KR101637229B1 - Apparatus and method for extracting feature point based on SIFT, and face recognition system using thereof - Google Patents

Abstract

The present invention relates to a SIFT-based image recognition technology, and a SIFT-based feature point extraction apparatus according to the present invention is capable of extracting a contrast threshold of a candidate pixel used for extracting a feature point corresponding to an input image A concentration threshold value calculation unit to calculate the concentration threshold value; And a feature point descriptor generation unit for generating a feature point descriptor using all candidate pixels having the density threshold value or more among the candidate pixels, thereby increasing the number of feature points to be extracted and thereby improving the image recognition rate do.

Description

[0001] SIFT-based feature point extraction apparatus and method, and face recognition system using the same [0002]

The present invention relates to an image recognition technology based on SIFT, and more particularly, to an apparatus and method for extracting feature points robust to a fluid environment by calculating an optimal concentration threshold value of a candidate pixel used for extracting feature points according to an input image, And methods.

In recent years, techniques for extracting features from images obtained using a camera have been widely used in the field of computer vision, such as personal authentication through face recognition, 3D reconstruction, and self-tracking.

The feature points of the image should be invariant to the size change and rotation transformation of the image. Also, it should be partially unaffected by changes in lighting and pose of the camera.

Conventional methods using PCA (Principal Component Analysis) or ICA (Independent Component Analysis) have a high recognition rate for general images having many edges, but they are suitable for application to facial images and endoscopic images which are difficult to extract feature points .

Particularly, existing methods have a problem that the face recognition rate in a limited situation is somewhat satisfactory but it is difficult to apply to a fluid situation. For example, there is a problem that the recognition rate is remarkably deteriorated not only for images with different sizes, but also for images including backgrounds.

SUMMARY OF THE INVENTION Accordingly, the present invention has been made in view of the above problems, and it is an object of the present invention to provide an image processing apparatus and method that can improve the recognition rate by increasing the number of extracted feature points by calculating an optimal density threshold value of a candidate pixel used for feature point extraction, A feature point extracting device.

A second object of the present invention is to provide a face recognition system using the feature point extraction apparatus.

A third object of the present invention is to provide an image processing apparatus and method capable of increasing the number of extracted feature points and improving the recognition rate by calculating an optimized concentration threshold value of a candidate pixel used for extracting feature points according to an input image based on SIFT And to provide a minutiae point extraction method.

A fourth object of the present invention is to provide a face recognition method using the feature point extraction method.

According to an aspect of the present invention, there is provided an apparatus for extracting feature points of an input image based on SIFT, the apparatus comprising: a density threshold value calculating unit for calculating a contrast threshold; And a minutiae point descriptor generating unit for generating a minutiae point descriptor using all candidate pixels having the concentration threshold value or more among the candidate pixels.

The density threshold value calculation unit calculates a density value having a maximum number of pixels through the histogram analysis of a DOG (Difference of Gaussian) image generated from the input image as the density threshold value.

The minutia descriptor generation unit generates the minutiae descriptor using the minutia information including the position, size, and azimuth information with neighboring pixels based on the candidate pixel having the concentration threshold value or more.

The feature point extracting apparatus may include a DOG image generating unit that generates a DOG image using a Gaussian pyramid of the input image; And a candidate pixel extracting unit for searching the maximum or minimum point of the DOG image to extract the entire candidate pixels.

According to another aspect of the present invention, there is provided a method for calculating a contrast threshold of a candidate pixel used for extracting a feature point based on SIFT, A feature point extracting unit for extracting feature points using candidate pixels having the density threshold value or more; And a face recognition unit for recognizing the input face image by matching feature points of the input face image with feature points of a face image stored in a database.

Wherein the feature point extracting unit extracts, from the input face image, a histogram of a Difference of Gaussian (DOG) image generated from the input face image to obtain a density value having the maximum pixel count as the density threshold value A threshold value calculation unit; And a minutiae descriptor generator for generating a minutiae point descriptor by using candidate pixels having the concentration threshold value or more among all candidate pixels.

In the single face image, the face recognition unit matches the input face image and the stored feature point of the stored face image through clustering by Hough Transform.

In the single-sided case, the face recognition unit recognizes the input face image in consideration of the number and distribution information of the matching feature points.

