KR101145278B1 - Method, apparatus and computer-readable recording medium for choosing representative images among similar images - Google Patents

Abstract

PURPOSE: A method for selecting a representative, an apparatus thereof, and computer readable medium are provided to increase searching speed, to reduce the number of image nodes, and to select a representative image. CONSTITUTION: A similarity measuring unit(510) obtains image similarities between image nodes with reference to image information. A similarity comparing unit(520) compares the image similarity with critical similarity. A representative image selecting unit(550) merges image nodes corresponding to image similarity. The representative image selecting unit selects a first feature image as the representative image.

Description

METHOD, APPARATUS AND COMPUTER-READABLE RECORDING MEDIUM FOR CHOOSING REPRESENTATIVE IMAGES AMONG SIMILAR IMAGES}

The present invention relates to a method, an apparatus and a computer readable recording medium for selecting a representative image from similar images. More specifically, by matching the image or information about the image to the image nodes, and then merging the image nodes having a high image similarity with each other to reduce the number of image nodes, image search while maintaining the accuracy of the search A method, apparatus and computer readable recording medium for selecting a representative image from similar images, which can reduce the number of images used for indexing.

When a user inputs a specific image as a query, there is a problem that the search may take a long time because the search has to be performed while calculating the similarity for all images stored in the image database. In addition, in order to accurately provide search results that match the user's query image, it is necessary to store and maintain as many images in the image database as possible, and accordingly, the storage space is consumed excessively, and the calculation is performed due to the similarity calculation. There was a problem of slowing down.

For example, related technologies include Ondrej Chum, James Philbin, Josef Sivic, Michael Isard and Andrew Zisserman Visual Geometry Group, Department of Engineering Science, University of Oxford Microsoft Research, Silicon Valley "Total recall: autimatic query expansion with a generative feature model for object retrieval ", we see that the image expansion is applied to the query expansion technique that is widely used for text search. Specifically, referring to FIG. 1 of the document, a search extension technique is used to search for multiple images including the same object.

However, in order to apply the search extension technique to the image search, since many images containing the same object must be stored in the database, a large amount of storage space is consumed and the search speed becomes slow. In addition, in order to expand the search, images containing the same object are stored in the database of images with high similarity, but storing the images with high similarity does not significantly affect the accuracy of the search results. This requires a lot of computation, which causes problems that slow down the search.

An object of the present invention is to solve the above-mentioned problem.

According to the present invention, images or information about them are respectively mapped to image nodes, and then image nodes having large image similarities are merged with each other to reduce the number of image nodes so that a representative image is selected. Another goal is to reduce the number of images used in the image search index while maintaining.

In addition, it is another object to improve the search speed by reasonably reducing the number of images used in the image search index.

In order to accomplish the above object, a representative structure of the present invention is as follows.

According to an aspect of the present invention, (a) when at least one image or information on the image corresponds to an image node, image similarities between a plurality of image nodes may correspond to the image corresponding to each of the image nodes or the Obtaining information about an image with reference, (b) comparing a critical similarity with each of the obtained image similarities, and (c) if there is an image similarity greater than the critical similarity among the image similarities, Merging image nodes corresponding to an image similarity larger than the threshold similarity to form an image node, and selecting a first specific image as the representative image from an image corresponding to the merged image node or an image indicated by the information about the image. Provided is a method comprising the steps of:

According to another aspect of the present invention, when at least one image or information on the image corresponds to an image node, image similarities between a plurality of image nodes may be compared to an image corresponding to each of the image nodes or the image. A similarity measurer which obtains information with reference, a similarity comparator that compares each of the obtained image similarities with a critical similarity, and an image similarity greater than the threshold similarity among the image similarities, which is greater than the threshold similarity. A representative image selector configured to merge image nodes corresponding to image similarities to form an image node, and to select a first specific image as the representative image among an image corresponding to the merged image node or an image indicated by the information about the image; An apparatus is provided.

