KR101715649B1 - Image recognition system using transformed images - Google Patents

Abstract

An image recognition system using a transformed image and a method for providing the same are disclosed. The method for providing an image recognition system using the transformed image corresponds to each of the original feature points of the original feature points corresponding to the original image corresponding to the recognition object and a plurality of transformation images corresponding to the original image, Specifying transformed minutiae points corresponding to the original minutiae points; And the image recognition system generating at least one integrated feature point from the original feature point and the transformed feature points, wherein the transform feature points are specified based on the original feature point.

Description

[0001] Image recognition system using transformed images [

The present invention relates to an image recognition system and a method for providing the image recognition system, and more particularly, to a system and method for providing an image recognition system, And a system capable of reducing the size of a DB while increasing a recognition rate by using the system and a method for providing the same.

Image recognition algorithms and systems therefor are well known. In general, the image recognition algorithm determines whether input data, that is, input image, exists in reference data existing in a specific DB.

For example, an image recognition algorithm such as a Scale Invariant Feature Transform (SIFT) algorithm is an algorithm that recognizes or searches for a specific feature included in a reference image, and is a well-known algorithm in the field of computer vision.

Although the technical idea of the present invention is not necessarily applied to the SIFT, the SIFT algorithm will be described herein for convenience of explanation.

The SIFT algorithm was proposed by David Lowe in 1999 and is described in the paper "Lowe, David G. (1999)." Object recognition from local scale-invariant features. "Proceedings of the International Conference on Computer Vision. International Journal of Computer Vision, 60, 2, pp. 91-110, 2004 ", and "Lowe, DG, " Distinctive Image Features from Scale-Invariant Keypoints ", pp. 1150-1157. Doi: 10.1109 / ICCV.1999.790410. There is. The above articles may be incorporated by reference herein.

As is known in the literature, the SIFT algorithm is an algorithm that can recognize certain features regardless of scale and orientation. The SIFT algorithm is an algorithm of recognizing the recognition target by matching keypoints of recognition targets with reference data, that is, minutiae points of a reference image.

At this time, when the recognition object is digitized data such as a predetermined image, a group of unique points is extracted for each recognition object.

However, when an object to be recognized is an object existing in a natural state, that is, a natural object, various transformation features may exist even for the same object to be recognized.

FIG. 1 is a view for explaining the difficulty of image recognition when the recognition object is a natural object. Referring to FIG. 1, when the predetermined recognition object is a natural object (e.g., a mark) The image may be represented by a plurality of images according to the state or angle of the means for acquiring the image of the recognition target. For example, as shown in FIG. 1, an image representing the same recognition target may be obtained not only from any one of the images 10 but also from various images (for example, 10-1, 10-2 , 10-3, 10-4).

That is, even when any one of the images 10 corresponding to a specific recognition object exists in the predetermined DB as the reference data (for example, the image 10 or the minutiae points of the image 10) The recognition rate of the recognition object, which is a natural object, can not but be reduced. This is because the image representing the recognition object which is a natural object does not exist only in the image 10 as shown in FIG. 1, but various images (for example, 10-1, 10-2, 10-3, 10-4), and any one of these transformed images (e.g., 10-1, 10-2, 10-3, 10-4) Because it can be input as an input image.

Therefore, in order to prevent such a problem, it may be necessary to preliminarily input various images (or minutiae points of an image) representing the recognition target in advance as reference data to a specific recognition target.

FIG. 2 is a diagram for explaining a problem when each of a plurality of images corresponding to a specific recognition target is included in reference data. As shown in FIG. 2, the images corresponding to a specific recognition object , 10, 11, 12, etc.) can be many. The feature points corresponding to the plurality of images (for example, 10, 11, 12) may be different from each other (i.e., the feature points extracted by the predetermined feature extraction algorithm) 11, and 12 are simply stored in the DB storing the reference data, the amount of data stored in the DB becomes too large, which may lead to a reduction in the speed and resources of performing the image recognition.

The technical idea for solving such a problem is disclosed in Korean Registered Patent (Registration No. 10-1265694, "Image recognition system using DB efficiency and method for providing the same ", hereinafter referred to as " registered patent ") filed and registered by the present applicant There is a bar. In this specification, the registered patent may be included as a reference.

In the patent, a plurality of various images are collected, and the matching points are searched (i.e., matching is determined) among the minutiae extracted from the respective images, and the matched points are combined and stored in the DB, And a higher recognition rate can be obtained.

However, it may not be easy to collect a plurality of various images for a specific recognition target, and even if a plurality of various images are collected, a considerable amount of time and resources .

Therefore, a technical idea that can improve the recognition rate by directly generating a plurality of various images using any specific image of the recognition object and reducing the size of the DB efficiently using the same may be required.

Accordingly, it is a need of the present invention to provide a system and method that do not need to collect various images of a recognition target according to an embodiment of the present invention.

In addition, the present invention provides a system and a method for efficiently rationalizing the DB (downsizing the DB size) and raising the recognition rate of the recognition object.

