KR101038706B1 - Method and apparatus for authenticating image - Google Patents

Abstract

The present invention discloses an image authentication method and apparatus. The present invention registers a plurality of images of the same user to reflect various environmental changes, and sets authentication criteria for each registered image by reflecting the correlation between the characteristic quantity distribution at the time of self referencing and the characteristic quantity distribution at the time of referencing another person. Improved image authentication performance. To this end, the present invention converts the registered images into LBP (Locally Binary Pattern) images, and for each LBP image, compares the registered LBP images of the same user with each other and compares the number of bits of the pixel value of each pixel at the time of self-referencing. Calculate the distribution, compare the comparison with the registered LBP images of other users, calculate the distribution of the number of coincidence bits of the pixel value of each pixel when referencing others, and then compare the distribution of the coincidence bits with self reference and the distribution of coincidence bits when referencing others. By using this, the threshold number of coincidence bits of each pixel of the registered LBP image is set. Thereafter, the authentication image input for image authentication is converted into an LBP image, compared with the registration image for each pixel unit, and authentication is performed by applying a threshold set for each pixel of the registration image.

Description

Image authentication method and apparatus {Method and apparatus for authenticating image}

The present invention relates to an image authentication method and apparatus, and more particularly, to an image authentication method and apparatus using an LBP image.

As the performance of a computer improves with the development of information and communication technology, image recognition and image authentication technology, which has been difficult to apply in the real life because a lot of computation time is required, is gradually applied to the real life. Representative examples used in real life include taking a picture of a user's fingerprint or iris, and opening and closing the door by authenticating the user compared to a registered user's fingerprint or iris image, and recently recognizing a user's face. It is expanding to the way.

However, the conventional image recognition and image authentication method adopts a method of simply comparing a photographed image for authentication according to a uniform standard with a previously registered image, so that the registered user is not recognized correctly, and other registered users or Malfunctions that are recognized as outsiders often occur.

For example, in the case where image authentication is performed by setting the similarity criterion of a uniform image without considering the characteristics of registered users, in a group with a large number of users having similar biological characteristics (face, fingerprint, iris, etc.). The result of performing image recognition and authentication is more likely to cause malfunctions than the result of performing image recognition and authentication in a group of users with similar biometric characteristics.

In addition, since the conventional technology performs image authentication based on only one registered image of each user, there is a problem that accurate image recognition and authentication cannot be performed when the environment for generating an image for image recognition and authentication is changed.

The problem to be solved by the present invention is to provide an image authentication method and apparatus capable of accurate image recognition and authentication by adaptively changing the authentication criteria used for image authentication in consideration of the characteristics of the target group to perform the image authentication will be.

The image authentication method of the present invention for solving the above problems comprises the steps of: (a) generating a plurality of LBP images for each of the target users to register in the database; (b) comparing registered LBP images of the same user with each other to calculate a distribution of coincidence bits of pixel values of each pixel upon self-referencing; (c) comparing registered LBP images of different users with each other to calculate distribution of coincidence bits of pixel values of each pixel in reference to another person; (d) setting and storing the threshold number of coincidence bits of each pixel of the registered LBP image by using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference for each pixel of each registered LBP image. ; (e) generating an authentication LBP image by converting the authentication user image into an LBP image when an authentication user image is input; And (f) comparing the authentication LBP image with each registered LBP image stored in the database on a pixel basis, and generating an authentication score according to the number of coincidence bits and the coincidence bit threshold.

Further, in the step (d), the probability that the owner of the image is recognized as the other person and the probability that the other person is recognized as the owner of the image is calculated for each matching bit number of the pixel value, and the difference between the two probabilities is the smallest number of matching bits. It can be set to the match bit threshold.

Further, in the step (d), the probability that the owner of the image is recognized as the other person increases as the threshold of the number of coincidence bits of the pixel value increases, and the probability that the other person is recognized as the owner of the image is the same bit of the pixel value. It may decrease as the threshold of the number increases.

Further, in the step (f), each pixel of the authentication LBP image is compared with a corresponding pixel of each registered LBP image, and authentication is performed according to the number of pixels whose number of matching bits is equal to or greater than the threshold number of matching bits of the corresponding pixel of the registration image. You can generate a score.