According to the present invention, images or information about them are respectively mapped to image nodes, and then image nodes having large image similarities are merged with each other to reduce the number of image nodes so as to select a representative image, thereby making the search accurate. It is possible to achieve the effect of improving the search speed while maintaining the.

1 is a flowchart of a method for selecting a representative image according to an embodiment of the present invention.
2 is an exemplary diagram to which a method for selecting a representative image according to an embodiment of the present invention is applied.
3 is a diagram schematically illustrating a configuration of an entire system for providing an image matching service using an image search index including a representative image according to an embodiment of the present invention.
4 is a diagram illustrating an internal configuration of the image matching system 500 according to an embodiment of the present invention.

DETAILED DESCRIPTION The following detailed description of the invention refers to the accompanying drawings that show, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain features, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in connection with an embodiment. It is also to be understood that the position or arrangement of the individual components within each disclosed embodiment may be varied without departing from the spirit and scope of the invention. The following detailed description, therefore, is not to be taken in a limiting sense, and the scope of the present invention, if properly described, is defined only by the appended claims, along with the full range of equivalents to which such claims are entitled. Like reference numerals in the drawings refer to the same or similar functions throughout the several aspects.

DETAILED DESCRIPTION Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement the present invention.

According to one embodiment of the invention, a method of selecting a representative image from similar images may be used to build an image search index.

1 is a flowchart of a method for selecting a representative image from similar images in accordance with one embodiment of the present invention.

2 is an exemplary diagram applying a method of selecting a representative image from similar images according to an embodiment of the present invention.

Referring to FIGS. 1 and 2A, image similarities S12, S13, and S23 between the plurality of image nodes IN1, IN2, and IN3 are obtained, respectively (110).

An image node (eg, IN1) corresponds to image I1. The image node IN1 may include the corresponding image I1 itself or may include information about the corresponding image I1. The information about the image I1 may be, for example, information about feature information of the object portion displayed on the image I1, information about a name of the image I1, or path information indicating the image I1. Etc., but this is merely an example and may be any information about the image I1. In addition, the number of images included in the image node may be singular or plural. It is assumed that the number of images included in one image node is singular unless specifically referred to as plural.

The image similarity (eg, S12) is a numerical representation of the degree of similarity between the image I1 corresponding to the image node IN1 and the image I2 corresponding to the image node IN2. In the present specification, the image similarity is a real number between 0 and 1, and as the numerical value increases, the degree of similarity is described. However, this is merely a mere example and the manner of expressing the image similarity may be variously modified. On the other hand, it is noted that the expression image similarity between image nodes described herein may mean image similarity between images corresponding to the image nodes.

Here, the similarity of an image may mean a similarity of a face of a person included in an image or a similarity of an object included in an image, wherein the similarity is matched between these feature information by using feature information extracted from each image. You can judge by looking at the degree. For example, if the image contains the face of a person, the feature image can be extracted from the face image, such as the eye's midpoint, the end points of each eye, the end points and midpoints of the eyebrows, and the end points of the lips. will be. Techniques that define methods such as the similarity of faces or the similarity of things are well known techniques, and detailed descriptions thereof will be omitted.

Returning to FIG. 2A, the image similarities S12, S13, and S23 may be obtained based on the feature information of the object part included in the images I1, I2, and I3. Specifically, in order to obtain an image similarity (for example, S12), the feature information of the object portion displayed in the image I1 and the feature information of the object portion displayed in the image I2 are compared with each other and accordingly the image similarity ( S12) can be obtained.

In addition, as another embodiment of the present invention, image similarities S12, S13, and S23 may be obtained based on names of the images I1, I2, and I3. Specifically, in order to obtain the image similarity (eg, S12), the image similarity S12 may be obtained according to the degree of similarity between the name of the image I1 and the name of the image I2. This will serve as an example to show that the image similarities S12, S13, and S23 are not necessarily obtained by the degree of similarity of the feature information, and in addition to obtaining the image similarities S12, S13, and S23. Of course, various examples can be assumed.