According to an aspect of the present invention, there is provided an image recognition system, comprising: an image feature extraction unit for extracting feature points corresponding to an original feature point corresponding to an original image corresponding to an object to be recognized and a plurality of converted images corresponding to the original image, A) identifying the corresponding transformed feature points, and b) generating the at least one integrated feature point from the original feature points and the transformed feature points, wherein the transformed feature points are based on the original feature point Is specified.

Wherein the step a) includes the steps of sequentially converting the original image by a predetermined unit angle to generate the plurality of converted images, extracting feature points from each of the generated plurality of converted images, And specifying the transformed minutiae points corresponding to the original minutiae extracted from the original image from the minutiae points based on the converted angle between the original image and each of the plurality of transformed images.

The method of providing an image recognition system using the transformed image may include replacing the generated at least one integrated feature point with corresponding corresponding feature points, wherein the corresponding feature points are feature points integrated into the integrated feature point from the original feature point and the transformed feature points. And c) storing the data in the DB.

Wherein the step (b) comprises the steps of (b-1) specifying correspondent minutiae points in which a predetermined arithmetic value calculated based on the feature values of the original minutiae points and the transformed minutiae points satisfy an integration criterion, and And b-2) generating the integrated feature point based on the feature value of the feature point.

Wherein the step b-1) sets any one of the original feature point and the transformed feature point as a reference feature point, and when at least one candidate feature point included in remaining feature points other than the set reference feature point satisfies the integration criterion (B-1-1) of sequentially extracting the candidate minutiae and the reference minutiae, and (b-1-2) a step of identifying at least one candidate minutia and the reference minutiae sequentially extracted until the integration criterion is satisfied, as the corresponding minutiae can do.

The method for providing an image recognition system using the transformed image may further include a step b-1-3) of generating second integrated feature points on the original feature points and the excluded feature points except for the corresponding feature points, Wherein the step (b-1-3) includes the steps of performing the step (b-1-1) for the minutiae points and the step (b-1-1) 2 < / RTI > corresponding feature points.

In the step b-1-1), the candidate feature points are sequentially extracted in order of decreasing conversion angle based on the reference feature points.

Wherein the computed value is a sum of the variance of each of the corresponding attribute values of the corresponding feature points - each feature value of the corresponding feature points has n (n is a natural number or more) attribute values, And the criterion may be a criterion in which the calculated value becomes a predetermined reference value or less.

According to an aspect of the present invention, there is provided a method of providing an image recognition system using a transformed image, the method comprising: generating an original feature point corresponding to an original image corresponding to an object to be recognized and a plurality of transforms B) specifying conversion feature points corresponding to each of the images and corresponding to the original feature points, and b) generating, by the image recognition system, at least one integrated feature point from the original feature points and the conversion feature points, Wherein the at least one integrated feature point is generated based on a feature value of each of the corresponding feature points that satisfies the integration criterion, the predetermined operation value calculated on the basis of the feature value of each of the original feature point and the transformed feature points .

According to an aspect of the present invention, there is provided a method of providing an image recognition system using a transformed image, the method comprising: extracting first feature points corresponding to a first image corresponding to a recognition object; Extracting second feature points from each of the plurality of feature points, and the image recognition system integrating at least one of the first feature points and the second feature points to generate integrated feature points, And the second images are images generated by rotation-transforming the first image.

A method of providing an image recognition system using the transformed image may be stored in a computer-readable recording medium having recorded thereon a program.

According to an aspect of the present invention, there is provided an image recognition system using a transformed image, the image recognition system comprising: an original feature point corresponding to an original image corresponding to an object to be recognized and a plurality of transformed images corresponding to the original image; A feature point specifying module for specifying transformed feature points corresponding to the original feature point and a control module for generating at least one integrated feature point from the original feature point and the transformed feature points, And the conversion characteristic points are specified.

Wherein the feature point specifying module sequentially rotates the original image by a predetermined unit angle to generate the plurality of converted images and extracts the feature points from each of the generated plurality of converted images, From the feature points, the transformed feature points corresponding to the original feature points extracted from the original image, based on the converted rotation angles between the transformed images.

The image recognition system using the transformed image replaces the generated at least one integrated feature point with corresponding corresponding feature points, wherein the corresponding feature points are minutiae points integrated into the integrated feature point among the original feature points and the converted minutiae points, And a management module for storing the management information.

Wherein the control module specifies corresponding feature points that satisfy a merging criterion by a predetermined arithmetic operation value calculated on the basis of the feature value of each of the original feature point and the transformed feature points and based on the feature value of each of the specified corresponding feature points The integrated feature point can be generated.

Wherein the control module sets at least one of the original feature point and the transformed feature point as a reference feature point and at least one candidate feature point included in remaining feature points other than the set reference feature point, And at least one candidate feature point and reference feature point sequentially extracted can be specified as the corresponding feature points until the integration criterion is satisfied.

