KR102124244B1 - Method for producing unique ID of object - Google Patents

Abstract

The present invention relates to a method of generating a unique ID of an object, and more specifically, by generating a unique ID considering characteristics of an object extracted from a plurality of object images, generation of a unique ID that can identify all objects such as animals and objects as well as people. It is possible to compare the input object image and the registered object image by generating a unique ID consisting of numbers generated from the object image, so there is no need to register and store the object images of all objects separately, and the unique ID generated from the input object image And a unique ID of an object capable of identifying all objects by comparing the registered unique IDs.

Description

Method for producing unique ID of object}

The present invention relates to a method of generating a unique ID of an object, and more specifically, by generating a unique ID considering characteristics of an object extracted from a plurality of object images, generation of a unique ID that can identify all objects such as animals and objects as well as people. It is possible to compare the input object image and the registered object image by generating a unique ID consisting of numbers generated from the object image, so there is no need to register and store the object images of all objects separately, and the unique ID generated from the input object image And a unique ID of an object capable of identifying all objects by comparing the registered unique IDs.

In the field of authentication or security, an ID is used to identify a user or object. For example, ID is used in various real life and industries such as access to buildings, banking where security is required, authentication of use of smart devices, and identification of objects.

Such an ID is generally composed of a number of characters and symbols arbitrarily set by a user or an administrator, or it is common to use the user's biometric information itself as a unique ID.

In the case of an ID composed of letters and symbols, the user has a problem of having to use the ID, and the ID is easily exposed to a third party and abused.

On the other hand, when the biometric information itself is used as a unique ID, the biometric information of all users is obtained by obtaining biometric information and comparing the obtained biometric information and pre-stored biometric information itself to authenticate the user with whether the biometric information matches. The information must be stored and registered, and the user's biometric information is extremely personal information, which has a problem that can cause huge damage when exposed to a third party.

In addition, the conventional ID is for identifying a user, and the user generates an ID with symbols and characters for authentication or registration, and does not reflect the user's characteristics. Moreover, the conventional ID has a problem in that it cannot be generated as an ID for identifying all objects including an animal or object by reflecting the characteristics of the animal or object.

The present invention is to solve the problems of the conventional ID generation method mentioned above, and an object of the present invention is to provide a method for generating a unique ID of an object by reflecting the characteristics of the object.

Another object of the present invention is to provide a method of generating a unique ID that can identify all objects such as animals, objects, as well as humans by generating a unique ID in consideration of characteristics of an object extracted from an object image.

Another object to be achieved by the present invention is to extract a feature point representing an object from a plurality of object images, remove a background area, and generate an unique object ID that is unchanged from unique feature points common to a plurality of object images among the extracted feature points. Is to provide a way.

Another object of the present invention is to provide a method for accurately generating a unique ID of an object with a small amount of computation from a descriptor vector for a feature point of an object image generated using a Scale Invariant Feature Transform (SIFT) algorithm.

In order to achieve the object of the present invention, a method for generating a unique ID according to the present invention estimates lighting components from a plurality of different object images photographing the same object and normalizes lighting by a plurality of object images based on the estimated lighting components Generating a generated image, extracting a keypoint for each light-normalized object image using a Scale Invariant Feature Transform (SIFT) algorithm, and generating a descriptor of the extracted feature point, and feature points of each light-normalized object image Extracting the unique feature points of the objects that match each other, removing the feature points of the background region from the light normalized object image for each object image from the distance between the feature points based on the descriptor, and the unique feature points of each light normalized object image And generating a unique ID of the object from the candidate ID of each illumination normalized object image generated from the vector matrix of the descriptor of the Korean.

Preference, according to an embodiment of the present invention, the illumination normalized object image inverts the object image to generate an inverted object image, and applies an average filter to the inverted object image to extract the low-frequency illumination component and estimates the inversion. Calculating an illumination component; reversing the inversion estimated illumination component to calculate the estimated illumination component; adding the estimated illumination component and the object image to each other to generate an illumination normalized inversion object image; and the illumination normalized inversion. Generating an inverted smoothed image in which the pixel values of the illumination normalized inverted object image are diffused and distributed by histogram equalization of the object image, and generating an illumination normalized object image by reversing the inverted smoothed image. It is characterized by.