Meanwhile, referring back to FIG. 1, in operation 110, image similarities S12, S13, and S23 are directly obtained through feature information comparison or name comparison, but image similarities S12, S13, and S23 that have already been obtained. It may be obtained and used. In operation 110, some of the image similarities S12, S13, and S23 may be directly obtained, and the remaining values may be obtained and used.

Next, each of the image similarities S12, S13, and S23 between the plurality of image nodes IN1, IN2, and IN3 may be compared with the threshold similarity Sref (120). Although the critical similarity Sref is shown as 0.5 and the image similarities S12, S13, and S23 are respectively 0.9, 0.2, and 0.3 in FIG. 2A, the numerical values may be variously modified. to be.

Referring to FIGS. 1 and 2B, when a comparison result of operation 120 of FIG. 1 shows that there is an image similarity S12 that is greater than a threshold similarity Sref among the image similarities S12, S13, and S23, the image The image nodes IN1 and IN2 corresponding to the similarity S12 are merged to form one image node IN4, and one of the images I1 and I2 corresponding to the image node IN4 is represented as a representative image. Select 130.

In order to select the representative image, one of the images I1 and I2 included in the image node IN4 may be randomly selected, and among the images I1 and I2 included in the image node IN4, An image I1 having low image similarity between the image I3 and the non-image I3 may be selected as the representative image. In this case, the similarity between the image I1 selected as the representative image and the image I3 not merged can be lowered, so that a more accurate search can be made.

Next, an image similarity S34 between the image I3 included in the unmerged image node IN3 and the image node IN4 formed by merging is obtained (140). The image similarity S34 is the maximum among the image similarities between the image I3 included in the unmerged image node IN3 and each of the images I1 and I2 included in the image node IN4 formed by merging. Can be set to a value. Specifically, the image similarity S13 between the image I3 included in the image node IN3 and the image I1 included in the image node IN4, and the image I3 included in the image node IN3, and The image similarity S23 between the images I2 included in the image node IN4 may be compared and set to the largest value. In FIG. 2A, the image similarity S13 and the image similarity S23 are shown to be 0.2 and 0.3, respectively. In FIG. 2B, the image similarity S34 is set to 0.3, which is the maximum value of 0.2 and 0.3. Is shown. Of course, the present invention is not limited thereto, and among the images I1 and I2 included in the image node IN4, the image I1 having a low image similarity between the unmerged images I3 is selected to select the representative image. If is selected as the representative image, 0.2, which is an existing image similarity S13 between the image I3 and the image I1, may be selected as the value of the image similarity S34.

Next, the image similarity S34 may be compared with the threshold similarity Sref again (120 of FIG. 1). As a result of the comparison, since the image similarity S34 is smaller than 0.5 which is the critical similarity Sref, the process is terminated without performing steps 130 and 140 of FIG. 1.

On the other hand, unlike shown in FIG. 2B, if the newly set image similarity is greater than the threshold similarity Sref, steps 130, 140, and 120 of FIG. 1 may be performed again. This process will be described with reference to FIG. 2C, and the image similarity between the image node IN4 formed by merging the image node IN1 and the image node IN2 and the non-merged image node IN5 S45. Is greater than the threshold similarity Sref. In this case, the image node IN4 and the image node IN5 are merged to form the image node IN6 (130). Next, an image similarity S36 between the image node IN3 and the image node IN6 is obtained (140) and compared with the threshold similarity Sref (120). Although FIG. 2C illustrates that the process is repeated twice, this process may be repeated three or more times depending on the number of image nodes and the image similarity between each other.

Next, since the image similarity S36 is smaller than the threshold similarity Sref, it may be easily understood that the steps 130 and 140 of FIG. 1 are not performed any more by referring to the foregoing description. Omit.

