KR20170077081A - System and method for object recognition based automatic picture color balance, recording medium thereof - Google Patents

Abstract

An object recognition-based automatic image color balance system and method and a recording medium therefor are disclosed. The disclosed automatic image color balance system includes an object database in which recognition information of objects and intrinsic reflection spectrum information are registered; An image acquisition unit receiving an image; An object recognition unit for recognizing an object in the input image using recognition information of the object database; A parameter extraction unit for extracting intrinsic reflection spectrum information of the recognized object from the object database; A gain calculator for calculating a gain value for color correction of the input image using the extracted eigenfrequency spectral information; And an image correction unit for correcting the input image based on the gain value. According to the disclosed system, an object can be recognized and color balance can be performed even if there is no white object, and an accurate color balance can be performed using an inherent reflection spectrum of an object.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to an object-recognition-based automatic image color balance system and method,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic image color balance system and method, and a recording medium therefor, and more particularly, to an automatic image color balance system and method based on object recognition and a recording medium therefor.

Conventional image color balance techniques are based on algorithms for finding regions or objects that are assumed to be white in the image. If an object or region estimated to be white in the image is recognized, white balance is performed on the entire image based on the recognized color of the object.

However, since it is assumed that a white subject is always present in the image, if there is no white subject in the image, the color correction can not be performed properly. The reliability is somewhat low.

One aspect of the present invention is to perform color balance on an entire image based on eigen-reflection spectrum information through object recognition even when there is no white subject.

Another aspect of the invention is to use a previously registered eigenfrequency spectrum to perform more accurate color balance than the prior art.

According to an aspect of the present invention, there is provided an object database in which recognition information of objects and intrinsic reflection spectrum information are registered. An image acquisition unit receiving an image; An object recognition unit for recognizing an object in the input image using recognition information of the object database; A parameter extraction unit for extracting intrinsic reflection spectrum information of the recognized object from the object database; A gain calculator for calculating a gain value for color correction of the input image using the extracted eigenfrequency spectral information; And an image correction unit for correcting the input image based on the gain value.

The recognition information of the object database includes a reference image; A feature point descriptor, and coordinates of at least one point of interest.

The object database may further include sensitivity information of the camera and power spectral information of the illumination.

The object recognition unit may generate a minutiae descriptor on the input image, compare the minutiae descriptor with the minutiae descriptors of the object database, and recognize the object having the most similar minutiae descriptor as the object in the input image .

Wherein the object recognition unit obtains a matrix representing a linear transformation relationship between an object in the input image and a reference image of an object recognized as the same object, and then coordinates the point of interest in the reference image of the object recognized as the same, And obtaining coordinates of a point of interest in the input image.

And the parameter extracting unit further extracts camera sensitivity information of the input image and power spectrum information of illumination from the object database.

Wherein the gain calculator calculates an ideal pixel value of a point of interest in the reference image of the object recognized as the same by using the extracted eigenfrequency spectral information, the camera sensitivity information, and the power spectrum information of the illumination, And a gain value for color correction of the input image is calculated by comparing the input pixel value with a pixel value of a point of interest in the input image.

Wherein the gain calculator calculates a gain value for each of the points of interest when the point of interest in the reference image of the object recognized as the same is greater than one and then calculates a gain value for color correction of the input image by averaging the values using the least squares method .

According to another embodiment of the present invention, there is provided an image processing apparatus including an image acquiring unit acquiring an image through photography; A communication unit that transmits the image acquired by the image acquisition unit to a balance adjustment server and receives a gain value for color balance from the balance adjustment server; And an image correction unit that corrects the image using the gain value. The balance adjustment server calculates an gain value for the color balance using the eigenfrequency spectrum information of the object after recognizing the object in the transmitted image And transmits the corrected image to the image correction unit.

According to another embodiment of the present invention, there is provided an object database in which recognition information and eigen-reflection spectrum information of objects are registered; A server communication unit for receiving an image; An object recognition unit for recognizing an object in the received image using recognition information of the object database; A parameter extraction unit for extracting intrinsic reflection spectrum information of the recognized object from the object database; And a gain calculator for calculating a gain value for color correction of the received image using the extracted eigenfrequency spectral information, wherein the calculated gain value is transmitted through the server communication unit A server is provided.