Preferably, the step of extracting a unique feature point of an object according to an embodiment of the present invention is based on a feature point descriptor for a feature group consisting of feature points of two selected light normalized object images among a plurality of light normalized object images Extracting candidate unique feature points that match each other from the feature point group from the distance between them, and unique feature points that are commonly matched in all feature point groups based on candidate unique feature points extracted from all feature point groups generated from a plurality of lighting normalized object images It characterized in that it comprises a step of extracting.

In the step of extracting the candidate unique feature points, the candidate unique feature points are extracted by the following equation (1).

[Equation 1]

Here, v _1n is a feature point of the first illumination normalized object image, v _2i , v _2j are two feature points of the second illumination normalized object image closest to v _1n , and d _R is an extraction weight , When satisfying Equation (1) v _1n , v _2i is characterized by being extracted as a candidate unique feature point.

Preferably, the step of generating a unique ID according to an embodiment of the present invention includes the largest number and the largest number in the 8-dimensional principal component of each partial vector matrix generated from the descriptor vector matrix for the unique feature point of each illumination normalized object image. Determining a small number of positions, and generating candidate IDs for each light normalized object image, consisting of first and second values representing the largest and smallest positions of each partial vector matrix. And, it characterized in that it comprises the step of generating a unique ID of the object consisting of a number common to each other from the candidate ID for each normalized object image.

Preferably, the step of extracting a feature point according to an embodiment of the present invention includes generating a scaled spatial image blurred by a Gaussian function having a different level of variance, and a difference between adjacent images among the scaled spatial images. Generating a Gaussian difference image, and determining a maximum or minimum pixel value from a neighboring pixel adjacent to the current pixel among the Gaussian difference images, and a neighboring adjacent pixel adjacent to the current pixel among upper and lower adjacent neighboring Geuntian difference images among the Gaussian difference images The branch is characterized by including the step of determining a candidate feature point, and removing the unstable candidate feature point from the candidate feature point to extract the feature point.

Calculating the azimuth and size between the feature point and the adjacent pixel of the feature point using adjacent pixels of the feature point around the feature point, determining a reference orientation of the feature point from the calculated histogram of the feature point, and surrounding the feature point based on the reference orientation And generating a feature point descriptor expressed in a gradient direction and size of the local area located at.

Here, it is characterized in that noise is removed from three different object images photographing the same object to obtain a normalized object image for each of the three object images.

The method for generating a unique ID of an object according to the present invention has various effects as follows.

First, in the method for generating a unique ID of an object according to the present invention, by generating a unique ID considering characteristics of an object extracted from an object image, it is possible to generate a unique ID that can identify all objects such as animals and objects.

Second, the method for generating a unique ID of an object according to the present invention extracts a feature point representing an object from a plurality of different object images and generates a unique ID of the object from unique feature points common among the plurality of object images among the feature points, The unique ID of the object can be generated accurately by reflecting the characteristics.

Third, the method for generating a unique ID of an object according to the present invention does not compare the input object image and the registered object image by generating a unique ID consisting of numbers generated from the object image, so it is necessary to separately register and store the object images of all objects. There is no, and all objects can be identified by comparing the unique ID generated from the input object image with the registered unique ID.

Fourth, in the method of generating a unique ID of an object according to the present invention, a unique ID is generated using a descriptor vector for a feature point of an object image generated by applying a Scale Invariant Feature Transform (SIFT) algorithm, so that the unique ID of the object is accurately calculated with a small amount of computation. You can create an ID.

1 is a functional block diagram illustrating a device for generating a unique ID according to an embodiment of the present invention.
2 is a view for explaining an example of a normalized image generation unit according to the present invention.
3 is a functional block diagram illustrating an example of a unique feature point generating unit according to the present invention.
4 is a flowchart illustrating a method for generating a unique ID of an object according to an embodiment of the present invention.
5 is a view for explaining an example of generating a gown cyan difference image.
FIG. 6 is a diagram for explaining an example of determining a feature point having a maximum or minimum pixel value in a next neighboring pixel adjacent to a current pixel in a difference image.
7(a) is a diagram showing a gradient direction and a size of a local area located around a SIFT feature point, and FIG. 7(b) is a diagram showing a gradient direction and a size after applying a Gaussian weight.
8 is a flowchart illustrating an example of a step of extracting unique feature points according to the present invention.
9 shows an example of unique feature points extracted from the first to third illumination normalized object images.
10 is a flowchart for explaining an example of the step of generating a unique ID according to the present invention.
11 illustrates an example of a feature point descriptor vector matrix for N eigen feature points in an illumination normalized object image.
12 shows an example of first and second values for each normalized object image.