Further, in the step (f), each pixel of the authentication LBP image is compared with the corresponding pixel of each registered LBP image to calculate the number of coincidence bits for each pixel, and a graph connecting the number of coincidence bits for each pixel, An authentication score may be generated by calculating a similarity with a graph connecting the coincidence bit number threshold of the registered LBP image.

In addition, if an LBP image is additionally registered in the database, the matching bit number threshold may be updated by performing steps (b) to (d) on all LBP images registered in the database.

On the other hand, the image authentication apparatus of the present invention for solving the above problems, the database for storing the LBP image of the user and the threshold value for each pixel of each LBP image; An LBP conversion unit for converting images of users into LBP images and registering them in the database, and converting user images for authentication into LBP images and outputting them; Comparing registered LBP images of the same user with each other, calculating the distribution of the number of coincidence bits of pixel values of each pixel in self-referencing, and comparing registered LBP images of different users with each other, matching the pixel values of each pixel in reference to others A coincidence bit number distribution calculator for calculating a bit number distribution; For each pixel of each registered LBP image, a coincidence bit number threshold of each pixel of the registered LBP image is set and registered in the database using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference. A threshold setting unit; And performing authentication by comparing the LBP image for authentication input from the LBP conversion unit with each registered LBP image stored in the database on a pixel basis, and generating an authentication score according to the number of matching bits and the threshold number of matching bits. Contains wealth.

In addition, the threshold setting unit calculates the probability that the owner of the image is recognized as the other person and the probability that the other person is recognized as the owner of the image by the number of matching bits of the pixel value, and counts the number of matching bits having the smallest difference between the two probabilities. Can be set as a threshold.

In addition, the probability that the owner of the image is recognized as another increases as the threshold of the number of coincidence bits of the pixel value increases, and the probability that the other is recognized as the owner of the image increases as the threshold of the number of coincidence bits of the pixel value increases. Thus it can be reduced.

In addition, the authentication unit compares each pixel of the LBP image for authentication with a corresponding pixel of each registered LBP image, and generates an authentication score according to the number of pixels whose number of matching bits is equal to or greater than a threshold number of matching bits of the corresponding pixel of the registration image. Can be.

In addition, the authentication unit compares each pixel of the authentication LBP image with the corresponding pixel of each registered LBP image, calculates the number of matching bits for each pixel, and calculates a graph connecting the number of matching bits for each pixel, the registered LBP image The authentication score may be generated by calculating the similarity with the graph of the concatenation of the number of coincidence bits.

The present invention registers a plurality of images of the same user to reflect various environmental changes, and sets authentication criteria for each registered image by reflecting the correlation between the characteristic quantity distribution at the time of self referencing and the characteristic quantity distribution at the time of referencing another person. Improved image authentication performance.

To this end, the present invention converts the registered images into LBP (Locally Binary Pattern) images, and for each LBP image, compares the registered LBP images of the same user with each other and compares the number of bits of the pixel value of each pixel at the time of self-referencing. Calculate the distribution, compare the comparison with the registered LBP images of other users, calculate the distribution of the number of coincidence bits of the pixel value of each pixel when referencing others, and then compare the distribution of the coincidence bits with self reference and the distribution of coincidence bits when referencing others. By using this, the threshold number of coincidence bits of each pixel of the registered LBP image is set. Thereafter, the authentication image input for image authentication is converted into an LBP image, compared with the registration image for each pixel unit, and authentication is performed by applying a threshold set for each pixel of the registration image.

Thus, the present invention registers a plurality of images for each user, compares the self-referencing image in the group with another reference image on a pixel basis, and correlates the feature amount distribution at the time of self-referencing with the feature amount distribution at the time of another reference. By setting the threshold value of each pixel reflecting this and repeating this process each time a registered image is added, the accuracy of image authentication is improved by adaptively changing the threshold value of each image.

Before describing the image authentication method and apparatus according to the preferred embodiment of the present invention, the concept of image authentication of the present invention will be described with reference to FIGS.