According to another aspect of the present invention, there is provided a method of extracting feature points of an input image based on SIFT in a computer system, the method comprising the steps of: A concentration threshold calculating step of calculating a concentration threshold; And a minutiae point descriptor generating step of generating a minutia point descriptor using all candidate pixels having the concentration threshold value or more as a candidate pixel.

According to a fourth aspect of the present invention, there is provided a method for recognizing facial images based on SIFT in a computer system, the method comprising the steps of: a feature point extracting step of calculating a contrast threshold of a candidate pixel and extracting a feature point using a candidate pixel having the density threshold value or more; And a face recognition step of recognizing the input face image by matching minutiae points of the input face image and minutiae points of a face image stored in a database.

The SIFT-based feature point extracting apparatus and method and the face recognition system and method using the SIFT according to the present invention calculate an optimal density threshold value of a candidate pixel used for extracting feature points according to an input image, By 30% or more, thereby improving the recognition rate.

Prior to the description of the concrete contents of the present invention, for the sake of understanding, an overview of the technical problem to be solved by the present invention will be given first.

The present invention is based on a Scale Invariant Feature Transform (SIFT) algorithm to create a robust face recognition system in a fluid environment.

SIFT, which is a representative algorithm for extracting feature points of images, is called "DG Lowe", "Distinctive image features from scale-invariant keypoints", Int. J. Comput. Vision 60 (2), 91-110, 2004. " SIFT has properties that are invariant to image rotation, scaling, translation, partial illumination changes, and projective transforms. The SIFT extracts attributes such as position, scale, and direction of the feature in consideration of the local image characteristic. That is, the first step is to select the sample point (or candidate pixel) by searching the maximum and minimum extrema in the scale space generated by the Difference-Of-Gaussian (DOG) function ( Scale - space extrema detection ) . As a second step, keypoints are selected based on the stability value ( Keypoint localization ) . As a third step, one or more directions are assigned to each keypoint ( Orientation assignment ) . As a final step, a keypoint descriptor is created using local image gradients ( Keypoint descriptor ) .

In particular, SIFT is used to find a fundamental matrix (F), which is the first step in finding three-dimensional information in face recognition. That is, feature points unchanged in size change are extracted by SIFT, and the basic matrix F is obtained using the feature points. In order to solve the one-to-one correspondence problem by eliminating outliers in the process of obtaining the basic matrix F from the minutiae points, keypoint matching, clustering by Hough Transform (clustering) step.

Through this process, the number of matching points and distribution information between the two images are grasped and the face is recognized.

However, since the simple SIFT algorithm applies the reference value for extracting the feature points fixed to any image, there is a limit in that the number of feature points that can be extracted is small in the face image and the endoscopic image having small features.

FIG. 1 shows feature points extracted by applying SIFT to two general images.

FIG. 2 shows feature points extracted by applying SIFT to two endoscopic images.

1 and 2 show the case where the contrast threshold of a candidate pixel used for extracting a feature point is set to 0.02 and the curvature threshold is set to 10.

As shown in FIG. 1, 485 and 414 feature points are extracted from the left image, respectively.

However, as shown in FIG. 2, 18 and 13 feature points are extracted from the left image for the endoscopic image, and it is found that there is an upper limit for the application of the feature image.

On the other hand, in order to overcome the limitation of the SIFT, the present invention appropriately calculates the reference value of the feature point extraction according to the input image, thereby increasing the number of feature points extracted and improving the recognition rate.

In FIG. 3, feature points extracted by applying SIFT with a density threshold of 0.01 and a curvature threshold of 10 are shown for two endoscopic images.

As shown in FIG. 3, 111 and 110 feature points are extracted from the left image, respectively, and the performance is improved.

FIG. 4 shows feature points extracted by applying a SIFT in which the density threshold value is set to 0.01 and the curvature threshold value is set to 50 for two endoscopic images.

As shown in FIG. 4, 126 and 123 feature points are extracted from the left image, respectively, and it can be seen that the feature points of the points seen as the boundary are slightly increased.

As can be seen from the results of FIGS. 3 and 4, the performance of the SIFT is greatly influenced by the concentration threshold value as compared with the curvature threshold value.

