KR101280439B1 - Method for recognizability assessment of facial images acquired by cameras on automated teller machines - Google Patents

Abstract

In the method of determining the recognition possibility of the face image according to the present invention, in a cash machine operation situation in which a transaction is allowed only when the user's face is recognizable, the recognition possibility of the face in various situations of face masking and face posture change is determined. In order to judge, a face with one eye and one mouth is defined as a recognizable face to allow financial transactions.

Description

Method for recognizability assessment of facial images acquired by cameras on automated teller machines}

The present invention relates to a face recognition possibility determination method for determining whether there is a recognizable face in an image acquired by a camera of an ATM, in order to reject a situation in which the ATM is unjustly used while covering the face.

Recently, as the use of automated cash machines increases, crimes such as cash extortion or illegal money transfer using them also increase. Criminals who want to use ATM machines unfairly cover their faces to hide their identity. Therefore, in order to prevent a person who covers a face from using a cash dispenser so that the recognition of a user is impossible, a technique of detecting a face occlusion or determining a face's recognition possibility has become a hot topic, and various methods have been invented according to such a need. .

However, since the conventional methods block the use of any part of the face, it is judged as an unfair dealing with natural facial obstruction and various facial postures caused by hair or accessories frequently occurring in actual situations. Have

Japanese Patent Application No. 4-76690 describes a method of determining whether a user's face is normal by taking an image when the user is at a normal distance using a sensor. Here, it is determined whether the face is normally recognizable by comparing a human face with a list of criminals, determining whether the face is masked using a mask or sunglasses, or covering the camera with a hand.

Japanese Patent Application No. Pyeongseong 5-35992 discloses a technique of determining the degree of suspicion by taking a face image of a user, extracting features of eyes, mouth, ears, nose, etc., and weighting them according to the extracted detection degree. .

Korean Patent Application No. 1998-0082154 describes a specific method of properly using flesh color information and background image information to extract feature points such as eyes, nose, and mouth from a face image of a user photographed from a surveillance camera attached to an ATM. It explains.

Korean Patent No. 10-0293897 describes a method of determining whether a face is covered by binarizing color information obtained from a face image photographed by an ATM.

Korean Patent No. 10-0473598 describes a method for detecting face occlusion using a principal component analysis technique and a support vector machine when a face is detected in an image taken by an ATM.

Korean Patent No. 10-0564372 describes a method of confirming that a front face is obtained and generating a face code from it.

Korean Patent No. 10-0825689 describes a method for identifying the presence of the eyes and the mouth only in each area after distinguishing the candidate areas from the front face.

However, these conventional inventions involve two problems as follows. First, the conventional inventions implicitly assume a case in which the user is looking directly at the camera, so that the user is looking at the direction rather than the front or face. When faces of various postures, such as a change of expression, are input, an error of not recognizing a face is generated. Secondly, since the conventional inventions can be recognized only when there is no cover at all on the face, the natural part without the intention of crime, such as when the hair flows down and one side of the eye is covered or the eye patch is worn due to a disease, etc. The offense of judging the user's transaction that the baggage originated as improper use is committed.

Due to these two problems, in order to use the conventional inventions effectively, in order to photograph the front face of the user without obscuring, the user must make a compulsive request to stare at the camera, thereby preventing user inconvenience. Will result.

Accordingly, an aspect of the present invention is to provide a method for determining a facial recognition possibility that can flexibly cope with natural partial occlusion that frequently occurs in actual situations.

Further, the present invention is to provide a face recognition possibility determination method that is enhanced by the user's ease of use by not being instructed to force the user to stare at the camera by being robust to various face postures that do not look directly at the camera.

In the face recognition possibility determination method according to the present invention, one eye and one mouth are identified to secure more than half of the face elements in order to maximize the convenience of the user and to be robust to natural part obstruction and various face postures. In this case, it is defined that a recognizable face exists, and it is determined whether a recognizable face exists in the captured image through face element based face candidate detection and verification technique. In addition, in order to increase the efficiency and completeness of a system for implementing a face recognition possibility determination method, a structure that separates an ROI search and an entire area search and a structure for handling exceptional situations is combined.