1 is a view illustrating a local binary pattern (LBP) used for extracting relative feature quantities from an image in the present invention. Referring to FIG. 1, LBP is a method frequently used for texture analysis, and an output value is determined by a relative value of a feature amount between a pixel of interest and a peripheral pixel. 1 conceptually describes an LBP, where the LBP is information indicating a relative value of a feature amount with a pixel around it for one pixel of interest Pij. For a 3 * 3 pixel block containing one attention pixel extracted from the image and eight neighboring pixels adjacent to the pixel, assign 1 to the upper left pixel and 0 to the smaller left pixel. The MSB outputs a bit block consisting of 8 bits while rotating clockwise. LBP in the example of FIG. 1 becomes 11010011, which corresponds to decimal 229.

2 illustrates an example in which eyes are extracted from a face image input for image authentication, only a face region is extracted, and a face region image is converted into an LBP image. The similarity of the images can be measured by calculating the coincidence of the bit array of the bytes (pixels) of each LBP image thus obtained.

를 이미지 A 및 B의 유사도 스코어로서 채택할 수도 있다.3 is a diagram illustrating the concept of calculating the degree of agreement of the LBP between the reference image and the referenced image used in the present invention. Referring to FIG. 3, the LBP of the pixel Aij located in the i row j column of the reference image A shown on the left side of FIG. 3 is 10011001, and is located in the i row j column of the referenced image B shown on the right side of FIG. 5. The LBP of the pixel Bij becomes 00011100 so that the second, third, fourth, fifth, and seventh bits coincide with each other, and thus the similarity score Cij of the corresponding pixel becomes five. This process is then applied to all pixels of the reference image and the referenced image

May be adopted as the similarity score of images A and B.

However, the experiments confirmed that the misrecognition occurred frequently because these methods did not sufficiently reflect the various factors of face variation. Accordingly, the present invention improves the recognition rate by determining the similarity using the degree of coincidence distribution in self-referencing and the degree of coincidence distribution in others.

4 is a diagram illustrating a concept of setting a threshold value for each pixel of each image according to the present invention. Referring to FIG. 4, in the case where a plurality of images (LBP images) for B, C, and D are registered in advance, when registering a new user A, first, a plurality of LBP images are generated for the new user A. do. In the example shown in FIG. 4, five LBP images were generated for each user.

Thereafter, each LBP image of the user to be newly registered is compared with another LBP image of the same user to calculate a distribution of feature amounts when referring to the self image.

In addition, by comparing each LBP image of the user to be newly registered with each LBP image of the users B, C, and D already registered, the feature amount distribution in reference to others is calculated. At this time, the LBP images of the users B, C, and D previously registered are also compared with the LBP images of the user A to be newly registered, and the feature amount distribution at the reference of another person is recalculated.

In the present invention, the number of coincidence bits of the LBP value of each pixel is used as the feature amount distribution. The LBP image registered in the database is called DIC, the user number (personal number) is i, the LBP image number of the user is j, and the j th LBP image of a specific person i is called DICij, and is shown in FIG. As described above, when the degree of coincidence between DICij and DIC i'j 'is determined, the number of coincidence bits can be arranged in a form as shown in Table 1 below.

Pixel number One 2 3 4 5 6 7 .. .. 1278 1279 1280 Number of bits matched 8 3 5 7 3 4 5 .. .. 7 8 8

The distribution of the number of coincidence bits arranged as shown in Table 1 is shown in FIGS. 5 and 6. 5 shows the distribution of the number of coincidence bits when the images of the same person are compared with each other, and FIG. 6 shows the distribution of the number of coincidence bits when the images of others are compared with each other. In the graphs shown in Figs. 5 and 6, the horizontal axis represents pixels and the vertical axis represents the number of coincidence bits.

As shown in FIG. 5 and FIG. 6, when the number of coincidence bits is 5, the number of pixels whose coincidence bits exceed the criterion in self-referencing is greater than the number of pixels in which the coincidence bits exceed the criterion when referring to others. It can be seen that there is a great deal, and accordingly, it is possible to grasp the correlation between the bit agreement degree at the time of self-referencing and the reference to others.