As such, the process may be repeated until there is no image similarity greater than the threshold similarity Sref (ie, until all image similarities are less than the critical similarity Sref), but all image similarities are Even before it becomes smaller than Sref, the process may be stopped. For example, referring to the second drawing of FIG. 2C, in a situation where the image similarity S45 is greater than the threshold similarity Sref, the process is stopped and corresponding to the image nodes IN3, IN4, and IN5. One image may be selected as a representative image among image nodes. As the above process is repeated, the number of image nodes decreases, and thus, a representative representative image among various images can be selected. Thereby, the size of the image search index for storing the representative image can be reduced, and the effect of improving the search speed while not having a great influence on the accuracy of the search results can be achieved.

In the above, the embodiment of the present invention has been described to start to be applied in a state in which one image node corresponds to one image. For example, referring to the first drawings of FIGS. 2A and 2C, each of the image nodes IN1, IN2, IN3, and IN5 is connected to one image I1, I2, I3, and I5. It has been shown correspondingly, and in this state embodiments of the present invention have been described as starting to apply. However, embodiments of the present invention may begin to be applied in a state where one image node corresponds to a plurality of images (image groups). For example, as shown in the second diagram of FIG. 2C, in the state in which two images are mapped to one image node I4, an embodiment of the present invention may be applied.

Meanwhile, in the embodiment of the present invention, as the threshold similarity Sref is lowered, the number of remaining image nodes can be reduced and search time can be shortened. However, if the threshold similarity (Sref) is excessively lowered, since the image nodes corresponding to images having low image similarity are merged with each other, the accuracy of the search may be reduced, and thus the threshold similarity (Sref) may be adjusted in consideration of this point. will be. In this case, the threshold similarity may be adaptively changed. For example, if the number of nodes merged among the total number of nodes for a predetermined time is calculated as a probability, and the probability exceeds the predetermined first threshold, the threshold similarity may be changed. (Sref) may be automatically increased, or if the probability is less than the second predetermined threshold, the threshold similarity (Sref) may be automatically lowered. (An exit criterion other than threshold similarity can be defined by how much merging has taken place. That is, if the input is 10 images and the exit criterion is 30%, the termination criterion terminates when there are three remaining nodes. Can be used as

On the other hand, in Figure 2 (a), (b), (c) it is shown that two image nodes are merged, but according to the embodiment of the present invention it is also possible to merge three or more image nodes at a time. For example, referring to FIG. 2C, all the image similarities S12, S15, and S25 among the three image nodes IN1, IN2, and IN5 are higher than the threshold similarity Sref. In this case, three image nodes IN1, IN2, and IN5 may be merged at once. In this case, the image similarity S36 between the image node IN6 formed by merging and the image node IN3 that is not merged is the image nodes IN1, IN2, and IN5 that are merged with the image node IN3 that is not merged. ) May be set to a maximum value among the image similarities S13, S23, and S35.

2 (a), (b) and (c) show the formation of one image node at a time, it is also possible to form two or more image nodes together. For example, according to an embodiment of the present invention, the image nodes may be merged to form a first image node, and other image nodes other than the image nodes may be merged to form a second image node (three or more). It is also possible to form image nodes together). In addition, a specific image may be selected as a representative image among an image corresponding to two or more merged image nodes or an image indicated by the information about the image.

Next, image similarities between two or more merged image nodes may be obtained, wherein if there is an image similarity larger than the threshold similarity among the acquired image similarities, the image corresponding to the image similarity larger than the threshold similarity The nodes may be merged again to form an image merge node, and again, a specific image may be selected as a representative image from among images indicated by the image or information corresponding to the merged image node.

According to an embodiment of the present invention, the method of selecting a representative image from similar images may further include storing only each representative image selected in step 130 of FIG. 1 in the image search index. As a result, the size of the image search index can be reduced, and the search speed can be improved.

In addition, according to an embodiment of the present invention, in the method of selecting a representative image from similar images, when a predetermined image is input as a query, the input image is compared with the selected representative images and the comparison result is referred to. The method may further include providing a matched representative image among the representative images as a search result for the query.