To this end, the present invention uses the following means to increase the user's convenience and recognizability determination performance at the same time.

First, a recognizable face is defined as a face with one eye and one mouth. The reason for this definition is that if one eye and one mouth can be secured, more than half of the face elements are secured. Even if only one eye is secured, the other eye can be restored to some extent based on the symmetry of the face. It is easy to identify the person who is shown in the video. As a result, natural face covering by eye patch, hair, or the like can be allowed, and user convenience can be achieved. In this case, a function of determining the posture of the face, in particular, the lean face, and a corresponding face element is added so as to robustly respond to various postures of the face with respect to the camera.

Second, we used a method of detecting and verifying face candidates based on face elements. In particular, instead of using the entire face area to determine the overall recognition possibility of the face, it is possible to determine the recognition possibility using only a few face elements including features that are effective for face recognition. It is also resistant to natural parts of the forehead that appear on the forehead.

Third, a face detector is used to set a region of interest on a part of the face, and allow a transaction as soon as a recognizable face element is found in the region of interest. This configuration greatly improves the efficiency of the system by dramatically improving the processing speed of face recognition possibility judgment.

Fourth, to reduce the errors such as the blindfold due to thick frames or misunderstanding the light reflected from the sunglasses, the system is able to cope with exceptional situations such as glasses and sunglasses to improve the performance of the system. Added exception handling.

The present invention is a method for determining the recognition possibility of a face image acquired by an ATM machine for achieving the above object, and includes the following configuration. First, a method of determining a face recognition possibility of a system interworking with an ATM, comprising: confirming a user's approach to or access to an ATM; Photographing a user's face to obtain N images; Selecting any one of N images and extracting face elements of a user's face from the selected images; And if the extracted facial element is at least one eye and one mouth, allowing subsequent manipulation of the user; otherwise, blocking the manipulation.

The extracting of the facial elements may include extracting at least two candidate facial elements from the selected image; Setting a region of interest in each of the extracted candidate face elements; Connecting the set regions of interest to form an inverted triangular shape; And extracting a face element from each inside of the set ROI when the configured inverted triangle shape is an inverted triangle shape according to a normal face proportion.

The method may further include removing the region of interest and re-extracting an arbitrary face element from the entirety of the extracted image when the facial element extracted from each of the set regions of interest is not any one of eyes and mouths; And checking whether the eye and the mouth of the re-extracted facial elements are extracted.

In addition, when at least one eye and one mouth are not detected from the selected face image, selecting another face image of the N pieces and extracting the face elements from the another face image. Repeat the step further.

The extracting of the facial elements may include applying a principal component analysis method or a local binary pattern to an image of the extracted facial elements to verify false detection of the extracted facial elements (at least one eye and one mouth). The method may further include extracting the feature vector using the extracted feature vector and applying the extracted feature vector to a support vector machine classifier.

In addition, the ROI includes a glasses / sunglasses region, and when glasses are detected in the glasses / sunglasses region, one eye is considered to be detected, and when sunglasses are detected in the glasses / sunglasses region, It further includes the step of deemed not found.

The extracting of facial elements may include extracting at least two candidate face elements corresponding to any one of eye, nose and mouth from the selected face image; Setting a region of interest including the extracted candidate face elements; Connecting the regions of interest to form a triangular pyramid shape; And when the triangular pyramid shape is a triangular pyramid shape according to a normal face proportion, determining the direction of the face of the user using the triangular pyramid shape.

According to the present invention, the recognition face is defined as a face having at least one eye and one mouth, and then the facial element-based face detection and verification method is used to evaluate the recognition possibility of the face. Done. In addition, instead of verifying the recognition possibility using the entire face, the detection and verification is performed based on the face elements, so that the face recognition possibility may be evaluated for various postures where the face is bowed down or the other face is viewed. This feature enhances the user's convenience and increases system performance.