It should be noted that the technical terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. In addition, technical terms used in the present invention should be interpreted as meanings generally understood by a person having ordinary knowledge in the technical field to which the present invention belongs, unless otherwise defined in the present invention. It should not be interpreted as a meaning or an excessively reduced meaning. In addition, when the technical term used in the present invention is a wrong technical term that does not accurately represent the spirit of the present invention, it should be understood as being replaced by a technical term that can be correctly understood by those skilled in the art.

In addition, the singular expression used in the present invention includes a plural expression unless the context clearly indicates otherwise. In the present invention, terms such as “consisting of” or “comprising” should not be construed to include all of the various components, or various stages, described in the invention, including some of the components or some stages It may or may not be construed as further comprising additional components or steps.

In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention and should not be interpreted as limiting the spirit of the present invention by the accompanying drawings.

Hereinafter, a method of generating a unique ID according to the present invention will be described in more detail with reference to the accompanying drawings.

1 is a functional block diagram illustrating a device for generating a unique ID according to an embodiment of the present invention.

Referring to FIG. 1, the image photographing unit 110 photographs an image of an object. Here, the object photographed through the image photographing unit 110 is an object having different characteristics for each object, for example, a human face, an iris, a vein pattern, etc., can be an object, and in the field to which the present invention is applied Accordingly, the face, vein pattern, iris, etc. of a pet, such as a puppy or a cat, may be an object, or objects having different characteristics from each other may be objects. The image photographing unit 110 acquires a plurality of object images by photographing objects with a time difference from each other. Preferably, the image photographing unit 110 acquires three object images by photographing the object three times with a time difference from each other.

As an example of the image photographing unit 110, a near infrared ray irradiated from an illumination unit configured such that a plurality of near infrared LED lamps are arranged in a plurality of columns and rows transmits or reflects an object, and photographs the object image using the provided light. Charge coupled device (CMOS) or complementary metal-oxide semiconductor (CMOS) cameras can be used.

The normalized image generating unit 130 estimates a lighting component from each of the plurality of object images obtained from the image photographing unit 110 and generates a lighting normalized image for each of the plurality of object images based on the estimated lighting component. The image capturing unit 110 acquires an object image using an infrared light source, and the infrared light source may be focused and irradiated to a specific part of the object according to the arrangement position of the infrared light source, and the object image obtained due to the difference in object tissue or thickness The lighting component at is unbalanced.

The feature point generator 150 extracts a SIFT keypoint representing an object feature from each light normalized image using a Scale Invariant Feature Transform (SIFT) algorithm, and generates a descriptor of the extracted SIFT feature point. The descriptor of the extracted feature points is the gradient direction and size of the local area located around the SIFT feature points, and the location information of the SIFT feature points.

The unique feature point generation unit 170 determines characteristic points that match each other in each light normalized image from the distance between the feature points based on the feature point descriptors of each light normalized image to generate unique feature points common to each other in the light normalized image . Here, the unique feature points are common feature points that match each other in each normalized image, and by removing feature points that do not match each other among the feature points of each normalized image, only the object region excluding the background region can be extracted from the normalized image. .

The unique ID generator 190 generates candidate IDs from a vector matrix of descriptors for unique feature points for each light normalized image, and generates unique IDs of objects from candidate IDs common to each other in the candidate IDs of each light normalized image. .

2 is a view for explaining an example of a normalized image generation unit according to the present invention.

Referring to FIG. 2 in more detail, the first inverting unit 131 generates an inverted image I'(x,y) by inverting the object image I(x,y), and the lighting component estimator (133) calculates the inverted estimated illumination component (R(x,y)) by convolving the average filter to extract the low-frequency illumination component. Since the illumination component appears as a low-frequency component in the object image, the illumination component is estimated from the inverted original image by convolution of the average filter on the inverted original image and blurring the inverted original image. Here, the average filter has an m×n size and is convolved with an M×N sized inverted original image to generate an inverted estimated illumination component.

The second inversion unit 135 re-inverts the inverted estimated illumination component R(x,y) to calculate the estimated illumination component R'(x,y).