Therefore, the present invention intends to improve the face recognition rate by increasing the number of extracted feature points by automatically calculating an optimized concentration threshold value of a candidate pixel used for feature point extraction according to an input image.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to clarify the solution of the technical problem of the present invention. In the following description, well-known functions or constructions are not described in detail since they may obscure the subject matter of the present invention. In addition, the terms described below are defined in consideration of functions of the present invention, and may be changed according to intentions or customs of a user, an operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

FIG. 5 is a block diagram illustrating an example of a SIFT-based face recognition system according to the present invention.

FIG. 6 is a flowchart illustrating an example of a SIFT-based face recognition method according to the present invention.

5 and 6, a face recognition system 500 according to the present invention includes a feature point extraction unit 502 and a face recognition unit 504. The face recognition system 500 includes a camera unit 510, a video signal processor 520 And a face area detecting unit 530. [0033] FIG.

First, the camera unit 510 converts an optical signal of a subject into an electrical signal using an image sensor, such as a CCD sensor or a CMOS sensor (S610).

The image signal processing unit 520 converts the electrical analog image signal input from the camera unit 510 into a digital data image signal (S620).

The face region detection unit 530 searches for and extracts a face region from a video signal input from the video signal processing unit 520 (S630). For the implementation of the face area detection unit 530, a face extraction technique using a contour line, a skin color, a texture, a template, and the like may be used. For example, the face region detection unit 530 can search for a face region using the face color in the input image. The face region detection unit 530 includes a conversion module for enhancing facial color, a conversion module for enhancing black and white, a preprocessing module for enhancing an image of a specific portion to extract facial feature information, A background labeling module that separates the background area, and the like.

Next, the feature point extracting unit 502 calculates a contrast threshold of a candidate pixel based on SIFT, which is used for extracting feature points corresponding to the input face image, The feature points are extracted using the above-described candidate pixels (S640).

The feature point extracting unit 502 can be implemented by a feature point extracting apparatus and method according to the present invention.

FIG. 7 is a block diagram of a SIFT-based feature point extracting apparatus according to the present invention.

FIG. 8 is a flowchart illustrating a SIFT-based feature point extraction method according to the present invention.

7 and 8, the feature point extraction apparatus according to the present invention includes a DOG image generation unit 710, a candidate pixel extraction unit 720, a density threshold value calculation unit 730, and a minutiae point descriptor generation unit 740 .

The DOG image generation unit 710 generates a DOG image using a Gaussian pyramid of the input image (S810).

FIG. 9 shows a process of generating a DOG image using a Gaussian pyramid of an input image.

As shown in FIG. 9, the DOG image generating unit 710 passes the input image through a Gaussian filter having variance values of various scales, and sequentially subtracts the images according to adjacent scales. The resulting image is called a Difference Of Gaussian (DOG) image. The DOG image is subsampled at 2: 1 and passed through a Gaussian filter having various scales. Through this process, DOG images of pyramid structure can be obtained.

Next, the candidate pixel extracting unit 720 extracts sample points corresponding to the maximum or minimum of each DOG image, that is, all candidate pixels to be feature points (S820).

In Fig. 10, pixels of a DOG image to be compared are extracted to extract candidate pixels.

As shown in FIG. 10, the maximum or minimum position is extracted as the candidate pixel when the maximum or minimum position is maximum or minimum compared with 26 pixels in all, 8 pixels in the periphery and 9 pixels in the upper and lower scale DOG images.

Next, the density threshold value calculation unit 730 calculates a density threshold value of a candidate pixel used for extracting the minutiae corresponding to the input image (S830).

In the conventional SIFT, candidate pixels on the edge line of which the DOG value is small or strong are removed in order to remove the elements due to noise or the like. That is, the candidate pixels smaller than the density threshold value are removed as shown in Equation (1).

,

,

,

,

,

,

.

.

는 localization을 통해 정확성이 향상시킨 위치 정보, 그리고

는 localization된 픽셀에 대한 DOG를 의미한다.In Equation (1), D is a variable representing DOG, X is information on the position of the selected pixel,

Location information that improves accuracy through localization, and

Means a DOG for a localized pixel.

As shown in Equation (1), the smaller the density threshold value, the more feature points can be extracted. However, if the concentration threshold value is reduced so as to increase the number of minutiae extracted more than necessary, there arises a problem of inefficiency of calculation as well as a risk of mismatching.

값의 분포가 다르므로 입력 영상에 따라 적절한 농도 임계점을 산출할 필요가 있다.In addition,

Since the distribution of values is different, it is necessary to calculate an appropriate concentration threshold according to the input image.