In addition, before applying the face element-based face detection and verification method to the entire image, set the region of interest with the entire face detector that can detect the face of the posture that occurs frequently in the ATM operation, and then apply this method only within the region of interest. By using the system structure of the application method and the method of terminating the determination as soon as the image judged to be recognizable while the plurality of images are sequentially processed, the average processing speed can be significantly improved.

Moreover, the exception of constructing and using the glasses and sunglasses detectors to cope with errors that cannot find the eyes due to the thick frame of the frame, or the false detection of the reflected light reflected on the sunglasses by the eye, a weakness of the facial element based recognition possibility determination method. By using the treatment method, the completeness of the system can be improved.

1 is a view showing an example of an unrecognizable face and a recognizable face classified based on the definition of a recognizable face used in the present invention. Here, (a) shows unrecognizable face images, and (b) shows recognizable face images.
2 illustrates a method of detecting a face candidate including face elements required for recognition. (a) shows the average face used to define the face candidate. (b) shows an example of detection of the left eye, right eye, and mouth. (c) to (h) show various examples of facial elements that can be obtained from the detection example of (b).
3 is a diagram illustrating face candidates in which arrangement of face elements among face candidates detected in FIGS. 2C to 2H satisfy geometrical conditions.
4 is a flowchart illustrating a processing flow of a facial element-based recognition possibility determination method according to the present invention.
5 is a diagram illustrating an exception handling method proposed by the present invention. The left side shows a case where the face detection and verification method fails by glasses and sunglasses, and the right side shows that the face candidate is finally determined by applying the exception processing method according to the present invention to the same face image.
6 shows various examples of processing when the method described in the present invention is applied to an ATM operation situation. The first, second, and third rows show examples of face images determined to be recognizable. The left two of the last row are the faces that were not detected by the eye but were judged to be recognizable by the spectacle detector. The last two right rows show examples of faces that incorrectly detected eyes but were judged to be wearing sunglasses by the sunglasses detector and found to have occlusion.

The present invention relates to a method (or face element based face candidate detection and verification method) of determining a recognition possibility of a face image photographed by an ATM, and has five characteristics as follows.

First, based on the results of various studies on face recognition and analyzing face masks of criminals, we want to define a face condition that can be recognized as a face with one eye and one mouth, and use an ATM machine. If there is at least one eye and mouth from the face image of the user, it is assumed that the ATM is intended to be used without covering the face.

In addition, it is possible to robustly respond to various postures of the face to the camera. In particular, each face element can be determined even in the face of the bowing posture in which the user looks at the input keyboard of the ATM.

The second is to detect and verify the face based on the face elements. In this process, a viola-jones detector is used to detect eyes and mouths as face elements, and then face candidates are set based on geometric conditions using one eye and one mouth.

Third, in order to reconfirm whether the face element detected in the above process is a misdetected face element, the eye and mouth regions are verified using a principal component analysis method, a local binary pattern, and a support vector machine.

Fourthly, in order to improve the amount of computation, the region of interest is first set by the entire face detector, and the face-based elements are searched by applying the element-based facial element detection and verification scheme to the set region of interest. If the face element that is not recognizable in the set ROI is not found, the face element is searched for the entire face.

Fifth, special exception processing is executed to reduce undetected or false detection errors caused by glasses and the like.

The present invention, which applies such a procedure, is robust against various occlusions, facial postures, and lighting changes that may occur in real situations, and at the same time requires a small amount of computation, thereby effectively coping with an actual ATM operation situation.

On the other hand, here, setting the face candidate means selecting one or a plurality of face images to be determined as a recognizable face among the plurality of face images.

The above technical content will be described in more detail below with reference to the accompanying drawings.