According to another embodiment of the present invention, there is provided a method of detecting an object, comprising: storing recognition information and eigen-reflection spectrum information of objects in an object database; Receiving an image; Recognizing an object in the received image using recognition information of the object database; Extracting intrinsic reflection spectrum information of the recognized object from the object database; And calculating a gain value for color correction of the received image using the extracted eigenfrequency spectral information and transmitting the calculated gain value.

According to another embodiment of the present invention, there is provided a computer-readable recording medium on which a program for performing the automatic image color balance method is recorded.

Since the present invention uses the eigenfrequency spectrum information of widely used objects, it is possible to perform more accurate color balance than the existing technology.

Further, there is an advantage that accurate color balance can be achieved even when there is no white subject.

와 전송된 영상

의 선형 변환 관계를 예시한 것이다.
도 5는 호모그래피 행렬

에 의해 워핑된 관심점을 예시한 것이다. 1 is a block diagram showing the configuration of an automatic image color balance system according to an embodiment of the present invention.
FIG. 2 is a flowchart showing the steps of an automatic image color balance method according to an embodiment of the present invention, according to the flow of time.
Fig. 3 shows the structure of the object database 1210. Fig.
Fig.

And transmitted video

As shown in FIG.
FIG. 5 is a diagram illustrating a homography matrix

Lt; RTI ID = 0.0 > of < / RTI >

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram showing the configuration of an automatic image color balance system according to an embodiment of the present invention.

As shown in FIG. 1, the automatic image color balance system may include an image correction device 1100 and a balance data server 1200. The image correction apparatus 1100 may include an image acquisition unit 1110, a communication unit 1120 and an image correction unit 1130. The balance data server 1200 includes an object database 1210, a server communication unit 1220, An object recognition unit 1230, a parameter extraction unit 1240, and a gain calculation unit 1250. [

Hereinafter, a process of recognizing an object in an input image and performing color balance using the automatic image color balance apparatus according to an exemplary embodiment of the present invention will be described in detail with reference to FIG.

The image correcting apparatus 1100 acquires an image through the image acquiring unit 1110. The image acquisition unit 1110 may be a camera module mounted on a digital camera, a smart phone, or a tablet PC. The communication unit 1120 transmits the acquired image to the server 1200 for color balance.

The server 1200 receives an image through the server communication unit 1220. The object recognition unit 1230 compares the object of the transmitted image with the object information stored in the object database 1210.

Fig. 3 shows the structure of the object database 1210. Fig.

As shown in FIG. 3, the object database 1210 stores reference images, minutiae descriptors, and points of interest information for object recognition. A unique ID is assigned to each object so as to be identifiable, and an image of the reference object is stored. The feature point descriptor information may be stored to identify the feature point of the object and the coordinates of the point of interest of the reference image may be stored for color balance of the image. In addition, the object database 1210 may also store inherent reflection spectra of stored objects, parameters of color balance, camera ID and sensitivity information, and power spectral information of illumination. The intrinsic reflection spectrum is an unchanging characteristic of an object and can be easily used for color balance. Also, the ID of the camera that captures the image, the corresponding sensitivity information, and the power spectrum of various lights may be stored.

The object recognizing unit 1230 recognizes an object included in the received image using the information of the object database 1210. [

Here, the object recognition means to determine which object among the objects of the object database 1210 in which various objects captured in the image are stored in advance.

Object recognition included in the image may be performed in various ways. Various known algorithms for object recognition exist and object recognition may be performed in any manner.

For example, the object recognition unit 1230 may recognize an object using a learning-based deep learning method. The object recognizing unit 1230 may set a feature point suitable for expressing a specific object through learning and store the feature point descriptor in the object database 1210 in advance.

Here, the minutiae descriptor is information that is a reference to distinguish a specific object from other objects. The minutiae descriptor may also be generated in a variety of ways and may be used in any manner.