However, the conventional SIFT has a problem of extracting feature points using fixed reference values regardless of the input image.

In contrast, the density threshold value calculator 730 of the present invention calculates a density value having a maximum number of pixels through a histogram analysis of a DOG (Difference of Gaussian) image generated from the input image, .

The histogram analysis refers to a method of analyzing the number of pixels having a corresponding density level for each level of the image, or the ratio of the number of pixels to the total number of pixels. This histogram analysis result can be displayed in a bar graph in which the ordinate axis indicates the density value and the ordinate axis indicates the number of pixels. The histogram analysis can be used to investigate which density pixels the image data is composed of.

On the other hand, the condition of the candidate pixel to be removed in relation to the curvature threshold r may be derived as shown in Equation (2).

As shown in Equation (2), in the case of the curvature threshold value, the larger the value of r, the smaller the number of feature points extracted at the boundary portion.

Next, the minutiae descriptor generator 740 generates a minutiae point descriptor using the candidate pixels of the concentration threshold value or more among all the candidate pixels (S840).

The minutiae descriptor generator 740 generates the minutiae descriptor using minutia information including position, size, and azimuth information with neighboring pixels on the basis of a candidate pixel having the concentration threshold value or more. That is, a 16x16 block is set around the finally selected candidate pixel, and the directionality is measured with respect to the pixels in the block. The direction and the size of each pixel are measured to estimate the direction that exists in the block. And sets this direction to the direction of the corresponding feature point. In order to obtain the final SIFT feature vector, a 16x16 block is set centering on each keypoint again, and again divided into 4x4 blocks, and the direction of change of pixel values is measured for each pixel in the block. The change direction is set to be proportional to the change slope, and a probability distribution with respect to the direction is obtained so that the probability distribution in the direction increases as the magnitude of the change increases.

FIG. 11 shows a process of extracting a directional distribution for 8 by 8 blocks and quantizing the 8 direction by 8 directions.

As shown in FIG. 11, the directional probability distribution is divided into eight directions and stored. As a result, the directional probability distribution is represented by a 128-dimensional vector form with respect to 16x16 blocks. The vector representing the directional probability distribution is a structure of the SIFT feature vector, and is stored together with parameters such as image coordinates, directionality, and Gaussian scale of the feature points. The 128-dimensional SIFT feature vector thus obtained is expressed as a codebook by performing vector quantization. The initially obtained SIFT feature vector is an arbitrary real number having a floating point, and a relatively large transmission capacity and time are required when transmitting the SIFT feature vector or comparing distances. Therefore, in order to overcome this disadvantage, vector quantization is performed on the 8-dimensional vector of the SIFT divided by each 8-direction unit, and the 8-dimensional directional distribution represented by the real value is associated with the 1-dimensional codebook. By performing such a vector quantization process, the real-valued SIFT feature vector of 128 dimensions can be converted into a vector of 16-dimensional integer coefficients. The SIFT feature vector quantized by the 16-dimensional vector is transmitted with low data amount, and when calculating the difference between the vectors, it is reduced to the 128-dimensional vector of the quantized codebook for comparison.

As described above, when the feature point extracting unit 502 extracts feature points from the input face image, the face recognizing unit 504 matches the feature points of the face image stored in the database 540 with the feature points of the input face image The input face image is recognized (S650).

That is, the face recognition unit 504 may perform clustering by Hough Transform to match feature points of the input face image and the stored face image, and may perform matching using the number of matching feature points, And recognizes the image having a large number of feature points as a face image. In addition, the face recognizing unit 504 recognizes not only the number information of the matching feature points but also the face image considering the distribution information of the matching feature point such as the position of the matching feature point, the length and slope of the line segment connecting the matching feature points, Can be recognized.

Particularly, when there is facial image information having a matching point that is greater than or equal to a predetermined reference value from the input facial image among the facial image information stored in the database 540 in the personal authentication, system security, and the like, the facial recognition unit 504 recognizes, And an authentication module (not shown) that acknowledges the access authority and denies the access authority otherwise.

In one embodiment, the SIFT-based feature point extraction apparatus and the face recognition system using the SIFT according to the present invention may be implemented as a system on chip on a microprocessor, Face recognition can be performed. When the present invention is implemented by a microprocessor, the size of various systems can be reduced, the assembly process can be simplified, and manufacturing costs can be reduced.