According to an aspect of the present invention, there is provided an image recognition system using a transformed image, the image recognition system comprising: an original feature point corresponding to an original image corresponding to an object to be recognized and a plurality of transformed images corresponding to the original image; And a control module for generating transformed minutiae corresponding to the original minutiae points, and a control module for generating at least one integrated minutiae from the original minutiae points and the transformed minutiae points, And the at least one integrated feature point is generated based on the feature value of each of the corresponding feature points, the predetermined operation value calculated based on the feature value satisfying the integration criterion.

According to an aspect of the present invention, there is provided an image recognition system using a transformed image, the image recognition system including a feature point specifying module and a control module for extracting first feature points corresponding to a first image corresponding to a recognition object,

Wherein the feature point specifying module extracts second feature points from a plurality of second images, the control module integrates at least one of the first feature points and the second feature points to generate integrated feature points, And the feature point specifying module rotates the first image to generate the plurality of second images.

An image recognition system using a transformed image and a method of providing the same according to an embodiment of the present invention may be configured such that when a plurality of images corresponding to an object to be recognized So that a more diverse image corresponding to the recognition object can be included in the reference data, thereby increasing the recognition rate.

In addition, when a plurality of images are actually collected, image capturing for this purpose must be performed. Not only is it not easy to collect images in various and precise angular units, but if images are missing from a specific angle, have. However, according to the present invention, since a desired image can be generated as much as desired, the recognition rate can be increased.

In addition, since the DB can be prevented from being enlarged by including a plurality of images in the reference data, the recognition time is increased due to a waste of cost due to the increase of the DB capacity and a remarkable increase of the amount of data stored in the DB And deterioration of recognition performance can be prevented.

In addition, since it is not necessary to determine which of the minutiae matching out of the plurality of minutiae extracted from the plurality of images for the purpose of efficiency of the DB is, it is possible to reduce resources (time and cost, etc.) There is an effect.

In addition, even if the rotation angle is large, even a transformed image generated from the original image may have a feature that is significantly different from the original image. Even in this case, any one of the specific points extracted from the original image, If all the feature points extracted from the transformed images corresponding to the original feature points are combined into one, there may be a problem that the recognition performance may be rather lowered. According to the present invention, it is possible to solve these problems rationally by dividing and combining the feature points according to the degree of deformation of the feature points.

BRIEF DESCRIPTION OF THE DRAWINGS A brief description of each drawing is provided to more fully understand the drawings recited in the description of the invention.
FIG. 1 is a diagram for explaining the difficulty of image recognition when the recognition object is a natural object.
FIG. 2 is a diagram for explaining a problem when each of a plurality of images corresponding to a specific recognition target is included in the reference data.
3 is a diagram showing a schematic configuration of an image recognition system using a transformed image according to an embodiment of the present invention.
4 to 5 are diagrams for explaining a concept of generating transformed images from an original image according to an embodiment of the present invention.
6 is a diagram for explaining a concept of performing DB efficiency through an image recognition system using a transformed image according to an embodiment of the present invention.
7 is a diagram for explaining a method of integrating feature points according to an embodiment of the present invention.
8 is a diagram for explaining a criterion for integrating minutiae according to an embodiment of the present invention.

In order to fully understand the present invention, operational advantages of the present invention, and objects achieved by the practice of the present invention, reference should be made to the accompanying drawings and the accompanying drawings which illustrate preferred embodiments of the present invention.

Also, in this specification, when any one element 'transmits' data to another element, the element may transmit the data directly to the other element, or may be transmitted through at least one other element And may transmit the data to the other component.

Conversely, when one element 'directly transmits' data to another element, it means that the data is transmitted to the other element without passing through another element in the element.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be described in detail with reference to the preferred embodiments of the present invention with reference to the accompanying drawings. Like reference symbols in the drawings denote like elements.

The technical idea of the present invention can be applied to all algorithms for performing image recognition by matching images (data) to be recognized in a reference image (data) using feature points such as SIFT and SURF (Speed Up Robust Features). However, for convenience of explanation, the SIFT algorithm will be described in this specification.

FIG. 3 shows a schematic configuration of an image recognition system 100 using a transformed image according to an embodiment of the present invention.

Referring to FIG. 3, an image recognition system (hereinafter referred to as an image recognition system 100) using a transformed image according to an embodiment of the present invention includes a control module 110 and a feature point specifying module 120. The image recognition system 100 using the transformed image may further include a management module 130 and a DB 140.

The image recognition system 100 may be implemented as a data processing apparatus (e.g., a computer, a server, a mobile terminal, or the like) capable of performing a method of providing an image recognition system according to the technical idea of the present invention. In addition, the image recognition system 100 may be defined to include not only a system for performing image recognition but also a system for generating a DB according to the technical idea of the present invention for image recognition.