3 is a diagram schematically illustrating a configuration of an entire system for providing an image matching service using an image search index including a representative image selected from similar images according to an embodiment of the present invention.

As shown in FIG. 3, the entire system according to an embodiment of the present invention includes a communication network 400, an image matching system 700 including a representative image selection system 500, and a user terminal device 600. Can be configured.

First, the communication network 400 may be configured regardless of its communication mode such as wired and wireless, and may include a local area network (LAN), a metropolitan area network (MAN), and a wide area network (WAN). Network). Preferably, the communication network 100 according to the present invention may be a known World Wide Web (WWW).

According to an embodiment of the present invention, the image matching system 700 including the representative image selection system 500 receives images from the user terminal device 600 and performs image matching, the image having a large image similarity between each other By merging the nodes together, we are able to extract distinctive representative images from a number of images and use them in the image search index.

On the other hand, the user terminal device 600 according to an embodiment of the present invention is a digital device that includes a function to enable the user to communicate after connecting to the image matching system 700, a personal computer (for example, desktop Computer, laptop computer, etc.), a workstation, a PDA, a web pad, a mobile phone, etc., any memory having a memory means and equipped with a microprocessor equipped with arithmetic capability, any number of digital devices are adopted as the user terminal device 600 according to the present invention Can be.

Hereinafter, the internal structure of the representative image selection system 500 performing important functions for the implementation of the present invention and the function of each component will be described.

4 is a diagram illustrating an internal configuration of a representative image selection system 500 according to an embodiment of the present invention.

Referring to FIG. 4, the image matching system 500 according to an embodiment of the present invention may include a similarity measurer 510, a similarity comparator 520, a representative image selector 550, and an image search index 570. ) May be included.

The similarity measurer 510 may compare image similarities Sij between the plurality of images I1 to In or the plurality of image nodes IN1 to INn respectively corresponding to information about the images I1 to In or the images I1 to In. i and j respectively obtain different natural numbers less than n). The similarity comparator 520 compares each of the image similarities Sij acquired by the similarity measurer 510 with a threshold similarity Sref. The representative image selector 550 merges the image nodes corresponding to the large image similarity when the image similarity larger than the threshold similarity Sref among the image similarities Sij as a result of the comparison by the similarity comparator 520. An image node is formed and one of the images corresponding to the merged image node or the images indicated by the corresponding image information is selected as the representative image (IRk, k is a natural number). In this case, the similarity measurer 510 obtains image similarities between each of the image nodes that are not merged and the merged image nodes in the representative image selector 550, and the similarity comparator 520 performs the comparison process again. As mentioned earlier, you may merge all nodes that exceed the threshold similarity (Sref) in one step.

The image search index 570 may store selected representative images IRk and k are natural numbers, and may be used for image search.

According to an embodiment of the present invention, at least a portion of the similarity measurer 510, the similarity comparator 520, the representative image selector 550, and the image search index 570 may be a user terminal device 600. And program modules in communication with the. Such program modules may be included in the representative image selection system 500 in the form of operating systems, application modules, and other program modules, and may be physically stored on a variety of known storage devices. Such program modules may also be stored in a remote storage device capable of communicating with the representative image selection system 500. On the other hand, such program modules include, but are not limited to, routines, subroutines, programs, objects, components, data structures, etc. that perform particular tasks or execute particular abstract data types, described below, in accordance with the present invention.

The representative image selection system 500 according to an embodiment of the present invention may further include a communication unit (not shown) and a control unit (not shown).

The communication unit performs a function of enabling the representative image selection system 500 to communicate with an external device such as the user terminal device 600.

The controller performs a function of controlling the flow of data between the similarity measurer 510, the similarity comparator 520, the representative image selector 550, the image search index 570, and the communicator. That is, the control unit controls the flow of data from the outside or between each component of the image matching system 500, such that the similarity measurer 510, the similarity comparator 520, the representative image selector 550, and the image searcher. The index 570 and the communication unit each control to perform a unique function.