A shift descriptor may be used as an example of the minutiae descriptor, and a shift descriptor may be created at the minutiae point of the object using the well-known Fast R-CNN deep processing method. The shift technician has the advantage that the object can be recognized even when the object moves or rotates. The object recognition unit 1230 may store the generated shift descriptors of the objects in the object database 1210 in advance.

The object recognition unit 1230 recognizes an object in the transmitted image using the established object database 1210.

The object recognizing unit 1230 will generate the minutiae descriptor in the transmitted image in the same manner as the method of constructing the minutiae descriptor in the object database 1210. [ The object recognizer 1230 may compare the generated minutiae descriptor with the minutiae descriptors of the objects in the object database 1210, and may find several objects with similar minutiae descriptors. In such a case, an object having the most similar minutiae descriptor among the relevant objects is found.

The object recognizing unit 1230 recognizes the object having the most similar minutiae descriptor as an object in the received image.

Next, the object recognition unit 1230 calculates a linear transformation relation between the reference image of the recognized object and the received image. The linear transformation relation can be expressed as a homography matrix. Using the homography matrix, it is possible to find out which coordinates of the transmitted point correspond to the point of interest in the reference image. Thus, the color balance can be performed based on the pixel value of the point of interest and the pixel value of the corresponding point. In order to calculate the linear transformation relation of the image, it is preferable to use the generated minutiae descriptor information.

와 전송된 영상

의 선형 변환 관계를 예시한 것이다.Fig.

And transmitted video

As shown in FIG.

In the linear transformation relation as shown in FIG. 4, the following equation is established.

는 물체 데이터베이스(1210)의 기준 영상,

은 전송받은 물체의 영상,

는 호모그래피 행렬이다. 수학식 1에 따르면, 호모그래피 행렬을 이용해,

의 관심점들을

`의 대응하는 점으로 워핑할 수 있다.here

A reference image of the object database 1210,

The image of the transmitted object,

Is a homography matrix. According to Equation (1), using a homography matrix,

Interests

`Can be warped to the corresponding point of`.

The homography matrix can be expressed in more detail by the following equation.

와

의 대응되는 4개의 특징점 기술자들을 사용하는데, 특징점 기술자들 중에 4개 쌍을 선택할 수 있다. 선택하는 방법은 공지된 다양한 방법 중 어느 것을 사용해도 무방할 것이다. 일례로 RANSAC을 사용한다면, 보다 정확하게 대응되는 특징점 기술자들을 뽑아낼 수 있을 것이다. 대응되는 4개 쌍에 대해 하기의 DLT식을 풀 수 있다.The DLT equation can be used as a method for obtaining the respective components of the homography matrix. The DLT expression

Wow

, Four pairs of minutiae descriptors can be selected. Any of a variety of known methods may be used for selection. For example, if you use RANSAC, you will be able to extract more precise matching feature descriptors. The following DLT equations can be solved for the corresponding four pairs.

는

에서의

번째 특징점의 좌표이며

는 이에 대응되는 좌표이다. 두 번째 행렬의

는 수학식 2의 호모그래피 행렬의 각

번째 행의 전치행렬을 나타낸다.

는

와

간의 대응되는 특징점을 나타낸다.In Equation (3)

The

In

Coordinate of the second minutiae point

Is a coordinate corresponding thereto. Of the second matrix

≪ / RTI > of the homography matrix < RTI ID = 0.0 >

Lt; th > row.

The

Wow

Respectively.

The DLT equation of Equation (3) can be briefly expressed by the following equation.

는

와

의 대응되는

번째 특징점을 나타낸다,

The

Wow

Corresponding to

Th feature point,

Finally, a homography matrix can be obtained using the following SVD equation.

가

행렬이면

와

는

의 특이벡터들을 나타내는 직교행렬로써 각각

,

행렬이며,

는 특이값을 나타내는

대각 행렬이다. 수학식 5를 이용하여 가장 작은 특이값에 해당하는 특이벡터를 근사해로 선택하면 최종적으로 호모그래피 행렬

를 구할 수 있다. The SVD equation in equation (5) is used to solve the linear equation through singular value decomposition.

end

If the matrix is

Wow

The

≪ / RTI > are orthogonal matrices representing the singular vectors of

,

Matrix,

Indicates a specific value

Diagonal matrix. If a singular vector corresponding to the smallest singular value is selected by approximate solution using Equation (5), a homography matrix

Can be obtained.