According to the method described above, LBP images of the same user are compared with LBP images 2 to 5, respectively, to calculate the number of matching bits for each pixel, and the corresponding data for each pixel is calculated for each pixel using the calculated data. If the probability of coincidence bit number is found, the result shown in Table 2 can be obtained (for example, comparing DIC11 and DIC12 to DIC15, respectively).

 Pixel number 0bit 1 bit 2 bit 3bit 4 bit 5 bit 6 bit 7bit 8bit One 0 0 0 0 0 0.25 0.5 0 0.25 2 0 0 0 0 0 0 0.5 0.5 0 3 0 0 0 0 0 0 0 0.75 0.25 4 0 0 0 0 0 0.25 0 0.25 0.5 5 0 0 0 0 0 0.25 0.25 0.25 0.25  6 0 0 0 0 0 0 0 0.5 0.5 ... ... ... ... ... ... ... ... ... ... 1275 0 0 0 0 0 0.5 0 0.5 0  1276 0 0 0 0 0 0 0 0.75 0.25 1277 0 0 0 0 0 0.25 0.25 0.25 0.25 1278 0 0 0 0 0 0.5 0.5 0 0 1279 0 0 0 0 0 0 0 0.75 0.25 1280 0 0 0 0 0 0 0.25 0.75 0

In addition, if the probability of the corresponding number of coincidence bit occurs for each coincidence bit at the time of reference of the other person in the same manner, the result shown in Table 3 can be obtained. compare)

Pixel number 0bit 1 bit 2 bit 3bit 4 bit 5 bit 6 bit 7bit 8bit One 0 0 0.15 0.1 0.2 0.25 0.15 0.1 0.05 2 0 0 0 0.3 0.1 0.35 0.2 0.05 0 3 0 0 0.1 0.25 0.35 0.25 0.05 0 0  4 0 0.1 0.05 0.35 0.25 0 0.15 0.05 0.05  5 0 0.05 0.15 0.25 0.35 0.05 0.15 0 0  6 0 0.05 0.15 0.35 0.2 0.15 0.1 0 0 ... ... ... ... ... ... ... ... ... ...  1272 0.05 0.3 0.3 0.05 0.05 0.15 0.1 0 0  1273 0 0.05 0.6 0.05 0.1 0.1 0.05 0 0.05  1274 0 0.1 0.15 0.35 0.2 0.15 0.05 0 0  1275 0.05 0.2 0.3 0.25 0.05 0.1 0.05 0 0  1276 0 0 0.25 0.35 0.3 0.1 0 0 0 1277 0 0 0.25 0.45 0.25 0.05 0 0 0 1278 0 0 0.25 0.3 0.4 0.05 0 0 0 1279 0 0 0.25 0.35 0.25 0.1 0.05 0 0 1280 0 0.05 0.2 0.2 0.25 0.15 0.1 0 0.05

Referring to Tables 2 and 3, the probability of occurrence of the corresponding number of bits for each pixel is displayed graphically as shown in FIG. In the graph of FIG. 7, the horizontal axis represents the number of coincidence bits, and the vertical axis represents the probability of occurrence of the coincidence bit number.

Referring to FIG. 7, in the case of self-referencing, the probability that the number of coincidence bits is 5 or more is significantly higher than the probability that the number of coincidence bits is less than 5, but in the case of another reference, the probability that the coincidence bit number is 5 or more is less than 5. .

Based on the meaning of the graph shown in FIG. 7, the results shown in Table 4 below may be derived for each pixel.

 Match Bits Threshold FRR FAR ABS (ERR) 0 0.0 One One One 0.05 0.69 0.64 2 0.13 0.66 0.53 3 0.16 0.61 0.45 4 0.27 0.45 0.18 5 0.31 0.38 0.07 6 0.43 0.24 0.19 7 0.78 0.12 0.66 8 One 0.0 One

In Table 4, the FRR (False Rejection Rate) indicates the probability that the owner of the image will be rejected and rejected (hereinafter referred to as rejection), and the FAR (False Acceptance Rate) indicates the probability that another person will be mistaken as the owner of the image. Indicates. For example, if the match bits threshold is set to 5 bits (that is, authentication is matched only if the number of match bits is 5 bits or more), the probability that the owner of the image is rejected is 31%, and the other person is mistaken as the owner of the image. Your chances are 38%.