In the present invention, the recognizable face is defined as a face having one eye and one mouth. The reason for this definition is as follows. Analyzes of the behavior of criminals (unfair users) who want to use ATM machines unfairly show that almost no one of their face elements covers their nose, and in general, their mouth is secured because they cover their nose at the same time. In this case, it can be assumed that the nose is also secured (see Fig. 1 (a)). Also, if one eye is secured, the other eye is almost approximated to both real eyes by symmetrical of the secured one eye. Therefore, if one eye and one mouth are secured, more than half of the face is secured, thereby making it easy to check the identity of the person in the image based on the symmetry of the human face. In addition, by using such a definition, the user does not recognize as a malicious face obstruction a natural obstruction that may be caused by eye patch, hair falling, a posture where the face is not looking at the camera, and shadows caused by ambient light. Therefore, it is possible to flexibly allow natural face covering that may occur even if it is not intended to be intentionally intended, thereby enabling convenience of use.

In order to determine whether the recognizable face defined above exists in the image acquired from the ATM camera, the face element-based face detection method is used in the present invention. This method finds at least one eye and one mouth with a Viola-Jones detector and then generates their position combinations and selects only those candidates that satisfy the geometrical conditions of the face.

2 shows a face element based face detection step. First, FIG. 2 (a) shows that the average face shape is generated by using various human face shapes. In addition, it can be seen that the geometric positional relationship of the inverted triangular shape connecting the face contour, the two eyes and the mouth is set from the average face shape.

If a user of the ATM machine photographs N times and acquires N images, one of the N images is selected, and the left eye (LE) and the right eye are selected using a viola-jones detector on the selected image. (RE), it detects candidate face elements that are likely to be mouths (M). 2B shows an example of the candidate face elements thus detected.

The region of interest is set for the candidate face element. In the present invention, the candidate face element, which may be any one of the face elements, is detected, and the procedure of verifying that the detection element is the actual face element (described later) is not performed in advance, but is simply set as the region of interest. When the candidate face element included in the set ROI is determined to be the face element, the candidate face element is accurately analyzed to verify whether the face element is correct.

Subsequently, a region of interest including the detected candidate face elements is connected to generate various combinations of one eye and one mouth, and virtually using positional and symmetrical relationships of each face element defined in FIG. An inverted triangular shape is configured, and a virtual candidate face area including a face outline is set according to the configured inverted triangular shape. 2 (c) to (h) show candidate face elements generated by this method and candidate face regions by their positional relationship. In the drawings, candidate face regions of various combinations are set using candidate face elements that may be face elements, and therefore, a case that is appropriately set to an actual face area and a case that is incorrectly set are mixed.

Then, the size ratio of the face and the eye, the ratio of the size of the face and the mouth, the positional relationship between the eyes and the mouth, and the minimum size of the face that can be photographed in the ATM machine operation state are calculated using (a) of FIG. Geometrical constraints such as maximum size are used to remove candidates that are not valid among the set candidate face regions. In this process, (d), (f), (g), and (h) of FIG. 2 are removed, and (c) and (e) are selected as face images having a reasonable face area. That is, when the mouth is positioned above the eyes, the case where the candidate face area is out of the range of the face image is removed.

On the other hand, if there are a plurality of finally selected face regions, it may be determined whether they can be merged with each other, and if they can be combined with each other, two eyes and one mouth may be additionally detected by combining the respective facial elements.

3 shows candidate faces finally generated by the face element based face detector. (A) of FIG. 3 is a candidate consisting of two eyes and one mouth, and FIGS. 3 (b) and (c) are candidates consisting of one left eye or one right eye and one mouth.

After the face region and the region of interest for the eyes and mouth therein are set by the face element-based face detection method, the authenticity of the candidate face elements included in the extracted region of interest is finally confirmed.

To do this, we extract feature vectors using principal component analysis and local binary patterns from images of the ROI that are considered to be eyes and mouths, and then use them as a support vector machine classifier. Classify as 'eye' and 'non-eye', 'mouth' and 'non-mouth'. Accordingly, the face candidate is finally determined to be recognizable only when both the extracted eyes and the mouth are determined to be the actual eyes and the mouth.

Since the face element-based face detection and verification method is used separately from each face, the face element-based face detection and verification method is frequently used in real situations, compared to the existing methods that the entire face must be completely recognizable. Various facial postures and natural partial occlusions can be strongly judged, and thus the possibility of face recognition can be determined.