When the coordinates of the point of interest of the object database 1210 are warped using the obtained homography matrix, coordinates corresponding to the points of interest in the transmitted image can be calculated.

에 의해 워핑된 관심점을 예시한 것이다. FIG. 5 is a diagram illustrating a homography matrix

Lt; RTI ID = 0.0 > of < / RTI >

의 관심점의 정보를 호모그래피 행렬

를 이용하여 워핑시켜 기준 영상

의 관심점과 대응되는 전송받은 영상

의 관심점의 좌표를 산출할 수 있다.As shown in FIG. 5, the object recognizing unit 1230 recognizes a reference image registered in the object database 1210

The information of the interest points of the homography matrix

To warp the reference image

Corresponding to the point of interest

Can be calculated.

The object recognizing unit 1230 provides the gain calculating unit 1250 with the coordinates of the point of interest in the calculated transmitted image.

In order to calculate a gain value for color balance, the gain calculating unit 1250, which will be described later, needs an ideal pixel value of the interest point in addition to the coordinates of the interest point provided from the object recognizing unit 1230. [ Knowing the coordinates of the point of interest in the transmitted image and the ideal pixel value of the point of interest, the gain value is calculated by calculating the ratio of the actual pixel value to the ideal pixel value of the corresponding coordinates and the color balance .

Therefore, the parameter extracting unit 1240 extracts information for calculating an ideal pixel value of the point of interest from the object database 1210. [

The parameter extracting unit 1240 receives the ID of the recognized object from the object recognizing unit 1230 and extracts the intrinsic reflection spectrum of the object from the object database 1210. The sensitivity information of the camera and the power spectral information of the illumination are also extracted something to do.

When capturing an image, the ID information of the camera and the information of the illumination may be stored and transmitted to the server together with the image. Therefore, the parameter extracting unit 1240 can extract the sensitivity information according to the ID of the camera and the power spectrum information of the corresponding illumination from the object database 1210. If the camera ID and illumination information of the image are not transmitted together with the image, the camera sensitivity information and the power spectrum information of the illumination may be used as default values in the server.

The parameter extracting unit 1240 provides the information extracted from the object database 1210 to the gain calculating unit 1250.

The gain calculating unit 1250 calculates gain using the coordinate information of the point of interest provided from the object recognizing unit 1230, the intrinsic reflection spectrum information provided from the parameter extracting unit 1240, the sensitivity information of the camera, .

First, the gain calculator 1250 calculates an ideal pixel value of a point of interest using the intrinsic reflection spectrum information of the object, the sensitivity information of the camera, and the power spectrum information of the illumination.

The intrinsic reflection spectrum information of the object, the sensitivity information of the camera, and the power spectrum information of the illumination can be used to determine the pixel value, which can be expressed by the following equation.

는 픽셀값,

은 물체의 고유 반사 스펙트럼,

는 카메라의 민감도,

는 조명의 파워 스펙트럼이다. In Equation (6)

A pixel value,

Is the intrinsic reflection spectrum of the object,

The sensitivity of the camera,

Is the power spectrum of the illumination.

Thus, by integrating the intrinsic reflection spectrum information, the sensitivity information of the camera, and the power spectrum information of the illumination, an ideal pixel value of a point of interest can be calculated.

Next, the gain calculator 1250 compares the ideal pixel value of the obtained interest point with the actual pixel value of the coordinates of the transmitted interest point, and calculates the ratio as a gain value.

The color balance changes the intensity of the RGB color, and is expressed by the following equation.