To this end, the image recognition system 100 may include a hardware resource and / or software necessary for implementing the technical idea of the present invention, and it may mean one physical component or one device It does not. That is, the image recognition system 100 may mean a logical combination of hardware and / or software provided to implement the technical idea of the present invention. If necessary, the image recognition system 100 may be installed in a separate apparatus, The present invention may be embodied as a set of logical structures for realizing the technical idea of the present invention. In addition, the image recognition system 100 may mean a set of configurations separately implemented for each function or role for implementing the technical idea of the present invention. Depending on the implementation, each of the configurations included in the image recognition system 100 may also be implemented by a combination of hardware and / or software installed in separate devices. For example, the control module 110 may refer to a configuration in which hardware and / or software installed in each of a plurality of different physical devices are organically combined through a network to perform functions defined in this specification.

In this specification, a module may mean a functional and structural combination of hardware for carrying out the technical idea of the present invention and software for driving the hardware. For example, the module may refer to a logical unit of a predetermined code and a hardware resource for executing the predetermined code, and does not necessarily mean a physically connected code or a kind of hardware. Can be easily deduced to the average expert in the field of the present invention.

The control module 110 may control the functions of other components included in the image recognition system 100 such as the feature point specifying module 120, the management module 130, and / or the DB 140, And / or resources.

The feature point specifying module 120 may extract a key point from an image. The feature point may be a point that can characterize a specific image, and a method of extracting feature points from a predetermined image may be different depending on the type of the image recognition algorithm. For example, the image recognition system 100 using the transformed image according to the embodiment of the present invention can perform image recognition through the SIFT algorithm. Then, the feature point extraction module 120 can extract feature points from a predetermined image using a feature point extraction method used in the SIFT algorithm. The manner in which the SIFT algorithm extracts feature points is described in detail in the above-mentioned papers, and a detailed description thereof will be omitted herein.

In addition, the feature point specifying module 120 may generate a plurality of transformed images using a specific image, that is, an original image.

Then, the feature point specifying module 120 may extract feature points from the original image. And extract feature points from the plurality of transformed images, respectively.

Then, the feature point specifying module 120 extracts feature points (hereinafter referred to as " feature points ") corresponding to any one of feature points extracted from the original image , Referred to as conversion characteristic points). The transformed minutiae corresponding to the original minutiae point is a set of transformed minutiae corresponding to the transformed minutiae when the original image is set as the reference data and the transformed image in which the transformed minutiae are extracted is used as the input image (i.e., the image (or object) May mean feature points that match each other. That is, a feature point (for example, a transformation feature point) corresponding to one of the feature points (for example, the original feature point) and the one of the feature points (for example, the original feature point) (Or points) in the object appearing in different images (e.g., transformed images) and the same point (or points) in the object appearing in other images (e.g., transformed images).

In the registered patent, since a plurality of images representing the same recognition target are collected, it is not known what corresponding feature points are among the extracted feature points from each of the plurality of images. Thus, the matching is performed, and the matching points are combined to generate a combining point (referred to as an integrated characteristic point in this specification). However, a relatively large amount of resources have been collected to search for a matching point among a plurality of minutiae. In addition, feature points to be matched in the process of matching may not be matched or may be misaligned. However, according to the present invention, since the image recognition system 100 transforms and generates a plurality of different images from any one original image, it is possible to match the feature points corresponding to any one original feature point extracted from the original image It is possible to know without performing. Therefore, there is an effect that rapid DB efficiency (or DB downsizing) can be achieved.

For example, the feature point specifying module 120 may receive an original image corresponding to a specific recognition target. The feature point specifying module 120 may then generate a plurality of transformed images from the original image. The transformed images may be an image obtained by performing a rotation transformation of the original image at a predetermined angle.

An example in which the feature point specifying module 120 generates transformed images will be described with reference to FIGS. 4 and 5. FIG.

4 to 5 are diagrams for explaining a concept of generating transformed images from an original image according to an embodiment of the present invention.

Referring to FIG. 4, the feature point specifying module 120 may receive an original image (for example, 20). The feature point specifying module 120 may then convert the original image (e.g., 20) to a predetermined angle to produce a transformed image (e.g., 21 to 28).

For example, the feature point identification module 120 may pre-determine camera calibration based on the original image (e.g., 20). Then, at least one rotation direction of the camera in the camera coordinate system can be determined, and the rotation angle can be determined. 4 shows transformed images (for example, 21 to 28) generated by sequentially rotating the camera in a plane direction with a predetermined unit angle (e.g., 10 degrees) with reference to the front surface of the original image (for example, 20) As an example.

5, converted images (e.g., images) generated by sequentially rotating the camera in a vertical direction in the vertical direction with respect to the front face of the original image (e.g., 20) in a predetermined unit angle (e.g., 10) 29 to 36) are shown as an example.

The average expert in the technical field of the present invention can easily obtain the rotation transformation image of the original image (e.g., 20) in various numbers and various directions by variously setting the unit angle and / or the rotation direction. You can do it. Further, the rotation matrix for obtaining the rotation-transformed image of the original image (for example, 20) is widely known to the average expert in the technical field of the present invention, and thus a detailed description thereof will be omitted.

Referring again to FIG. 3, the feature point specifying module 120 may thus generate a plurality of transformed images (e.g., 21 to 36) from the original image (e.g., 20).