The illumination normalization unit 137 adds the estimated illumination component R'(x,y) and the inverted image I'(x,y) to each other to maintain a pixel value between 0 and 255, and a pixel value of 0 or less. Set to 0, and a pixel value of 255 or greater is set to 255 to generate an illumination normalized inverted image I''(x,y), and the smoothing unit 158 illuminates the normalized illumination image I''(x, y)) is histogram equalization to diffusely distribute the pixel values of the illumination normalized inverted image I''(x,y). Here, the histogram of the light-normalized inverted image with poor contrast distribution is uniformly distributed through histogram smoothing, and a light-normalized object image with maximized contrast is generated.

The third inversion unit 139 re-inverts the inverted smoothing image to generate an illumination normalized object image E(x,y).

When the pixel value of the image has a value of 0 to 255, the conventional lighting normalization method subtracts the estimated lighting component generated by applying an average filter to the input image from the input signal, and adds 255 to the subtracted value. The lighting was normalized by dividing the value by 1/2. The normalized image generator according to the present invention sums the estimated illumination component R'(x,y) and the inverted image I'(x,y) to maintain the pixel values between 0 and 255 and maintains pixel values below 0 By setting to 0 and setting the pixel value over 255 to 255, the object image can be more clearly filtered and the accuracy of SIFT feature point extraction and matching can be improved.

3 is a functional block diagram illustrating an example of a unique feature point generating unit according to the present invention.

The group generating unit 171 generates a group from two different normalized object images selected from a plurality of normalized object images. That is, when three light normalized object images E1, E2, and E3 are generated from three object images, the first group E1, E2 from the two light normalized object images selected from the three light normalized object images ), the second group (E1, E3), and the third group (E2, E3) are created.

The candidate unique feature point generation unit 173 generates candidate unique feature points that match each other in each group based on the distance between the feature points of the lighting normalized object image constituting each group, and the unique feature point generation unit 175 generates each group In each of the candidate unique feature points matching each other, unique feature points common to each group are generated.

4 is a flowchart illustrating a method for generating a unique ID of an object according to an embodiment of the present invention.

Looking more specifically with reference to FIG. 4, lighting components are estimated from a plurality of obtained object images for the same object, and normalized lighting is generated for each of the plurality of object images based on the estimated lighting components (S10 ).

A keypoint of an object is extracted from each light normalized image using a SIFT algorithm, and a description of the feature point of each light normalized image is generated (S50).

Looking at the step of extracting the feature points in more detail, the blurring scale spatial image is obtained by convolution of a normalized finger vein image with a Gaussian function having a different level of variance and a peaced light normalized finger vein image. Produces like this:

[Equation 1]

L(x, y, σ)=G(x, y, σ)*E(x,y)

Where G(x, y, σ) is the gross cyan function, E(x,y) is the smoothed light normalized object image, and L(x, y, σ) is the Gaussian function with different levels of variance (σ). It means a blurred image.

In order to find a stable feature point in the scale space, a Gaussian difference image (DOG) is generated from the difference between adjacent images among the scale space images as shown in Equation (2) below. 5 is a view for explaining an example of generating a gown cyan difference image.

[Equation 2]

DOG=(G(x, y, kσ)*E(x,y)-G(x, y, σ)*E(x,y))

=L(x, y, kσ)-L(x, y, σ)

A feature point having a maximum or minimum pixel value is determined from a neighboring pixel adjacent to the current pixel among the Gaussian difference images, and a neighboring adjacent pixel adjacent to the current pixel among upper and lower adjacent neighboring Gownian difference images.

Referring to FIG. 6 in more detail, a gaussian difference image is used to compare 8 pixels in the current image and 9 pixels in the adjacent image. Assuming that the'X' position is the current reference pixel, the total including pixels included in a certain area (for example, 3x3) of the same position in the Gaussian difference image adjacent to the neighboring pixel around the'X' By comparing the pixel values of 26 pixels, if the reference pixel is the maximum or minimum value, the reference pixel is registered as a feature point. The above process is repeated using the Gaussian difference image for each octave in the scale space to find feature points.

Here, the function m(x,y) for the size value of the SIFT feature point (X(x,y)) and the function θ(x,y) for the direction value are shown in Equations (3) and (4) below. Is calculated by

[Equation 3]

[Equation 4]

The SIFT feature points whose size (m(x,y)) and direction (θ(x,y) are determined based on the surrounding pixel values are not changed, because the pixel values around the SIFT feature points (X') do not change even if the image is rotated. The direction of the SIFT feature points is rotated as the image rotates, so the reference direction of the SIFT feature points becomes an important criterion when creating feature descriptors that do not change even with the rotation of the image.

