, when fast operation is required, it can be configured to process the image using only the brightness component of the RGB color model.

At this time, the R, G and B channels of the RGB color model of the original image are converted into grayscale using only the Y (brightness) component, and set to 255 for all pixels in the image, And performs the binarization.

The Bezier Curve will be described.

The Bezier curve was named after a French engineer named Pierre Bezier who used this curve in the 1970s when designing a Renault car. This curve is applied to various fields because it is simple and the calculation is fast and stable because it is not out of set control points.

[Equation 4] below expresses the Bezier curve.

[Formula 4]

[Equation 4] is a continuous function for obtaining a curve generated by the control points of N + 1 starting from 0 (N is 2 if the point is 3).

For example, if there are three N, that is, four control points, P ₀ and P ₃ are the end points of the curve, the curve from P ₀ to P ₁ , the curve from P ₁ to P ₂ , and P ₂ to P ₃ , And curves are defined by four points.

A description will be given of a hatch feature and a face component detection system using a Bezier curve according to the present invention.

The present invention relates to a system for detecting facial components using Haar-like features and Bezier curves, and more particularly, to a system for detecting facial components using a Haar-like feature and a Bezier curve, Extracts the extracted feature region, generates a facial component region by binarizing the extracted specific region, and generates a Bezier Curve in the generated facial component region. And a face component detection system.

FIG. 3 is a diagram schematically illustrating a configuration of a face component detection system using a Hahake feature and a Bezier curve according to the present invention, and FIG. 4 is a block diagram of a face component detection system using a Hahake feature and a Bezier curve according to the present invention. FIG. 8 is a diagram illustrating a process for detecting a face component. FIG.

As shown in the accompanying drawings, the face component detection system using the haraki feature and Bezier curve according to the present invention includes a specific region extraction module 10, a binarization module 20, a noise removal filtering module 30, A detection module 40 and a face component display module 50.

The specific region extraction module 10 extracts a face region from a selected image (image or photograph) and extracts a specific region from the extracted face region. In this configuration, a Haar-like Feature Extract a specific region.

The face area extracted using the Haar-like Feature technique is extracted from the eyebrows except the hair area to a part of the jaw.

After the facial region is extracted, its peripheral region is extracted around the specific region.

At this time, the specific area may be composed of eyes and mouth.

The reason for extracting the eye and mouth peripheral region separately is to reduce the skin area as much as possible in the process of binarizing the next image to reduce errors such as noise that may occur due to environmental variables such as illumination and to improve processing speed.

FIG. 5 of the accompanying drawings shows an embodiment in which a specific area, that is, an eye area and an area around the mouth are extracted by using the HARQ feature technique.

The HARQ feature technique shows a robust extraction rate even under various environmental changes, and it is also relatively simple to implement using a library. Also, there is an advantage that the face region can be extracted quickly.

Three strong haar_cascade classifiers can be used to detect face, eye, and mouth, which are specific regions in the extracted face region.

A strong classifier starts with a weak classifier and sequentially searches through the entire image to determine where each component region of the face is. By repeating this process, learning becomes one strong classifier.

The binarization process is performed based on the extracted image as shown in FIG.

The binarization module 20 generates a facial component region by binarizing the hue of the detected specific region, and includes a YCbCr conversion section 21 and a gray scale conversion section 22.

The YCbCr conversion unit 21 converts an RGB color image of the extracted specific region into a YCbCr color image, and FIG. 6 of the accompanying drawings shows an embodiment in which an RGB color image is converted into a YCbCr color image.

The color of a specific region extracted in this process is converted into a YCbCr color image through the following [Equation 5].

[Equation 5]

Y = 0.299R + 0.587G + 0.114B

Cb = 128 + (- 0.169R + 0.331G + 0.5B)

Cr = 128 + (0.5R-0.149G-0.081B)

Here, Y is luminance (amount of light), Cb and Cr are color difference values, R is red, G is green, and B is blue.

The gray-scale conversion unit 22 converts the RGB color image of the extracted specific region into a monochrome image. FIG. 7 of the accompanying drawings shows an embodiment in which an RGB color image is converted into a monochrome image.

At this time, the brightness value is removed from the monochrome image converted through the gray scale conversion unit 22, and the RGB values are processed with the gray range values.

FIG. 8 is a diagram illustrating an image obtained by binarizing an image by grayscale conversion in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.

The color of the specific region extracted in this process is converted into gray scale through the following [Equation 6].

[Equation 6]

Y = 0.299R + 0.587G + 0.114B

R = G = B = Y

Here, Y is the luminance (amount of light), R is the red value, G is the green value, and B is the blue value.

Here, the YCbCr conversion unit 21 and the gray-scale conversion unit 22 constituting the binarization module 20 may be configured to binarize an image into a selected one. That is, the binarization process can be performed only through the YCbCr conversion unit 21 that converts the RGB color image into the YCbCr color image, and the binarization process is performed only through the gray scale conversion unit 22 that converts the RGB color image into the monochrome image .

Or two conversion units may be interlocked according to design conditions.

The noise elimination filtering module 30 removes surrounding noise (points) and the like from the binarized image (generated face component area). In particular, when the image is binarized through the YCbCr conversion part 21, Like black pixels can be distributed. Therefore, when such noise or the like is not removed, a problem may arise that an area other than the face component area is generated. Therefore, after the binarization process, noise removal filtering should be performed.

Depending on the design conditions, this noise cancellation filtering module 30 may be omitted.