Meanwhile, the detailed process of the SIFT-based feature point extraction method and the face recognition method using the SIFT according to the present invention can be explained in correspondence with the detailed operation of the configuration modules of the feature point extraction device and the face recognition system.

In addition, the SIFT-based feature point extraction method and the face recognition method using the SIFT according to the present invention can be implemented as computer-readable program codes on a computer-readable recording medium. When the present invention is executed through software, the constituent means of the present invention are code segments for executing necessary operations. The program or code segments may be stored in a processor readable medium or transmitted by a computer data signal coupled with a carrier wave in a transmission medium or a communication network.

The computer-readable recording medium includes all kinds of recording apparatuses in which data that can be read by a computer system is stored. Examples of the computer-readable recording medium include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like. In addition, the computer-readable recording medium can be stored and executed in a computer-readable manner in a distributed manner by being distributed to a network-connected computer system.

Hereinafter, the remarkable effects of the present invention will be verified.

12A and 12B show feature points and matching points extracted by applying the present invention to a general image, respectively.

As shown in FIGS. 12A and 12B, the number of extracted feature points was 687 and 772, respectively, from the left, and the calculated concentration threshold value was 0.0077 and 0.0056 respectively from the left. In addition, 135 matching points were obtained.

13A and 13B show feature points and matching points extracted by applying the present invention to endoscopic images, respectively.

As shown in FIGS. 13A and 13B, the number of extracted feature points was 560 and 509, respectively, from the left, and the calculated concentration threshold value was 0.0031 and 0.0032 respectively from the left. Also, the number of matching points obtained through this study was 145.

As described above, according to the present invention, it is possible to extract more feature points regardless of the type of the image, and thus it is possible to find more matching points.

Figs. 14A to 18B show the face recognition result using the conventional method and the present invention.

14A and 14B show characteristic points extracted by the conventional method and the present invention, respectively.

15A and 15B show matching points obtained by applying the conventional method and the present invention, respectively.

Figs. 16A and 16B show the results of facial recognition using the conventional method and the present invention, respectively.

As shown in FIGS. 14A to 16B, according to the present invention, the number of feature points extracted from the existing method increases by an average of 30% or more, thereby obtaining more matching points and consequently enhancing the face recognition rate .

17A to 18B show the face recognition result obtained by applying the existing method and the present invention to a face image having a large change in facial expression.

17A and 17B show feature points extracted by applying the conventional method and the present invention to facial images having many facial expressions.

Figs. 18A and 18B show the face recognition result obtained by applying the existing method and the present invention to face images having a large change in facial expression, respectively.

As shown in FIGS. 17A and 18B, when the present invention is applied to a face image having a large change in facial expression, it is possible to extract more feature points, thereby increasing the face recognition rate.

The present invention has been described with reference to the embodiments. However, it will be understood by those skilled in the art that the present invention may be embodied in various other forms without departing from the spirit or essential characteristics thereof. Therefore, the disclosed embodiments should be considered in an illustrative rather than a restrictive sense. That is, the true technical scope of the present invention is indicated in the appended claims, and all differences within the scope of equivalents thereof should be construed as being included in the present invention.

FIG. 1 is a view showing feature points extracted by applying SIFT to two general images. FIG.

2 is a view showing feature points extracted by applying SIFT to two endoscopic images.

FIG. 3 is a view showing feature points extracted by applying a SIFT in which a density threshold value is set to 0.01 and a curvature threshold value is set to 10 for two endoscopic images. FIG.

FIG. 4 is a view showing feature points extracted by applying a SIFT in which a density threshold value is set to 0.01 and a curvature threshold value is set to 50 for two endoscopic images. FIG.

5 is a block diagram illustrating an example of a SIFT-based face recognition system according to the present invention.

6 is a flowchart illustrating an example of a SIFT-based face recognition method according to the present invention.

FIG. 7 is a block diagram illustrating a SIFT-based feature point extraction apparatus according to the present invention. FIG.

8 is a flowchart illustrating a SIFT-based feature point extraction method according to the present invention.

9 is a diagram illustrating a process of generating a DOG image using a Gaussian pyramid of an input image.