The feature point specifying module 120 may then extract feature points from each of the original image (e.g., 20) and the plurality of transformed images (e.g., 21 to 36). The feature point specifying module 120 may store information on the extracted feature points (for example, information on the extracted feature points, locations of the feature points, feature values (e.g., discriptor) of the feature points, have.

The feature point specifying module 120 may specify feature points corresponding to any one of the feature points extracted from the original image (e.g., 20), that is, converted feature points. At this time, the feature point specifying module 120 can specify the converted feature points based on the original feature points. Specifying the transformed minutiae points based on the original minutia points means that transformed minutiae points corresponding to the original minutiae points can be specified from among a plurality of minutiae points using information (e.g., position (coordinates) For example, since the feature point specifying module 120 can know the rotation angle of the original image (for example, 20) and the predetermined first converted image (for example, 21), the position of the original feature point extracted from the original image It is possible to know the position of the transformed minutiae corresponding to the original minutiae among the minutiae extracted from the first transformed image (for example, 21). For example, (I.e., a rotation angle from the original image to the first transformed image) of the original feature point, It may be converted feature point corresponding to the feature point.

In this way, the feature point specifying module 120 can extract the transformed feature points corresponding to the original feature points from among the feature points extracted from each of the plurality of transformed images (for example, 21 to 36).

Of course, there is no guarantee that any one of the minutiae in the original image is extracted as minutiae in the rotated image of the original image. In most cases, however, since the minutiae point corresponds to a characteristic moiety in the corresponding image, Unless the feature of the image is completely changed, the feature points in the original image may be extracted as feature points in the transformed image.

As described above, according to the technical idea of the present invention, the minutia point specifying module 120 can specify the corresponding converted minutiae points based on the original minutiae points without performing matching as in the above-mentioned registered patent. As a result, the set of feature points that can be integrated (i.e., the original feature point and the transformed feature points corresponding to the original feature point) can be specified relatively quickly and quickly.

Then, the control module 110 may integrate (combine) the transformed minutiae points corresponding to the original minutiae points and the original minutiae points in a predetermined manner to generate integrated minutiae points for DB efficiency.

Such an example will be described with reference to FIGS. 6 to 8. FIG.

6 is a diagram for explaining a concept of performing DB efficiency through an image recognition system using a transformed image according to an embodiment of the present invention. 7 is a view for explaining a method of integrating feature points according to an embodiment of the present invention, and FIG. 8 is a diagram for explaining a reference for integrating feature points according to an embodiment of the present invention.

Referring to FIG. 6, information on the minutiae extracted by the minutia matching module 120 may be stored in the DB 140. For example, feature points (e.g., 50, 51, 52, 53) may be extracted from an original image (e.g., 20). And feature points (e.g., 50-1, 51-1, 52-1, 53-1) may be extracted from the first transformed image (e.g., 21). In addition, feature points (e.g., 50-2, 51-2, 52-2, 53-2) may be extracted from the second transformed image (e.g., 36). Then, the DB 140 may conceptually store feature points as shown in FIG.

(For example, 50-1, 50) corresponding to a predetermined original feature point (for example, 50) among the feature points extracted from the original image (for example, 20) by the feature point specifying module 120, -2) can be specified. (For example, 51-1, 51-2) corresponding to a second original feature point (for example, 51), third conversion feature points (for example, 52-1, 52-2, and fourth transform feature points (e.g., 53-1, 53-2) corresponding to a fourth original feature point (e.g., 53).

Then, the control module 110 can perform the integrated feature point generation process on the transformed minutiae points (e.g., 50-1 and 50-2) corresponding to any one original feature point (e.g., 50). Through the integrated feature point generation process, the control module 110 may replace the original feature point (e.g., 50) and the converted feature points (e.g., 50-1, 50-2) with at least one integrated feature point. Although only two transformation feature points (e.g., 50-1 and 50-2) corresponding to the original feature point (for example, 50) are shown in Fig. 6 for convenience, it is assumed that as many transformation feature points as the number of transformation images can be specified Of course.

The control module 110 may generate the original feature point (e.g., 50) and the transform feature points (e.g., 50-1, 50-2) as one integrated feature point. However, in this case, since the feature points of the feature points (for example, the descriptors of the SIFT algorithm) significantly changed by the rotation transformation are integrated into one integrated feature point, the recognition rate may be lowered rather. Accordingly, among the set of feature points that can be integrated (i.e., the original feature points (e.g., 50) and the converted feature points (e.g., 50-1 and 50-2)), only the feature points having similar feature values are integrated, May be desirable. Then, the management module 130 may store the generated integrated feature points in the DB 140 in place of corresponding corresponding feature points (i.e., integrated feature points). That is, the corresponding feature points may be deleted from the DB 140, and the integrated feature points may be stored in the DB 140. As a result, the efficiency of the DB 140 is enabled and the time required for recognition is reduced, and the recognition rate can be increased due to the integrated feature points. The reason why the recognition rate is increased when the minutiae having similar feature values are integrated is disclosed in detail in the patent, so that detailed description will be omitted.