,

,

를 곱하여 새로운 RGB색상을 얻는다. 기존에는 비교할 RGB색상이 없기 때문에 화이트 밸런스를 위한 게인값을 구하는 다양한 방법들이 제안되어 왔다. 그러나 본 발명에 따르면, 정확히 비교할 관심점의 픽셀값이 존재하므로, 하기의 수학식에 따라 쉽고 정확하게 게인값을 구할 수 있다.As shown in Equation (7), the color balance method uses a gain value

,

,

To obtain a new RGB color. Since there is no RGB color to be compared in the past, various methods for obtaining a gain value for white balance have been proposed. However, according to the present invention, since the pixel value of the point of interest to be accurately compared exists, the gain value can be obtained easily and accurately according to the following expression.

,

,

은 전송받은 영상의 픽셀값이고,

,

,

은 계산된 이상적인 픽셀값이다. In Equation (8)

,

,

Is the pixel value of the transmitted image,

,

,

Is the calculated ideal pixel value.

For accurate color balance of the transmitted image, a plurality of points of interest may be used. In order to obtain the gain value that minimizes the error with respect to the colors of the plurality of points of interest, the pixel value of each point of interest may be expressed by a matrix of the following equation, and then the gain value may be obtained by solving the equation by the least squares method.

,

,

는

번째 관심점의 픽셀값이다. 수학식 9를 통해

개의 관심점에 대해 가장 적은 오차를 갖는 게인값

,

,

을 구할 수 있게 된다. In Equation (9)

,

,

The

Th pixel of interest. Through equation (9)

Gain values with the smallest error for the points of interest

,

,

.

The gain calculator 1250 can calculate a gain value for color balance of the transmitted image in this way. The server communication unit 1220 transmits the calculated gain value to the image correction apparatus 1100. [

The image correction apparatus 1100 receives the gain value calculated through the communication unit 1120. [ The image correction unit 1130 can perform color balance on the entire image acquired by the image acquisition unit 1110 based on the received gain value.

FIG. 2 is a flowchart showing the steps of an automatic image color balance method according to an embodiment of the present invention, according to the flow of time.

The automatic image color balancing method according to an exemplary embodiment of the present invention includes storing S2100 of objects, receiving an image S2200, recognizing an object in an image S2300, (S2400), and calculating and transmitting a gain value (S2500).

Storing information of objects (S2100) is a step of storing information for object recognition, intrinsic reflection spectrum information of an object, sensitivity information of a camera, and power spectrum information of illumination in an object database 1210. The information for object recognition may include a reference image of various objects, a minutiae descriptor, and interest information.

The step of receiving the image (S2200) is a step of receiving the image from the server communication unit 1220. [ And may also receive the camera ID and illumination information together with the image photographed from the image correction device 1100. [

In the step S2300 of recognizing an object in the image, the object recognition unit 1230 recognizes an object in the received image using information for object recognition in the object database 1210. [ Recognizing the object using the minutiae descriptor, and using the point of interest information to determine the corresponding point of interest coordinates in the received image.

The step of extracting the eigenfrequency spectrum information (S2400) is a step of extracting the intrinsic reflection spectrum information of the object recognized by the parameter extraction unit 1240 in the object database 1210. Sensitivity information of camera and power spectral information of illumination can also be extracted.

In operation S2500, the gain calculator 1250 calculates a gain value and transmits the calculated gain value to the image correcting apparatus 1100 through the server communication unit 1220 so as to perform color balance . An ideal pixel value of the point of interest is calculated using the values extracted in step S2400 of extracting the eigenfrequency spectral information, and the corresponding interest point coordinates in the received image obtained in step S2300 of recognizing an object in the image And the gain value is calculated.

As described above, the automatic image color balance system according to an embodiment of the present invention uses information of various widely used objects, so that even when there is no white subject, the color balance can be performed by recognizing an object in the image.

In addition, since the eigenfrequency spectrum, which is an unchanging characteristic of an object, is used, there is an advantage of performing accurate color balance.