Meanwhile, the face element-based face detection and verification method according to the present invention uses face detection means capable of accurately detecting a face region by recognizing various directions to which the face faces. After setting an arbitrary region of interest in the detected face region, processing with a small amount of computation is performed first by applying a face element-based face detection and verification method only within the region of interest. And only when the face which can be recognized by this process is not found, the method (full area search) which searches each face element by searching the whole image precisely can also be used.

In the present invention, the front face detector and the bowed face detector (to be described later) are used simultaneously to detect the face area of the entire face. If it is determined that there is a recognizable face in the video, the method terminates all processing and allows the financial transaction to the user. However, when it is determined that there is no recognizable face, the method repeats the aforementioned face element detection using another captured image.

That is, if a face that is recognizable is not found even though all of the predetermined N images are searched one by one, the entire area search is performed for each of the N images again. In the full region search step, the element-based face candidate generation and verification procedure described above is performed not only in a specific region but in all regions of the image to determine whether a face can be recognized.

If this procedure determines that a recognizable face exists, the system immediately stops all processing and allows the user to finance. Conversely, if no face is found in all the N images, the system finally determines that the user is hiding or hiding the face, and asks the user to reject the financial transaction or remove the occlusion and retry the operation.

The face element-based face detection and verification method according to the present invention has several advantages mentioned above, but the eye is blocked by the thick glasses as shown in the upper left of FIG. 4, and the detector does not detect the eyes, or as shown in the lower left. There is a possibility that the reflected light on the sunglasses is misdetected by the eye, which may cause an error in the judgment of the face recognition possibility.

Therefore, an exception handling procedure is introduced to cope with such a situation. In the exception handling procedure, the glasses / sunglass detector which detects frequently occurring glasses and sunglasses is applied to the image being processed, as well as the eye and mouth combinations, as shown in the right images of FIG. Facial candidates can be generated using a combination of sunglasses and mouth regions.

In this case, if the glasses and the mouth are detected in the face image, it is determined that the face is recognizable, and if the sunglasses and the mouth are detected, the transaction is rejected by determining that the face is covered.

5 to 8 are diagrams for explaining the bowed face detector proposed in the present invention and a method for detecting each face element from the bowed face using the bowed face detector.

First, FIG. 5 is a diagram illustrating a result of detecting a face region using a front face detector capable of detecting only an existing front face and a bowed face detector according to the present invention with respect to the same image. In FIG. 5A, a conventional front face detector shows a state in which most faces are not recognized in an image photographed with a bowed head down. On the other hand, FIG. 5B shows that by using the bow face detector according to the present invention, even when the user is bowed down, the face area is extracted almost accurately.

FIG. 6 is a diagram illustrating a method of detecting a direction in which a face faces and a region of a face from an image of a face not looking at the front using a bowed face detector. First, as in (a), the geometric positional relationship of two eyes, one nose and one mouth is established from the average face. The position of both eyes and mouth is connected to form an inverted triangle shape, and the position of the nose is located inside the configured inverted triangle shape. This form is shown as a flattened form of triangular pyramid.

6 (b) shows a result of setting various candidate face elements that may be face elements from one image. In the drawing, the green area is a candidate area of the eye, the blue area is a candidate area of the mouth, and the yellow area is a candidate area of the nose.

Next, the eye candidate region and the nose candidate region, or the eye candidate region and the mouth candidate region are connected (see FIG. 6 (c)), and the connected line segments are applied to the triangular pyramid shape of FIG. 6 (a) to generate a face shape. By using the positional relationship of each facial element, it is possible to determine the viewing direction of the photographed face as shown in FIG.

FIG. 6 (e) shows that a face shape is set using various candidate areas set for the same image. The figure of the rectangle enclosed by the bold blue line in the figure corresponds to FIG. 6 (d) and shows that the face shape is correctly set.