To this end, the control module 110 can generate integrated feature points only when the feature values of the feature points to be integrated satisfy a certain reference, that is, the integration criteria. The integration criterion may be a criterion by which the feature values can extract similar feature points.

For example, in the case of the SIFT algorithm, the feature value may have 128 attribute values, and the feature value of each feature point may be interpreted as a coordinate where the feature point is located on a 128-dimensional space. Accordingly, the control module 110 can integrate only the minutiae points existing within a close range in the space according to a certain criterion.

For this, the control module 110 performs a predetermined calculation based on the feature points to be integrated, that is, the feature values of corresponding feature points corresponding to the integrated feature points, Standard) is satisfied. As a result of the determination, the feature points satisfying the predetermined criteria can be integrated into the integrated feature points.

According to an embodiment, the calculated value may be the sum of variances of each of the attribute values of corresponding feature points. And the integration criterion may be such that the sum of variances of each of the attribute values of the corresponding feature points is less than or equal to a predetermined value. That is, the large sum of the dispersions may mean that the feature points are relatively scattered in the 128-dimensional space as described above. Therefore, the dispersion can be used as a criterion for integrating only the relatively concentrated feature points in the space. However, it will be easily understood by those of ordinary skill in the art that various embodiments may be possible for integrating only feature points having similar feature values in addition to using such variance.

When the corresponding feature points satisfying the integration criterion are specified, the control module 110 may combine the corresponding feature points to generate integrated feature points. The feature value of the integrated feature point may be an average of the attribute values of the corresponding feature points. However, it goes without saying that various embodiments may be possible. For example, a feature value of one of the corresponding feature points may be used as a feature value of the integrated feature point.

7 and 8, the control module 110 will explain the process of generating the integrated feature point, according to one embodiment.

7 and 8, the control module 110 may include a set of feature points to be integrated (e.g., transform feature points corresponding to an original feature point (e.g., K11) and the original feature point (e.g., K11) , K13, K14, K1N) from the feature point specifying module 120. [ Where Ki1 may denote the i-th original feature point of the original image, and Ki2 to KiN may denote the conversion feature points of Ki1 extracted from each of the transformed images.

The control module 110 may extract transformed minutiae points (for example, K12 to K1N) one by one based on the original minutiae (for example, K11) to determine whether the unified criterion is satisfied. Of course, any feature point among the collectors (for example, K11 to K1N) can be used as a reference without referring to the original feature point (e.g., K11). According to an example, the control module 110 may extract feature points in the order of the transformed feature points extracted from the image having a small rotation angle from the original image. In this case, since the rotation angle is small, the difference of the feature values can be small, so that a larger number of feature points can be integrated into the integrated feature points. For convenience of explanation, K12, K13, ... , And K1N are extracted, and the order of extraction is assumed to be the order in which the rotation angle is small as described above. However, it goes without saying that various embodiments are possible.

The feature values of the original feature points (e.g., K11) are {8, 18, 8, 2, ... , 7}. The characteristic values of the first conversion feature point (for example, K12) are {9, 17, 9, 2, ... , 10}.

Then, the control module 110 determines whether a candidate feature point (for example, a first transformed feature point (e.g., K12)) extracted from the reference feature point (e.g., original feature point (e.g., K11) (S100). To this end, the first transformed feature point (e.g., K12) determines the variance of each of the corresponding property values of the feature points to be incorporated (i.e., the original feature point (e.g., K11) and the first transform feature point (e.g., K12) . Since the first attribute values are 8 and 9 respectively, the average may be 8.5 and the variance may be 2 x (0.5) ^ 2. In this way, the variance can be found for all property values, and the sum of the variances obtained can be found. The sum of the variances may be less than a predetermined reference value. Then, the control module 110 determines that the integration criterion is satisfied, and returns one feature point, that is, a second transformed feature point (for example, K13) as a candidate feature point (that is, a feature point to be a candidate for further integration) . If the integration criterion is not satisfied (S100), only the minutiae that satisfy the integration criterion (in this case only one original minutiae point K11) are integrated (S140) and the integrated minutiae point (in this case, For example, the feature point merging process may be performed again on the minutiae except minus one), i.e., minus minutiae (e.g., K12 to K1N) (S150). At this time, the first transformed feature point (e.g., K12) may become the reference feature point again. Then, when the original feature point (e.g., K11) becomes the first integrated feature point and the feature point integration process performed on the excluded feature points is performed, the first transformed feature point (e.g., K12) Can be generated.

When the second conversion feature point (e.g., K13) is extracted (S100), the control module 110 determines that the feature point (e.g., K11) and the first feature point It is possible to determine whether or not the integration criterion is satisfied even when the candidate feature point, i.e., the second transformed feature point (e.g., K13) is added to the transformed feature point (e.g., K12) The second transformed feature point (e.g., K13) is a feature point having a feature value of {7, 15, 11, 2, ... , 10}.