Embodiments of the present invention may be implemented in the form of program instructions that can be executed on various computer means and recorded on a computer readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and configured for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Such as magneto-optical, and ROM, RAM, flash memory, etc. Examples of program instructions recorded on the medium include machine code such as those produced by a compiler, Lt; RTI ID = 0.0 > and / or < / RTI > The hardware devices described above may be configured to operate as at least one software module to perform operations of one embodiment of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- Those skilled in the art will appreciate that various modifications and changes may be made thereto without departing from the scope of the present invention. Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

1100: Image correction device
1110:
1120:
1130: image correction unit
1200: balance data server
1210: Object database
1220: Server communication section
1230: Object recognition unit
1240:
1250: gain calculator

Claims (10)

An object database in which recognition information of objects and intrinsic reflection spectrum information are registered;
An image acquisition unit receiving an image;
An object recognition unit for recognizing an object in the input image using recognition information of the object database;
A parameter extraction unit for extracting intrinsic reflection spectrum information of the recognized object from the object database;
A gain calculator for calculating a gain value for color correction of the input image using the extracted eigenfrequency spectral information; And
And an image correction unit that corrects the input image based on the gain value, wherein the recognition information of the object database includes a reference image, a minutiae point descriptor, and coordinates of at least one point of interest, and the object recognition unit A minutiae descriptor is generated on an image and compared with the minutiae descriptors of the object database to recognize an object having the most similar minutiae descriptor as an object in the input image, Wherein the coordinates of a point of interest in the input image is obtained by obtaining a linear transformation relation between the reference image of the object and the input image.
The method according to claim 1,
Wherein the object database further includes sensitivity information of the camera and power spectrum information of the illumination.
The method according to claim 1,
Wherein the object recognition unit obtains a matrix representing a linear transformation relationship between an object in the input image and a reference image of an object recognized as the same object and then uses the matrix to warp coordinates of a point of interest in the reference image of the recognized object, And obtaining coordinates of a point of interest in the input image. The method of claim 3,
Wherein the parameter extracting unit further extracts camera sensitivity information of the input image and power spectrum information of illumination from the object database. 5. The method of claim 4,
Wherein the gain calculator calculates an ideal pixel value of a point of interest in the reference image of the object recognized as the same by using the extracted eigenfrequency spectral information, the camera sensitivity information, and the power spectrum information of the illumination, Wherein a gain value for color correction of the input image is calculated by comparing the input pixel value with a pixel value of a point of interest in the input image. 6. The method of claim 5,
Wherein the gain calculator calculates a gain value for each of the points of interest when the point of interest in the reference image of the object recognized as the same is greater than one and then calculates a gain value for color correction of the input image by averaging the values using the least squares method Wherein the color image is a color image.
An object database in which recognition information of objects and intrinsic reflection spectrum information are registered;
A server communication unit for receiving an image;
An object recognition unit for recognizing an object in the received image using recognition information of the object database;
A parameter extraction unit for extracting intrinsic reflection spectrum information of the recognized object from the object database; And
And a gain calculator for calculating a gain value for color correction of the received image using the extracted eigenfrequency spectral information,
The calculated gain value is transmitted through the server communication unit,
The recognition information of the object database includes a reference image, a minutiae descriptor, and coordinates of at least one point of interest. The object recognition unit generates a minutiae descriptor on the input image, compares the minutiae descriptor with minutiae descriptors of the object database, Recognizes that an object having a similar minutiae descriptor is the same as an object in the input image and obtains a linear transformation relationship between a reference image of the object recognized as the same object as the object in the input image and the input image, And obtains coordinates of a point of interest in the image.
14. The method of claim 13,
Wherein the object database further includes sensitivity information of the camera and power spectrum information of the illumination.
Storing identification information and eigenfrequency spectrum information of objects in an object database;
Receiving an image;
Recognizing an object in the received image using recognition information of the object database;
Extracting intrinsic reflection spectrum information of the recognized object from the object database; And
And calculating and transmitting a gain value for color correction of the received image using the extracted eigenfrequency spectral information,
Wherein the recognition information of the object database includes coordinates of a reference image, a minutiae descriptor, and at least one point of interest, and the step of recognizing the object includes generating minutiae descriptors on the input image, And recognizes that an object having the most similar feature point descriptor is the same as an object in the input image and obtains a linear transformation relationship between the object in the input image and the reference image of the object recognized as the same and the input image, Wherein coordinates of a point of interest in the input image are obtained. A computer-readable recording medium on which a program for performing the automatic image color balance method according to claim 9 is recorded.

Images