On the other hand, looking at the method of generating the feature point descriptor in more detail, a key point descriptor is generated based on the reference direction of the feature point.

Calculate the gradient direction and size of the local area located around the SIFT feature point. The gradient direction and size of the local area can be calculated through Equation (3) and Equation (4) described above.

To obtain an invariant descriptor for image rotation, the gradient direction and the coordinates of the descriptor are rotated based on the SIFT feature point direction, and a Gaussian weight is given to the rotated gradient around the SIFT feature point. Gaussian weighting is applied to prevent the gradient direction and size from being sensitively changed to small changes in the area around the SIFT feature point. By applying a Gaussian weight around the SIFT feature points, it is possible to emphasize the gradient around the feature points and minimize errors.

7(a) is a diagram showing a gradient direction and a size of a local area located around a SIFT feature point, and FIG. 7(b) is a diagram showing a gradient direction and a size after applying a Gaussian weight.

The Gaussian weighted gradient is reconstructed into a 4x4 array with each array having 8 directions. This directional histogram consists of 128 (8x4x4)-dimensional vectors. A normalization process can be performed to make the 128-dimensional vector a light-resistant descriptor. The 128-dimensional vector generated through this process becomes a feature point descriptor representing a SIFT feature point.

Referring to FIG. 4 again, referring to the method for generating a unique ID of an object according to the present invention, a background area in a light normalized object image for each object image from a distance between feature points based on a feature point descriptor of each light normalized object image The unique feature points of the objects that match each other are extracted by removing the feature points (S70 ).

A candidate unique ID for each light normalized object image is generated from a 128 vector matrix of feature point descriptors for the unique feature points of each light normalized object image, and from the candidate unique IDs for each light normalized object image to common IDs Generate a unique ID for the object (S90).

8 is a flowchart illustrating an example of a step of extracting unique feature points according to the present invention.

Referring to FIG. 8 in more detail, a group of feature points is generated as a feature point of two selected light normalized images among a plurality of light normalized images (S71). That is, when three lighting normalized object images E1, E2, and E3 are generated from three object images, the first group E1 from two different lighting normalized object images selected from the three lighting normalized object images , E2), the second group (E1, E3), and the third group (E2, E3).

Based on the distance between the feature points of the illumination normalized object image constituting each feature point group, candidate unique feature points matching each other are extracted (S73). For example, based on the distance between the feature points of the first illumination normalized object image E1 and the second illumination normalized object image E2 constituting the first group, and the feature points of the first illumination normalized object image Candidate unique feature points common to each other are extracted between the second illumination normalized object images.

Preferably, the candidate unique feature points that match each other in each group are calculated as in Equation (5) below.

[Equation 5]

Here, v _1n is a feature point of the first illumination normalized object image, v _2i , v _2j are two feature points of the second illumination normalized object image closest to v _1n , and d _R is an extraction weight ,

When Equation (5) is satisfied, v _1n , v _2i is extracted as a candidate unique feature point.

The candidate unique feature points of each feature point group are extracted in the same manner as described above for all the generated feature point groups.

Among the candidate unique feature points that match each other in each feature point group, unique feature points that are commonly matched with each other are extracted (S75). That is, as illustrated in FIG. 9, when three light normalized images are obtained from three object images, feature points of the first light normalized image E1 and feature points of the third light normalized image E2 A unique feature point common to all of the feature points of the field and the third illumination normalized image E3 is extracted.

10 is a flowchart for explaining an example of the step of generating a unique ID according to the present invention.

Looking more specifically with reference to FIG. 10, the positions of the largest number and the smallest number in the 8-dimensional principal component of the partial vector matrix generated from the vector matrix of the feature point descriptor for the unique feature points of each normalized object image are determined ( S91). As illustrated in FIG. 11, when N unique feature points (V _{1 -1} , V _1-2 ,...., V _{1 -N} ) are extracted from the first illumination normalized object image, each unique feature point In the above-described manner, a 128-dimensional feature point descriptor vector matrix is generated. The 128-dimensional feature point descriptor vector matrix is divided into 8 to generate a partial vector matrix (P _i ), and the average value of the magnitudes of principal components is calculated for each partial vector matrix. In the case, the position of the largest number and the smallest number in each average value is determined.