FIG. 7 is a diagram illustrating a result of determining the possibility of face recognition with respect to various images by a face element based face candidate detection and verification method according to the present invention. Up to the top three lines of the drawing, you are not facing the front, are wearing eyeglasses, are wearing glasses, part of your face is covered by a hat's window, you are touching part of your face with your hands, or your face with your hair In this case, any one eye and one mouth are detected for natural face obstruction, such that the face recognition possibility is successful.

In the bottom row of the drawing, two images on the left show a case in which one eyeglass region and one mouth region are detected and the possibility of face recognition is determined to be effective. The two images on the right show that the sunglasses region and one mouth region In this case, the possibility of face recognition is determined to be invalid.

As described above, in the method of determining the recognition possibility of the face image according to the present invention, in a cash machine operation situation in which a transaction is allowed only when the user's face is recognizable, various postures of the face are changed and face posture changes such as bowing the face. In order to be able to flexibly judge the recognition possibility of a face in the situation, a face with one eye and a mouth is visible and is defined as a face that allows financial transactions.

In addition, face element-based face that detects face using the positional relationship of found face elements first in the image to determine the recognition possibility of the face robustly to various part occlusion and face posture in ATM machine operation situation Instead of using the detection method and the entire face to determine the occlusion, a face element-based face verification method is used to determine the occlusion of the face using information of the found eyes and mouth area.

In order to use the face element based face candidate detection and verification method more efficiently, after setting the region of interest using the entire face detector which can detect the face posture that appears frequently in the ATM operation situation, only the face element based After performing the ROI search step that applies the face candidate detection and verification method, and then, if the recognizable face is not found, the whole area search step that applies the face element-based face candidate detection and verification method to the entire image is performed. Use a structure to proceed.

In addition, the face element-based face candidate detection and verification method uses an exception handling scheme in order to cope with a situation such as glasses, sunglasses, etc. that can make an error. The system including the exception handling method detects not only eyes and mouths but also glasses and sunglasses to generate face candidates composed of glasses and mouths, sunglasses and mouths, and intentionally obscurs when a face composed of glasses and mouths is detected. Without this, it is determined that the eye is not detected by the glasses, and the transaction is allowed. When the face composed of the sunglasses and the mouth is detected, it is determined that the intentional obstruction is present and the transaction is rejected.

Next, referring to FIGS. 8 and 9, the basic structure of the face recognition possibility determination system and the method for performing the face recognition possibility determination method (or face element-based face candidate detection and verification method) according to the present invention. Describe the procedure.

8 is a structure of a face recognition possibility determination system according to the present invention. Referring to the drawings, the face recognition possibility determination system 100 includes a viola-jones face detector 110, a face element based face detector 120, and a financial transaction determiner 130.

The viola-jones face detector 110 extracts face elements by analyzing images input from the outside (eg, a camera) by viola-jones algorithms. The extracted face element is a candidate face element since it is a state before it is correctly verified as being each face element.

The face element based face detector 120 forms a geometric shape by connecting candidate face elements extracted from the viola-jones face detector 110 (or a region of interest set to include each candidate face element), and the configured shape. It is determined using which region each candidate face element is.

When each facial element determined by the facial element-based face detector 120 is at least one eye and one mouth, the financial transaction determiner 130 assumes that the face image is an unobstructed face and the user uses an ATM. If the financial transaction is allowed to be executed, and at least one eye and one mouth are not detected, the face is considered to be artificially covered, and the face recognition possibility judgment result is output to block the subsequent financial transaction.

Next, FIG. 9 is a flowchart illustrating a procedure of executing a method for determining facial recognition possibility according to the present invention.

First, when images photographed by a camera or the like are sequentially input, a region of interest for a candidate face element is set in the first image, and a partial region search using the set region of interest is performed. At this time, by using the front face detection method that can set the face area with respect to the front face and the lean face detection method that detects the lean face and set the face area according to the lean posture, To be set.

When the face region is set, face elements are detected within the set face region. When face elements are detected, the face elements are verified by the geometric positional relationship to each face element and the shape analysis of each face element.