According to an embodiment of the present invention, the representative image stored in the image search index 570 may be usefully used in the visual search process by image matching. More specifically, the image matching system 700 including the representative image selection system 500 according to the present invention, when an image containing an object is input as a query, the image (or to a specific object) Feature information) and a reference image for any object stored in the reference image database (or feature information for any object), and the specific image for the object determined to be similar to the object with reference to the matching result. It may further include a query performing unit (not shown) that provides a as a search result for the query.

Embodiments according to the present invention described above may be implemented in the form of program instructions that may be executed by various computer components, and may be recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination. Program instructions recorded on the computer-readable recording medium may be those specially designed and configured for the present invention, or may be known and available to those skilled in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs, DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform the process according to the invention, and vice versa.

Although the present invention has been described by specific embodiments such as specific components and the like, but the embodiments and the drawings are provided to assist in a more general understanding of the present invention, the present invention is not limited to the above embodiments. For those skilled in the art, various modifications and variations can be made from these descriptions.

Accordingly, the spirit of the present invention should not be limited to the above-described embodiments, and all of the equivalents or equivalents of the claims, as well as the appended claims, fall within the scope of the spirit of the present invention. I will say.

400: communication network
500: representative image selection system
510: similarity measuring unit
520: similarity comparison unit
550: representative image selection unit
570: image search index
600: user terminal device
700: image matching system

Claims (27)