On the other hand, in Table 4, the probability that the owner of the image is recognized by another person increases as the threshold of the number of coincidence bits of the pixel value increases. When the coincidence bit number threshold is 1, since the coincidence bit count of most pixels is greater than or equal to the threshold value, most pixels of the authentication target image are authenticated. Therefore, it is very likely that the image of another person will be mistaken as the owner's image, but it is very unlikely that the owner of the image will be rejected.

On the other hand, when the coincidence bit number threshold is 7, since the coincidence bit number of most pixels is less than the threshold value, a large number of pixels of the authentication target image fail to authenticate. In other words, if the number of coincidence bits is 0 to 6, authentication fails. Therefore, it is very unlikely that the image of another person will be mistaken as the owner's image, but it is also highly likely that the owner of the image will be rejected.

Meanwhile, in Table 4, the probability that another person is mistaken as the owner of the image decreases as the threshold of the number of coincidence bits of the pixel value increases. When the coincidence bit number threshold is 1, since the coincidence bit count of most pixels is greater than or equal to the threshold value, most pixels of the authentication target image are authenticated. Therefore, it is very unlikely that the owner of the image will be rejected, but it is very likely that the image of another person is mistaken for the owner's image.

On the other hand, when the coincidence bit number threshold is 7, since the coincidence bit number of most pixels is less than the threshold value, a large number of pixels of the authentication target image fail to authenticate. In other words, if the number of coincidence bits is 0 to 6, authentication fails. Therefore, there is a high possibility that the owner of the image is misunderstood as another. In addition, it is very unlikely that another person's image will be misunderstood as the owner's image.

In Table 4, ABS (ERR) represents the magnitude of the difference between FRR and FAR.

8 is a graph conceptually showing the contents described in Table 4. FIG. Referring to FIG. 8, it can be seen that the coincidence bit threshold of the FRR and the FAR are sufficiently low that the coincidence bit threshold of the two graphs is about 5 bits, and thus the coincidence bit threshold of the corresponding pixel can be set as 5. have.

The above process is set for each pixel of every image. Therefore, every LBP image registered in the database has a threshold number of coincidence bits for each pixel by the number of pixels. The threshold number of coincidence bits of each of these images is preferably configured in a threshold table and stored in association with the image as shown in Table 5.

Pixel number One 2 3 4 5 6 7 .. .. .. 1279 1280 Threshold 7 5 5 6 4 5 5 .. .. .. 6 6

As described above, the process of setting the threshold of the number of coincidence bits for each pixel of each image is repeatedly performed every time a new user is registered, and the thresholds of all the pixels of all the images registered in the database are new. It is updated to reflect the registered image.

Meanwhile, after the coincidence bit threshold is set for each pixel of all registered images through the above-described process, when an image for authentication is input, the input image is converted into an LBP image, and each pixel of the LBP image for authentication is registered for each pixel. The number of coincidence bits is calculated by comparing with the corresponding pixels of the LBP image, and the ratio of "positive" to "positive" when the number of coincidence bits is greater than or equal to the threshold value of each pixel of the registered LBP image is "negative". Similarity can be calculated by For example, when the number of pixels having the same number of matching bits or more among the pixels is 80% or more, it may be determined as an image of the same person and authenticated.

In addition, as shown in FIG. 9, a unique waveform 901 is generated by connecting the threshold values set for each pixel of the registered LBP image, and comparing the input LBP image with the corresponding registered LBP image to calculate the number of coincidence bits. After generating the waveform 902 by connecting the number of matching bits, authentication of the input image may be performed by comparing similarities of the LBP waveforms using an algorithm such as DP (Dynamic Programming) Matching.

In addition, it will be understood by those skilled in the art that various ways of authentication may be performed using the thresholds of each pixel in the registered LBP image.

The authentication concept of the present invention has been described so far. Hereinafter, an image authentication apparatus and method according to a preferred embodiment of the present invention for implementing the concept of the present invention described above with reference to FIGS. 10 and 11 will be described. However, since the basic functional description has been described above with reference to FIGS. 1 to 9, a detailed description thereof will be omitted.