)과의 행렬곱(

)으로 얻어진 행렬의 고유값과 고유벡터를 계산하여 그 고유벡터들로부터 부분 벡터 행렬(P_i)의 열벡터들의 주성분을 계산하는 방법이다. Divide the 128-dimensional feature point descriptor vector matrix into 8 to generate a partial vector matrix (P _i ), and calculate the 8-dimensional principal component vector from each partial vector matrix to determine the location of the largest and smallest numbers in each principal component vector. do. Here, the principal component means a vector consisting of the average vector of the partial vector matrix P _i or the absolute value of the eigenvector corresponding to the largest number of eigenvalues obtained through Principle Component Analysis (PCA). Principal component analysis is based on the matrix P _i ,

Matrix product of)

) Is a method of calculating the eigenvalues and eigenvectors of a matrix obtained by) and calculating the principal components of column vectors of the partial vector matrix (P _i ) from the eigenvectors.

In FIG. 11, the average value of the principal component size of the largest size of the partial vector matrix P ₁ is 4.0, and the average value of the smallest principal component size is 2.6. The average value 4.0 is located in the eighth, and the average value 2.6 is located in the third position. The average values of the principal component sizes for each partial vector matrix generated from the feature point descriptor vector matrix of common feature points common to each other in the plurality of illumination normalized images may be different from each other, in which order the largest component is located and where the smallest component is located. It can be seen that the order of doing is almost the same.

Referring back to FIG. 10, candidate IDs for each illumination normalized object image, which are composed of first and second values indicating positions of the largest number and the smallest number of each partial vector matrix, are generated (S93 ). . As described above with reference to FIG. 11, in the first partial vector matrix P ₁ , the first value and the second value are 8 and 3, respectively, and the first and second values of the remaining partial vector matrix are determined to determine all parts. A candidate ID consisting of a first value and a second value for a vector matrix is generated. FIG. 12 shows an example of first and second values for each normalized object image, candidate ID1 consisting of first and second values of all partial vector matrices of the first normalized image. (ID ₁ ), a candidate ID (ID ₂ ) consisting of the first and second values of all partial vector matrices of the second illumination normalized image, the first and second values of all partial vector matrices of the third illumination normalized image The candidate ID (ID ₃ ) consisting of two values is as follows.

ID ₁ = (83, 71, 57, 34, 36, 62, 35, 41, 34, 12, 47, 65, 47, 21, 65, 27)

ID ₂ = (83, 71, 57, 35, 36, 52, 35, 41, 34, 12, 47, 65, 47, 23, 65, 27)

ID ₃ = (83, 71, 57, 35, 36, 52, 35, 41, 34, 12, 47, 65, 47, 24, 65, 27)

A unique ID of an object having a common number is generated from candidate IDs for each normalized object image (95). That is, when the candidate ID1 (ID ₁ ) of the first normalized light image, the candidate ID (ID ₂ ) of the _second normalized light image, and the candidate ID (ID ₃ ) of the third normalized light image match each other If the number of matches remains the same, and if they do not match, set to 0 to set the candidate ID 1 (ID ₁ ) of the first normalized image, the candidate ID (ID ₂ ) of the second normalized image, and normalize the third illumination A unique ID of the object is generated from the candidate ID (ID ₃ ) of the image.

According to an embodiment, if any of the candidate IDs generated from the plurality of normalized object images does not match each other, it may be set to 0. According to another embodiment, when all candidate IDs generated from the plurality of normalized object images are different from each other, a unique ID may be set to a matching number if a matching number exists.

In FIG. 12, one embodiment generated from the candidate ID1 (ID ₁ ) of the first normalized image, the candidate ID (ID ₂ ) of the _second normalized image, and the candidate ID (ID ₃ ) of the third normalized image The unique ID (I) according to the example is as follows.

ID=(83, 71, 57, 30, 36, 02, 35, 41, 34, 12, 47, 65, 47, 20, 65, 27)

Meanwhile, in FIG. 12, generated from candidate ID1 (ID ₁ ) of the first normalized light image, candidate ID (ID ₂ ) of the _second normalized light image, and candidate ID (ID ₃ ) of the third normalized light image The unique ID (I) according to another embodiment is as follows.