Then, the control module 110 determines whether or not each of the corresponding attribute values of the feature points to be integrated (original feature point (e.g., K11), first transform feature point (e.g., K12), and second transform feature point (e.g., K13) Can be determined. Since the first attribute values are 8, 9, and 7, respectively, the average is 8 and the variance can be 2. Also, the second attribute values are 18, 17, and 15, respectively, with an average of 17 and a variance of 2. In this way, we can obtain the variance for all attribute values and obtain the sum of the variances. Then, it can be determined whether the sum of the variances obtained is equal to or less than a predetermined reference value (S110).

If the integration criterion is satisfied (S110), the feature points (e.g., original feature points (e.g., K11), first transform feature points (e.g., K12), second transform feature points (e.g., K13) And a candidate feature point (third conversion feature point (e.g., K14)) can be extracted again. When the candidate feature point is added to the feature points (e.g., the original feature point K11, the first transform feature point K12, and the second transform feature point K13, for example), it is determined whether the integration criterion is satisfied (S120).

If the integration criterion is not satisfied (S110), only the feature points (in this case, the original feature point (for example, K11) and the first transform feature point (for example, K12)) are integrated (S140) The feature point integration process can be performed again on the feature points excluding the feature points (in this case, the original feature point (e.g., K11) and the first converted feature point (e.g., K12)), i.e., the excluded feature points (e.g., K13 to K1N) (S150). At this time, the second transformed feature point (e.g., K12) may become the reference feature point again. Then, a first integrated feature point where an original feature point (e.g., K11) and a first transform feature point (e.g., K12) are integrated, and a feature point integration process performed on the excluded feature points is performed, K13), and a second integrated feature point in which several feature points are integrated.

In this manner, the control module 110 may perform the integrated feature point generation process for both the original feature point (e.g., K11) and the corresponding transformed feature points (e.g., K12 to K1N). When the last transformed feature point (e.g., K1N) is extracted and satisfies the integrated criterion, an integrated point including the last transformed feature point may be generated (S160).

The integration criterion is satisfied up to the original feature point (e.g., K11), the first transformed feature point (e.g., K12), and the second transformed feature point (e.g., K13) (S110) and a third transformed feature point (e.g., K14) Assuming that the integration criteria are not satisfied in step S120, the control module 110 determines whether or not the feature points satisfying the integration criteria in the previous step S110 (i.e., the original feature point (e.g., K11) (E.g., K12), and a second transformed feature point (e.g., K13)) to generate integrated feature points. In this case, the integrated feature point may be an average of the attribute values of the feature points, and as shown in FIG. 8, {8, 17, 9, 2, ... , 9}.

In this manner, the control module 110 may perform an integrated feature point generation process on all the original feature points (e.g., K11, K21, K31, etc.) extracted from the original image (e.g., 20). Then, the efficiency of the DB 140 can be completed.

In addition to the embodiments described with reference to FIGS. 7 and 8, the image recognition system 100 may extract a predetermined number of feature points sequentially in accordance with a rotation angle, and integrate them into integrated feature points. It will be readily apparent to one of ordinary skill in the art that various embodiments are possible.

The method of providing an image recognition system using a transformed image according to an embodiment of the present invention can be implemented as a computer-readable code on a computer-readable recording medium. A 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, hard disk, floppy disk, optical data storage, and the like. In addition, the computer-readable recording medium may be distributed over network-connected computer systems so that computer readable codes can be stored and executed in a distributed manner. And functional programs, codes, and code segments for implementing the present invention can be easily inferred by programmers skilled in the art to which the present invention pertains.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

Claims (18)