10 is a block diagram showing a configuration of an image authentication apparatus 1000 according to a preferred embodiment of the present invention. Referring to FIG. 10, the image authentication apparatus 1000 according to the present invention includes a preprocessor 1010, an LBP converter 1020, a feature variable distribution calculator 1030, a threshold value setting unit 1040, and an authentication unit 1060. And a database 1050.

First, the preprocessing unit 1010 is a configuration generally used in the image authentication apparatus 1000. The preprocessing unit 1010 extracts a pattern (for example, a face, a fingerprint, an iris, etc.) to be authenticated from an image input from a camera or the like, and then applies the pattern. Generates and outputs an image of a constant size included. In addition, the preprocessor 1010 may perform various image processing generally used in pattern recognition, such as adjusting the contrast and lighting effect of the image.

The LBP converter 1020 converts an image input from the preprocessor 1010 into an LBP image and outputs the converted LBP image. If the input image is an image for registration, the LBP conversion unit 1020 outputs the converted LBP image to the feature variable distribution calculation unit 1030 and the database 1050, and in the case of the image for authentication, the LBP image. Convert to and output to the authentication unit 1060.

The characteristic quantity distribution calculating unit 1030 calculates the characteristic quantity distribution at the time of self-referencing and the characteristic quantity distribution at the time of another reference, and outputs it to the threshold value setting unit 1040. The feature distribution distribution calculation unit 1030 according to the preferred embodiment of the present invention compares the registered LBP images of the same user with each other, calculates the distribution of the number of coincidence bits of the pixel value of each pixel upon self-referencing, and registers different users. The LBP images are compared with each other, and the distribution of the number of coincidence bits of the pixel values of each pixel is calculated and then output to the threshold setting unit 1040.

The threshold setting unit 1040 calculates the coincidence bit threshold of each pixel of the registered LBP image by using the coincidence bit distribution in self-reference and the coincidence bit distribution in another reference for each pixel of each registered LBP image. It is set and output to the database 1050. The method of setting the threshold is as described above with reference to FIGS. 5 to 8.

The authentication unit 1060 compares the authentication LBP image input from the LBP conversion unit 1020 with each registered LBP image stored in the database 1050 in units of pixels, and authenticates the corresponding bit number and the bit number threshold. Generate scores to perform authentication.

In a preferred embodiment of the present invention, the authentication unit 1060 calculates the number of matching bits by comparing the LBP image for authentication input from the LBP conversion unit 1020 with each registered LBP image stored in the database 1050 in units of pixels. When the number of coincidence bits is equal to or greater than the bit number threshold set for each pixel of the comparison target registered image, authentication may be performed by counting the number of pixels determined as "positive" and counting the number of pixels determined as "positive". That is, when the number of pixels whose coincidence bit count is equal to or greater than the threshold is equal to or greater than a predetermined ratio of the total pixel count, the authentication image and the registered image may be determined to be the same person.

The database 1050 stores LBP images of users and threshold values set for each pixel of each LBP image. The coincidence bit number threshold set for each image may be stored in association with the corresponding image as a threshold table. In this case, the database 1050 includes an image storage unit 1052 and a threshold storage unit 1054, the image storage unit 1052 stores the registered images, and the threshold storage unit 1054 stores the respective images. A threshold table of registered images can be stored.

11A is a flowchart illustrating a user image registration process according to a preferred embodiment of the present invention, and FIG. 11B is a flowchart illustrating an image authentication method according to a preferred embodiment of the present invention.

First, in order to use the image authentication method of the present invention, images of registered users are registered in advance in a database. Referring to FIG. 11A, a database registration process will be described. The image authentication apparatus 1000 receives an image of users (registration target users) who want to use the image authentication service of the present invention, converts the image into LBP images, and converts the plurality of LBPs. An image is generated and registered in the database 1050 (S1110).

When the LBP images of registered users are registered, the image authentication apparatus 1000 compares the registered LBP images of the same user with each other, and calculates a distribution of the number of coincidence bits of pixel values of each pixel at the time of self-referencing (S1120).

In addition, the image authentication apparatus 1000 compares registered LBP images of different users with each other, and calculates a distribution of the number of coincidence bits of pixel values of each pixel when referring to another person (S1130).