If the face element is verified, it is determined which face part, and if at least one eye is detected and one mouth is detected, it is determined that face recognition is possible, and if at least one eye and one mouth are not detected It is determined that face recognition is not possible.

In this procedure, if it is determined that face recognition is possible (" valid " in the drawing), the result of the determination is immediately outputted so that the financial transaction can be executed without performing the subsequent determination procedure. On the other hand, if it is determined that face recognition is not possible (" invalid " in the figure), the flow proceeds to the following procedure to continue face recognition possibility determination.

In this way, the recognition possibility of each of the images is determined. If a search for all the N images input from the camera does not find an image in which the face is recognizable, the process proceeds to the next full area search.

Before that, the N images used for the partial region search may be arranged in the image quality order, and a verification procedure following the image having the highest image quality may be executed. In this case, the elements used for the image quality comparison may include resolution, brightness, sharpness, illumination direction, and the like.

In the full region search procedure, a face element is detected for the entire input image, the extracted face element is verified, and a recognition possibility is determined without undergoing a front face detection and a bowed face detection procedure. This procedure extracts and verifies facial elements over the entire image, which can take a long time and put a heavy load on the system's throughput, but enables accurate detection.

If at least one eye and one mouth are found in the whole area search, the face recognition possibility is determined to be valid, and if not, the search procedure is performed on another image.

In spite of searching for all N images, if no recognizable face is found, it is determined that the face recognition is impossible and finally the financial transaction is blocked.

Claims (8)

As a method of determining the face recognition possibility of a system linked with an ATM machine:
Confirming the user's access to the atm machine or the commencement of the manipulation;
Photographing a user's face to obtain N images;
Selecting any one of the N images and extracting at least two candidate face elements of the user's face from the selected images;
Setting a region of interest in each of the extracted candidate face elements;
Connecting the set regions of interest to form inverse triangle shapes;
Extracting a face element from each inside of the ROI set in an inverted triangle shape according to a normal face proportion among the configured inverted triangle shapes; And
If the extracted face element is at least one eye and one mouth, allowing subsequent manipulation of the user; otherwise, blocking the manipulation. delete The method of claim 1,
When the face element extracted from the ROIs set in the inverse triangle shape according to the normal face proportion does not include at least one eye and one mouth,
Removing the region of interest and re-extracting any facial elements from the entirety of the selected image; And
And determining whether any one of the re-extracted face elements is extracted with eyes and mouth. The method of claim 1,
If at least one eye and one mouth cannot be extracted from the selected image, selecting another one of the N images and repeating the step of extracting the face element from the another image. Face recognition possibility determination method, characterized in that further performing. The method of claim 1,
Extracting the face element,
In order to verify false detection of the extracted face elements (at least one eye and one mouth), a feature vector is extracted using a principal component analysis method or a local binary pattern on the extracted face element image, and the extracted And applying the feature vector to the support vector machine classifier. The method of claim 1,
The region of interest includes a glasses / sunglasses region. The method according to claim 6,
And when the glasses are extracted from the glasses / sunglasses area, one eye is considered to be extracted, and when the sunglasses are extracted from the glasses / sunglasses area, the eyes are not extracted. To determine the likelihood of facial recognition. As a method of determining the face recognition possibility of a system linked with an ATM machine:
Confirming the initiation of the user's access or manipulation of the atm machine;
Photographing a user's face to obtain N images;
Selecting one of the N images, and extracting at least two candidate face elements corresponding to face elements of eyes, nose, and mouth from the selected images;
Setting a region of interest including the extracted candidate face elements;
Connecting the regions of interest to form a triangular pyramid shape;
When the triangular pyramid shape is a triangular pyramid shape according to a normal face proportion, the triangular pyramid shape is used to determine a direction that the face of the user faces, and extracts a face element from the inside of the set ROI based on the determined direction. Doing; And
And allowing the subsequent operation of the user if the extracted face element is at least one eye and one mouth, and if not, blocking the subsequent operation.

Images