Wherein the image recognition system identifies transformed minutiae corresponding to each of the original minutiae of the original minutiae corresponding to the original image representing the recognition object and a plurality of transformed images corresponding to the original image, ) step; And
Wherein the image recognition system generates at least one integrated feature point from the original feature point and the transformed feature points,
The conversion feature points may include,
Wherein the rotation angle is specified based on a rotation angle between the original image and each of the plurality of transformation images.
The method of claim 1, wherein the step a)
The image recognition system sequentially converting the original image by a predetermined unit angle to generate the plurality of converted images;
Extracting feature points from each of the generated plurality of transformed images; And
And specifying the transformed minutiae points corresponding to the original minutiae extracted from the original image, from the minutiae points, based on the transformed rotation angle between the original image and each of the plurality of transformed images, / RTI >
The method as claimed in claim 1,
Storing the generated at least one integrated feature point in the DB in correspondence to corresponding feature points corresponding to the integrated feature points, wherein the corresponding feature points are minutiae points integrated into the integrated feature point among the original feature points and the converted minutiae points; Further comprising the steps of:
2. The method of claim 1, wherein step b)
B-1) specifying a corresponding feature point that satisfies an integration criterion by a predetermined computation value calculated based on the original feature point and the feature value of each of the transformed feature points; And
And b-2) generating the integrated feature point based on the feature value of each of the specified corresponding feature points.
5. The method of claim 4, wherein the step (b-1)
Wherein the feature point extraction unit is configured to set a feature point of any one of the original feature point and the transformed feature point as a reference feature point and to sequentially extract at least one candidate feature point included in remaining feature points other than the set reference feature point, 1-1) step; And
And b-1-2) specifying at least one candidate feature point and the reference feature point that are sequentially extracted until the integration criterion is satisfied, as the corresponding feature points.
[6] The method of claim 5,
And (b-1-3) generating second integrated feature points on the extracted feature points and the converted feature points, excluding the corresponding feature points, on the basis of the extracted feature points,
The step (b-1-3)
Performing the step b-1-1) for the minus minutiae points; And
And specifying the minutiae identified as a result of the b-1-1) as second corresponding minutiae corresponding to the second integrated minutiae.
6. The method of claim 5, wherein the step b-1-1)
And extracting the candidate feature points sequentially in order of decreasing conversion angles based on the reference feature points.
5. The method of claim 4,
Characterized in that the corresponding feature points - the feature values of each of the corresponding feature points - are each a sum of variances of respective attribute values of n (where n is a natural number or more)
The integration criterion may include:
Wherein the calculated value is a reference that is less than or equal to a predetermined reference value.
Wherein the image recognition system converts transformed minutiae corresponding to each of the source minutiae of the source minutiae corresponding to the original image representing the recognition object and the plurality of transformation images corresponding to the source image, And a rotation angle between each of the plurality of transformed images; And
Wherein the image recognition system generates at least one integrated feature point from the original feature point and the transformed feature points,
Wherein the at least one integrated feature point comprises:
And the predetermined calculation value calculated based on the feature value of each of the original feature point and the transformed feature point is generated based on the feature value of each of the corresponding feature points satisfying the integration criterion. Delivery method.

Extracting first feature points corresponding to a first image representing an object to be recognized by the image recognition system;
Extracting second feature points from the plurality of second images, respectively; And
Wherein the image recognition system integrates at least one of the first feature points and the second feature points to generate integrated feature points,
The plurality of second images may include:
An image generated by rotation transformation of the first image,
The image recognition system includes:
Wherein the integrated feature point is generated using the rotation angle between the first image and each of the plurality of second images.
A computer-readable recording medium recording a program for performing the method according to any one of claims 1 to 10.
A feature point specifying module that specifies any one of the original feature points corresponding to the original image representing the recognition target and the transformed feature points corresponding to each of the plurality of transform images corresponding to the original image and corresponding to the original feature point; And
And a control module for generating at least one integrated feature point from the original feature point and the transformed feature points,
The feature point specifying module includes:
Wherein the transformed feature points are specified based on the transformed rotation angle between the original image and each of the plurality of transformed images.
13. The method according to claim 12,
Sequentially converting the original image by a predetermined unit angle to generate the plurality of converted images,
Extracting feature points from each of the generated plurality of transformed images,
And a transformed image that specifies the transformed minutiae points corresponding to the original minutiae extracted from the original image based on the transformed rotation angle between the original image and each of the plurality of transformed images.
13. The image recognition system according to claim 12,
Further comprising a management module for replacing the generated at least one integrated feature point with corresponding corresponding feature points, wherein the corresponding feature points are feature points integrated into the integrated feature point among the original feature points and the converted feature points, Image recognition system using transformed images.
13. The apparatus of claim 12,
Wherein the feature point extraction unit specifies the corresponding feature points that satisfy the integration criterion and the predetermined operation value calculated based on the feature value of each of the original feature point and the transformed feature points, and determines the integrated feature point based on the feature value of each of the specified corresponding feature points. Image Recognition System Using Transformed Image.
16. The apparatus of claim 15,
Setting at least one of the original feature point and the transformed feature point as a reference feature point and sequentially extracting at least one candidate feature point included in remaining feature points other than the set reference feature point until the integration criterion is satisfied,
And a transformation image for specifying at least one candidate feature point and the reference feature point sequentially extracted until the integration criterion is satisfied as the corresponding feature points.
Wherein the transformed feature points corresponding to each of the original feature points corresponding to the original image representing the recognition target and the plurality of transformation images corresponding to the original image and corresponding to the original feature points, A feature point specifying module that specifies based on the converted rotation angle between each of the transformed images; And
And a control module for generating at least one integrated feature point from the original feature point and the transformed feature points,
The control module includes:
Characterized in that the at least one integrated feature point is generated based on each feature value of each of the corresponding feature points whose predetermined computed value calculated based on the feature values of the original feature point and each of the transformed feature points satisfy an integration criterion Image Recognition System Using Transformed Images.
A feature point specifying module for extracting first feature points corresponding to a first image representing a recognition target; And
A control module,
The feature point specifying module includes:
Extracting second feature points from the plurality of second images,
The control module includes:
Integrating at least one of the first feature points and the second feature points to generate integrated feature points,
The feature point specifying module includes:
Characterized in that the first image is rotated and converted to generate the plurality of second images,
And a transformed image that specifies at least one of the second feature points to be incorporated into the integrated feature point based on the converted rotation angle between the first image and each of the plurality of second images.

Images