Thereafter, the image authentication apparatus 1000 uses, for each pixel of each registered LBP image, the coincidence bit number threshold of each pixel of the registered LBP image by using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference. The value is set and stored (S1140). At this time, a threshold table that stores the threshold of each pixel may be generated and stored in association with the LBP image.

Steps S1110 to S1140 described above are performed to set threshold values for image registration and authentication of users in the database 1050.

Meanwhile, a process of additionally registering or authenticating a user in a database will be described with reference to FIG. 11B.

Referring to FIG. 11B, when the image authentication apparatus of the present invention receives an image (S1152), the image authentication apparatus 1000 checks whether the input image is an image for authentication or a registration image for new user registration. (S1154).

If the input image is for new user registration, a plurality of LBP images are generated and registered for the new registered user (S1162), and the LBP images of the new registered user are compared with each other, and the pixel values of each pixel in self-reference are matched. The bit number distribution is calculated (S1164).

In addition, the image authentication apparatus 1000 compares registered LBP images of different users with each other, and calculates a distribution of the number of coincidence bits of pixel values of each pixel when referring to another person (S1166).

Thereafter, the image authentication apparatus 1000 uses, for each pixel of each registered LBP image, the coincidence bit number threshold of each pixel of the registered LBP image by using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference. The value is set and stored (S1168).

On the other hand, if the input image is an image for authentication, the authentication user image is converted into an LBP image to generate an LBP image for authentication (S1170).

Thereafter, the number of matching bits is calculated by comparing the authentication LBP image stored in the database 1050 with each registered LBP image in units of pixels (S1180). When the number of matching bits of each pixel is calculated, the number of matching bits and the authentication LBP is calculated. Authentication is generated by generating an authentication score according to the match bit number threshold set for each pixel of the registered LBP image with which the image is compared (S1190). As described above, a method of performing authentication by using the coincidence bit number and the coincidence bit threshold value uses the ratio of the number of pixels in which the coincidence bit count is equal to or greater than the threshold value in all the pixels, or the waveform of the coincidence bit number and the coincidence bit threshold. A method of measuring similarity by comparing waveforms of values may be used, and various methods may be applied.

The present invention can also be embodied as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, which are also implemented in the form of a carrier wave (for example, transmission over the Internet). It also includes. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

So far I looked at the center of the preferred embodiment for the present invention. Those skilled in the art will appreciate that the present invention can be implemented in a modified form without departing from the essential features of the present invention. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. The scope of the present invention is shown in the claims rather than the foregoing description, and all differences within the scope will be construed as being included in the present invention.

1 is a view illustrating a local binary pattern (LBP) used for extracting relative feature quantities from an image in the present invention.

FIG. 2 is a diagram illustrating an example of detecting an eye from a face image input for image authentication, extracting only a face region, and converting a face region image into an LBP image.

3 is a diagram illustrating a concept of calculating the degree of agreement of LBP between a reference image and a referenced image used in the present invention.

4 is a diagram illustrating a concept of setting a threshold value for each pixel of each image according to the present invention.

5 is a diagram showing a distribution of coincidence bit numbers when images of the same person are compared with each other.

6 is a diagram showing a distribution of coincidence bit numbers when an image of another person is compared with each other.

FIG. 7 is a graph illustrating a probability of generating a corresponding number of corresponding bits for each pixel. FIG.

8 is a graph conceptually showing the contents described in Table 4. FIG.

FIG. 9 is a diagram illustrating an example of a method of performing authentication using the coincidence bit count and the coincidence bit threshold.

10 is a block diagram showing the configuration of an image authentication device according to a preferred embodiment of the present invention.

11A is a flowchart illustrating a user image registration process according to a preferred embodiment of the present invention, and FIG. 11B is a flowchart illustrating an image authentication method according to a preferred embodiment of the present invention.