A method of selecting a representative image from a plurality of images,
(a) When at least one image or information on the image corresponds to an image node in the similarity measuring unit, image similarities between a plurality of image nodes are displayed on the image corresponding to each of the image nodes or on the image. Acquiring with reference to,
(b) comparing a critical similarity with each of the obtained image similarities in a similarity comparator, and
(c) If an image similarity greater than the threshold similarity exists among the image similarities in the representative image selector, image nodes corresponding to the image similarity larger than the threshold similarity are merged to form an image node, and the merged image Selecting a first specific image among the image corresponding to the node or the image indicated by the information about the image as the representative image,
In step (c),
In selecting the first specific image among the image corresponding to the merged image node or the information indicated by the image as the representative image, a specific image node among the nodes other than the merged image node and the merged image node Selecting an image having a small image similarity between respective pre-merge image nodes to which each image corresponding to the image corresponds before the merge. The method of claim 1,
(d) acquiring image similarities between the merged image nodes in the similarity measuring unit, and
(e) If there is an image similarity larger than the threshold similarity among the image similarities between the merged image nodes in the representative image selector, image nodes corresponding to the image similarity larger than the threshold similarity are merged again. Forming a merge node and selecting, as the representative image, a second specific image from among an image corresponding to the merged image node or an image indicated by the information about the image;
Method comprising a. The method of claim 1,
(d) acquiring image similarities between each image node other than the image node and the merged image node in the similarity measuring unit;
(e) an image corresponding to an image similarity larger than the threshold similarity when the image similarity greater than the threshold similarity among the image similarities between each of the another image node and the merged image node is present in the representative image selection unit; Merging the nodes again to form an image merging node, and selecting a second specific image as the representative image from an image corresponding to the merged image node again or an image indicated by the information on the image;
Method comprising a. The method of claim 3,
The step (d)
Among the other image nodes, k (k is a natural number less than or equal to the number of nodes other than the merged image node) and the image similarity between the merged image node is
And each image corresponding to the k-th image node and the merged image nodes is set to the largest image similarity among the image similarities between the corresponding image nodes before merging. delete delete The method of claim 1,
The step (a)
And calculating the image similarity based on the feature information of the object part included in the plurality of images. The method of claim 1,
The step (a)
The image similarity is obtained by referring to names of the plurality of images. The method of claim 1,
(d) storing the selected representative image in an image search index in an image search index
≪ / RTI > 10. The method of claim 9,
(e) When a predetermined image is input as a query in the query execution unit, the input image is matched with the representative images stored in the image search index, and among the representative images with reference to the matching result. Providing a representative image or information about the matched representative image as a search result for the query
≪ / RTI > The method of claim 1,
In step (c),
And when there is an image similarity larger than the threshold similarity among the image similarities, all image nodes corresponding to the image similarity larger than the threshold similarity are merged together to form an image node. The method of claim 1,
The threshold similarity varies adaptively. The method of claim 12,
Calculate the probability occupied by the number of merged image nodes among the total number of image nodes for a predetermined time so that the threshold similarity is automatically increased when the probability exceeds a predetermined first threshold, and the probability is predetermined The threshold similarity is automatically lowered if less than a second threshold. An apparatus for selecting a representative image from a plurality of images,
When at least one image or information on the image corresponds to an image node, a similarity measure of acquiring image similarities between a plurality of image nodes with reference to an image corresponding to each of the image nodes or information on the image part,
A similarity comparison unit that compares each of the obtained image similarities with a critical similarity, and
If there is an image similarity greater than the threshold similarity among the image similarities, image nodes corresponding to the image similarity greater than the critical similarity are merged to form an image node, and the image corresponding to the merged image node or the image Representative image selection unit for selecting a first specific image of the image indicated by the information as the representative image,
The representative image selection unit,
In selecting the first specific image among the image corresponding to the merged image node or the information indicated by the image as the representative image, a specific image node among the nodes other than the merged image node and the merged image node And selecting as the representative image an image having a small image similarity between respective pre-merge image nodes to which each image corresponding to the sub-merge has corresponded before the merge. The method of claim 14,
The similarity measuring unit,
Obtain image similarities between the merged image nodes,
The representative image selection unit,
If there is an image similarity greater than the threshold similarity among the image similarities between the merged image nodes, image nodes corresponding to the image similarity greater than the threshold similarity are merged again to form an image merging node, and again And selecting a second specific image as the representative image from an image corresponding to a merged image node or an image indicated by the information about the image. The method of claim 14,
The similarity measuring unit,
Obtain image similarities between each of the other image nodes other than the image node and the merged image node,
The representative image selection unit,
If there is an image similarity greater than the threshold similarity among the image similarities between each of the another image node and the merged image node, the image merge node is further merged by merging the image nodes corresponding to the image similarity larger than the threshold similarity again. And select a second specific image among the image corresponding to the merged image node or the image indicated by the information about the image as the representative image. The method of claim 16,
The similarity measuring unit,
Among the other image nodes, k (k is a natural number less than or equal to the number of nodes other than the merged image node) and image similarity between the merged image node and
And set the largest image similarity among the image similarities between the kth image node and each pre-merge image node corresponding to each image corresponding to the merged image nodes. delete delete The method of claim 14,
The similarity measuring unit,
And obtaining the image similarity based on feature information of an object part included in the plurality of images. The method of claim 14,
The similarity measuring unit,
The image similarity is obtained by referring to names of the plurality of images. The method of claim 14,
And an image search index for storing the selected representative image. The method of claim 22,
When a predetermined image is input as a query, the input image is matched with the representative images stored in the image search index and the matched representative image or the matched representative among the representative images with reference to the matching result. Query execution unit that provides information about the image as a search result for the query
Device further comprising. The method of claim 14,
The representative image selection unit,
And when there is an image similarity larger than the threshold similarity among the image similarities, all image nodes corresponding to the image similarity larger than the threshold similarity are merged together to form an image node. The method of claim 14,
The threshold similarity varies adaptively. 26. The method of claim 25,
Calculate the probability occupied by the number of merged image nodes among the total number of image nodes for a predetermined time so that the threshold similarity is automatically increased when the probability exceeds a predetermined first threshold, and the probability is predetermined The threshold similarity is automatically lowered if less than a second threshold. A computer-readable recording medium having recorded thereon a computer program for executing the method according to any one of claims 1 to 4 and 7.

Images