ID=(83, 71, 57, 35, 36, 52, 35, 41, 34, 12, 47, 65, 47, 20, 65, 27)

Meanwhile, the above-described embodiments of the present invention can be written as a program executable on a computer, and can be implemented on a general-purpose digital computer that operates the program using a computer-readable recording medium.

The computer-readable recording medium includes magnetic storage media (eg, ROM, floppy disks, hard disks, etc.), optical reading media (eg, CD-ROMs, DVDs, etc.) and carrier waves (eg, the Internet). Storage media).

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those skilled in the art will understand that various modifications and other equivalent embodiments are possible therefrom. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

110: image capture unit 130: normalized image generation unit
150: feature point generator 170: unique feature point generator
190: unique ID generation unit 131: first inversion unit
133: illumination component extraction unit 135: second inversion unit
137: lighting normalization unit 138: smoothing unit
139: third inversion unit 171: group creation unit
173: candidate unique feature point generator 175: unique feature point generator

Claims (8)

Estimating a lighting component from a plurality of different object images photographing the same object and generating an illumination normalized image for each of the plurality of object images based on the estimated lighting component;
Extracting a keypoint representing an object feature for each illumination normalized object image using a Scale Invariant Feature Transform (SIFT) algorithm, and generating a descriptor of the extracted feature point;
A group is generated by selecting two light normalized object images from among the plurality of light normalized object images, and matching each other from the distance between the feature points based on the feature point descriptors of the two light normalized object images constituting each group, Determining a unique feature point for each group;
Determining unique feature points common to all groups generated from the plurality of lighting normalized object images among candidate unique feature points for each group; And
Generating candidate IDs of each normalized object image from a vector matrix of descriptors for unique feature points of each normalized object image, and generating unique IDs consisting of a common number of candidate IDs of each normalized object image. Unique ID generation method of the object, characterized in that it comprises. The method of claim 1, wherein the normalized lighting object image
Generating an inverted object image by inverting the object image;
Calculating an inverted estimated illumination component by applying an average filter to the inverted object image to extract a low-frequency illumination component;
Re-inverting the inverted estimated illumination component to calculate an estimated illumination component;
Generating an illumination normalized inverted object image by summing the estimated illumination component and the object image;
Generating an inverted smoothed image in which the pixel values of the normalized inverted object image are diffusely distributed by performing histogram equalization of the normalized inverted object image; And
And re-inverting the inverted smoothing image to generate an illumination normalized object image. delete The method of claim 1, wherein in the step of extracting the candidate unique feature points, the candidate unique feature points are extracted by the following equation (1):
[Equation 1]

Here, v _1n is a feature point of the first illumination normalized object image, v _2i , v _2j are two feature points of the second illumination normalized object image closest to v _1n , and d _R is an extraction weight ,
When Equation (1) is satisfied, v _1n , v _2i is a method of generating a unique ID of an object, characterized in that extracted as a candidate unique feature point. The method of claim 1, wherein the step of generating the unique ID
Determining a position of a largest number and a smallest number in an 8-dimensional principal component of each partial vector matrix generated from a descriptor vector matrix for a unique feature point of each normalized object image;
Generating candidate IDs for each illumination normalized object image, consisting of first and second values indicating positions of the largest number and the smallest number of each partial vector matrix; And
And generating unique IDs of objects having a common number from each other in the candidate IDs for the normalized object images. The method of claim 1, 2, 4 and 5, wherein the step of extracting the feature point is
Generating a blurred scale spatial image by a Gaussian function having different levels of variance;
Generating a Gaussian difference image from a difference of adjacent images among the scale spatial images;
Determining a candidate feature point having a maximum or minimum pixel value from adjacent pixels adjacent to the current pixel among the Gaussian difference images, and adjacent adjacent pixels above and below the Gaussian difference image adjacent to the current pixels; And
And extracting a feature point by removing an unstable candidate feature point from the candidate feature point. The method according to any one of claims 1, 2, 4 and 5,
Calculating an azimuth and a size between the feature point and the adjacent pixels of the feature point by using adjacent pixels of the feature point around the feature point;
Determining a reference orientation of the feature point from the calculated azimuth histogram; And
And generating a feature point descriptor expressed in a gradient direction and a size of the local area located around the feature point based on the reference orientation. The method according to any one of claims 1, 2, 4 and 5,
A method for generating a unique ID of an object, wherein noise is normalized for each of the three object images to obtain a normalized object image by removing noise from three different object images taken of the same object.

Images