Claims (12)

(a) generating a plurality of LBP images for each of the users to be registered and registering them in the database; (b) comparing registered LBP images of the same user with each other to calculate a distribution of coincidence bits of pixel values of each pixel upon self-referencing; (c) comparing registered LBP images of different users with each other to calculate distribution of coincidence bits of pixel values of each pixel in reference to another person; (d) setting and storing the threshold number of coincidence bits of each pixel of the registered LBP image by using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference for each pixel of each registered LBP image. ; (e) generating an authentication LBP image by converting the authentication user image into an LBP image when an authentication user image is input; And (f) comparing the authentication LBP image with each registered LBP image stored in the database on a pixel-by-pixel basis, and generating an authentication score according to the number of coincidence bits and the coincidence bit threshold, and performing authentication. Image authentication method. The method of claim 1, wherein in step (d), Calculates the probability that the owner of the image is recognized as the other person and the probability that the other person is recognized as the owner of the image by the number of coincidence bits of the pixel value, and sets the coincidence bit number having the smallest difference between the two probabilities as the coincidence bit threshold. Image authentication method. The method of claim 2, wherein in step (d), The probability that the owner of an image is recognized as another increases as the threshold of the number of coincidence bits of the pixel value increases, and the probability that another person is perceived as the owner of the image decreases as the threshold of the number of coincidence bits of the pixel value increases. The image authentication method characterized by the above-mentioned. The method of claim 1, wherein in step (f), Comparing each pixel of the LBP image for authentication with a corresponding pixel of each registered LBP image to generate an authentication score according to the number of pixels whose number of matching bits is equal to or greater than the threshold number of matching bits of the corresponding pixel of the registration image. Authentication method. The method of claim 1, wherein in step (f), Comparing each pixel of the LBP image for authentication with the corresponding pixel of each registered LBP image, calculating the number of coincidence bits for each pixel, and comparing the number of coincidence bits for each pixel, and calculating the coincidence bit threshold of the registered LBP image. And an authentication score is generated by calculating the similarity with the graph of the values. The method according to claim 1, wherein when LBP images are additionally registered in the database, steps (b) to (d) are performed on all LBP images registered in the database to update the coincidence bit number threshold. The image authentication method characterized by the above-mentioned. The recording medium which recorded the image authentication method of any one of Claims 1-6 with the program code which can be read by a computer, and is executable. A database for storing LBP images of users and threshold values of pixels of each LBP image; An LBP conversion unit for converting images of users into LBP images and registering them in the database, and converting user images for authentication into LBP images and outputting them; Comparing registered LBP images of the same user with each other, calculating the distribution of the number of coincidence bits of pixel values of each pixel in self-referencing, and comparing registered LBP images of different users with each other, matching the pixel values of each pixel in reference to others A coincidence bit number distribution calculator for calculating a bit number distribution; For each pixel of each registered LBP image, a coincidence bit number threshold of each pixel of the registered LBP image is set and registered in the database using the coincidence bit distribution in self-reference and coincidence bit distribution in another reference. A threshold setting unit; And An authentication unit for performing authentication by comparing the LBP image for authentication input from the LBP conversion unit with each registered LBP image stored in the database on a pixel basis, and generating an authentication score according to the number of matching bits and the threshold number of matching bits. Image authentication apparatus comprising a. The method of claim 8, wherein the threshold setting unit Calculates the probability that the owner of the image is recognized as the other person and the probability that the other person is recognized as the owner of the image by the number of coincidence bits of the pixel value, and sets the coincidence bit number having the smallest difference between the two probabilities as the coincidence bit threshold. An image authentication device. The method of claim 9, The probability that the owner of an image is recognized as another increases as the threshold of the number of coincidence bits of the pixel value increases, and the probability that another is perceived as the owner of the image decreases as the threshold of the number of coincidence bits of the pixel value increases. An image authentication device, characterized in that. The method of claim 8, wherein the authentication unit Comparing each pixel of the LBP image for authentication with a corresponding pixel of each registered LBP image to generate an authentication score according to the number of pixels whose number of matching bits is equal to or greater than the threshold number of matching bits of the corresponding pixel of the registration image. Authentication device. The method of claim 8, wherein the authentication unit Comparing each pixel of the LBP image for authentication with the corresponding pixel of each registered LBP image, calculating the number of coincidence bits for each pixel, and comparing the number of coincidence bits for each pixel, and calculating the coincidence bit threshold of the registered LBP image. And an authentication score is generated by calculating a similarity with a graph of values.

Images