Figure 10 shows pixels of a DOG image compared to extract candidate pixels;

11 is a diagram illustrating a process of extracting a directional distribution for 8 by 8 blocks and quantizing the 8 direction by 8 directions.

12A and 12B are diagrams showing feature points and matching points extracted by applying the present invention to a general image, respectively.

13A and 13B are diagrams showing feature points and matching points extracted by applying the present invention to an endoscopic image, respectively.

Figs. 14A to 18B are diagrams showing a face recognition result using the existing method and the present invention; Fig.

Claims (20)

delete delete delete delete SIFT, a contrast threshold value of a candidate pixel used for extracting a feature point is calculated corresponding to an input face image, and a feature point for extracting a feature point using a candidate pixel having the density threshold value or more An extraction unit; And And a face recognition unit for recognizing the input face image by matching feature points of the input face image and feature points of a face image stored in a database, Wherein the feature point extraction unit stores the vector of the real number coefficients indicating the directional probability distribution measured based on the candidate pixels not less than the density threshold value into a vector of integer coefficients through vector quantization, Wherein the face recognizing unit comprises: If the minutiae of the facial image matching the minutiae of the input facial image stored in the database matches the minutiae of the input facial image and has a matching point of a predetermined reference value or more, And an authentication module for denying access authority to the face recognition system. 6. The method of claim 5, The feature point extracting unit may extract, A density threshold value calculation unit for calculating a density value having a maximum number of pixels through the histogram analysis of a Difference of Gaussian (DOG) image generated from the input face image to the density threshold value; And And a minutiae point descriptor generating unit for generating a minutiae point descriptor using all candidate pixels having the concentration threshold value or more among the candidate pixels. The method according to claim 6, Wherein the feature point descriptor generating unit generates the feature point descriptor using feature point information including position, size, and azimuth information with neighboring pixels based on a candidate pixel having the density threshold value or more, . The method of claim 6, wherein The feature point extracting unit may extract, A DOG image generation unit for generating a DOG image using a Gaussian pyramid of the input face image; And And a candidate pixel extracting unit for searching the maximum or minimum point of the DOG image to extract the entire candidate pixels. 6. The method of claim 5, Wherein the face recognition unit matches the feature points of the input face image and the stored face image through clustering by Hough Transform. 6. The method of claim 5, Wherein the face recognition unit recognizes the input face image in consideration of the number and distribution information of the matching feature points. delete delete delete delete A method for recognizing facial images based on SIFT in a computer system, SIFT, a contrast threshold value of a candidate pixel used for extracting a feature point is calculated corresponding to an input face image, and a feature point for extracting a feature point using a candidate pixel having the density threshold value or more Extraction step; A face recognition step of recognizing the input face image by matching minutiae points of the input face image and minutiae points of a face image stored in a database; And If the minutiae of the facial image matching the minutiae of the input facial image stored in the database matches the minutiae of the input facial image and has a matching point of a predetermined reference value or more, A step of setting an access right denying the access right; / RTI > The feature point extracting step may include: Wherein a vector of a real coefficient indicating a directional probability distribution measured on the basis of a candidate pixel not less than the density threshold value is converted into a vector of integer coefficients through vector quantization so as to be stored in correspondence with a codebook. 16. The method of claim 15, The feature point extracting step may include: A density threshold value calculation step of calculating a density value having a maximum number of pixels through the histogram analysis of a Difference of Gaussian (DOG) image generated from the input face image to the density threshold value; And A minutiae point descriptor generating step of generating a minutiae point descriptor by using a candidate pixel having a concentration threshold value or more among all candidate pixels. 17. The method of claim 16, Wherein the generating of the minutiae descriptor is a step of generating the minutiae descriptor using the minutia information including the position, size, and azimuth information with neighboring pixels based on the candidate pixel having the concentration threshold value or more. Face recognition method. The method of claim 16, wherein The feature point extracting step may include: A DOG image generation step of generating a DOG image using a Gaussian pyramid of the input face image before the density threshold value calculation step; And Further comprising a candidate pixel extracting step of searching for a maximum point or a minimum point of the DOG image and extracting the entire candidate pixels. 16. The method of claim 15, Wherein the face recognition step is a step of matching feature points of the input face image and the stored face image through clustering by Hough Transform. 20. A recording medium on which a program for executing a method according to any one of claims 15 to 19 is recorded in a computer system, the recording medium being readable by the computer system.

Images