The face component detection module 40 detects a face component by applying a Bezier curve in the generated face component region.

9 is a diagram illustrating an example of detecting face components in a specific region in the face component detection system using the Hahake feature and the Bezier curve according to the present invention.

That is, the face component detection module 40 detects the points that determine the outline of the binarized face component region, applies the points to the Bezier curve, connects the points to the curve, and detects the face component.

The face component display module 50 displays the face components detected by the face component detection module 40.

To verify the performance of the face component detection system using the Halake feature and the Bezier curve according to the present invention, a verification was performed.

In contrast to the present invention, the entire face image is converted into a YCbCr color model without Haar-like technique, and then a Bezier curve of each face component is generated through binarization.

The images used in the verification were images of the "Face Recognition Homepage" website, which provides face images used in various research fields such as biometrics, computer vision and pattern recognition.

FIG. 10 is a diagram showing a result (A) of detecting a face component without applying the harking feature, and a face component detection result (B) using the harking feature according to the present invention and the face component detection system using a Bezier curve .

10A shows a result of detecting a face component in a conventional configuration, and B shows a result of detecting a face component in a configuration according to the present invention.

In Fig. 10, the result corresponding to each image is the result of detecting a Bezier curve of the left eye, right eye, and mouth.

As can be seen from the verification result, when the distance between the eye and the eyebrow is short in the original image, the face component detected by the configuration of A may include a part of the eyebrow in the eye area. Thus, in the process of performing binarization, the eye and eyebrows are converted into black together, and when the Bezier curve is generated, the position of the point is not correctly selected and the curve is shifted to one side, or the original shape of the face component is not maintained As a result.

Also, in the image database used, Westerners have a characteristic that the eyes are protruded more than the Oriental people like the original images in FIG. 10, resulting in a lot of shadows in the eyes. In this case, when the binarization is performed by the A method, the distribution of the black pixels is distributed not only in the eye region but also around the eye region, so that the outline of the face component can not be detected clearly in the process of generating the Bezier curve .

The result image of ④ and ⑤ is an example showing the result of images with this problem.

In addition, when the face component is detected by the configuration of A, there arises a problem that the curve is not properly detected in the mouth area. That is, when the binarization is performed, white pixels are formed between the upper lip and the lower lip, resulting in a problem that the lips are not connected or only the upper or lower lip is converted to black.

Due to such a problem, the face component detected through the Bezier curve can be detected only in the upper or lower lips, as in (2) and (5) of FIG.

As a result, the face component detected by the configuration of A has many problems such as noise and blank in the process of performing binarization, so that there is a problem that a sharp curve is not generated in the process of detecting the Bezier curve.

On the other hand, the face component detected with the configuration of B according to the present invention is generally detected with a definite boundary along the outline of the component. That is, even if the distance between the eye and the eyebrow is short by applying the Haarque feature technique, the eye component can be accurately detected by extracting only the eye except the eyebrows. In addition, compared with the face component detected by the A configuration, the noise distribution of the black pixels is remarkably reduced, and the binarization with the clear boundary between the skin and the components is performed, and more accurate face components are detected.

Likewise, the detection of facial components due to the facial component A configuration of the mouth is more effective than the detection of the facial component, so that problems such as the gap between the upper lip and the lower lip are eliminated, and thus the distinct facial component can be detected.

According to the present invention, there is an effect of detecting a specific component in the face of various races, not limited to Western and Oriental people.

In addition, the time required to detect a specific component on the face is shortened, and it can be applied to a drowsiness detection device and a specific face expression technology of a robot using the detected specific components, and furthermore, It is effective.

1 to 10 illustrate only the main points of the present invention. As far as various designs can be made within the technical scope thereof, the present invention is not limited to the configurations and functions of Figs. 1 to 10 It is self-evident that it is not limited.

10: Specific area detection module
20: Binarization module
21: YCbCr conversion section
22: Gray scale conversion section
30: Noise canceling filtering module
40: Face component detection module
50: Face component display module

Claims (5)

A specific region extraction module 10 for extracting a specific region of the face region using a Haar-like feature technique in the selected image;
A binarization module (20) for binarizing a color of the extracted specific region to generate a facial component region;
A noise removal filtering module 30 for removing noise from the generated face component region;
A face component detection module 40 for detecting a face component by applying a Bezier curve in the generated face component region; And
And a face component display module (50) for displaying a face component detected by the face component detection module (40)
The binarization module (20)
A YCbCr conversion unit 21 for converting the color of the detected specific area into a YCbCr color image through the following equation (A), and a YCbCr conversion unit 21 for converting the color of the detected specific area into a monochrome image through the following equation [B] And a face component region is generated by binarizing the image with a selected one of the gray scale conversion sections (22).
[Formula A]
Y = 0.299R + 0.587G + 0.114B
Cb = 128 + (- 0.169R + 0.331G + 0.5B)
Cr = 128 + (0.5R-0.149G-0.081B)
Here, Y is luminance, Cb and Cr are color difference values, R is red, G is green, and B is blue.
[Formula B]
Y = 0.299R + 0.587G + 0.114B
R = G = B = Y
Here, Y is luminance, R is a red value, G is a green value, and B is a blue value.
delete delete The method according to claim 1,
The specific region may be,
Wherein the face component is a peripheral region of a face including an eye or an mouth.
The method according to claim 1,
Wherein the face component comprises:
Characterized in that it is in the shape of an eye or a mouth.

Images