JP2002007469A - Retrieving method, retrieving device and image retrieving device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image retrieving device suitable for visually grasping the relation between images though the images approximate each other only in the values of the specific item of feature information in the case of performing the image retrieval of the images. SOLUTION: Plural retrieving images are arranged and displayed in a virtual space for displaying in accordance with the value (mix) of the item of feature information corresponding to them. Then, a user inputs retrieving images as retrieving key images from the retrieving images arranged in the virtual space for displaying and retrieves the retrieving images suited to the image common with the retrieving key images from the retrieving images of an image storage means based on each feature information corresponding to the plural inputted different retrieving key images. As a result, when a desired image is not obtained, specific items whose values are the same with or approximate each other among the items of each feature information corresponding to the plural inputted different retrieving key images are assigned to the axis of the virtual space for displaying.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a retrieval method which is based on a plurality of different retrieval key images given by a user as retrieval keys and which is adapted from a plurality of retrieval images to an image common to the retrieval key images. BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for retrieving an image, and particularly, when retrieving an image of an image, visually ascertains the relevance of images having only similar characteristic item values. The present invention relates to a search method, a search device, and an image search device that are suitable for:

[0002]

2. Description of the Related Art A new modeling based on a product concept has been demanded to build an image database by accumulating a large number of images such as paintings and photographs in a computer, and to be able to quickly extract an image oneself has imaged as necessary. Do a lot of industrial designers and planners have. In this case, from among a plurality of images to be searched, a plurality of images similar to the image imaged by the user are specified, and a search image suitable for an image common to the specified images can be narrowed down to obtain a desired image. It is relatively easy to find and very convenient.

Conventionally, in order to respond to such a user's request, for example, based on a plurality of different search key images given as search keys by the user, a plurality of search images are used in common with the search key images. There has been an image search apparatus that searches for a search image suitable for an image. This includes an image database storing a plurality of search images in association with feature information in which the features of the images are extracted as feature amounts for a plurality of items, an evaluation input unit for inputting an evaluation of the image database for the search image, An image for searching a search image in an image database for a search image that matches an image common to the evaluation target images based on the evaluation information input in the evaluation input unit and the respective characteristic information corresponding to the search images targeted for evaluation. It consists of a search section.

Here, the characteristic information includes, for example, a color distribution,
The feature of the image is extracted as a feature amount for each of a plurality of items such as a grayscale distribution or a frequency distribution. First, assuming that the number of items of feature information to be extracted is N, the i-th feature information Fi of the image database can be expressed by the following equation (1). The feature information Fi is obtained by normalizing the feature amounts extracted from the search image data. This is because it is necessary to normalize the value of each item in order to treat the value of each item of the feature information with the same weight as the value of the other items in the search.

[0005]

(Equation 1)

[0006]

(Equation 2)

Here, the value min of the n-th item among the items of the characteristic information will be considered. First, the feature amount extracted from the search image data is expressed as shown in the above equation (2), and when the total number of search images in the image database 44 is Nd, the average value mdn and the variance σ are obtained for the nth item of the feature information.
dn ² can be calculated by the following equations (3) and (4), respectively. Then, the value min of the n-th item of the feature information
Can be calculated by subtracting the average value mdn from the feature value Pin of the n-th item by the following equation (5), and dividing the subtraction result by the standard deviation σdn.

[0008]

(Equation 3)

[0009]

(Equation 4)

[0010]

(Equation 5)

Next, an evaluation V is input for some of the search images in the image database. The input of the evaluation V is performed by the user inputting a high degree of relevance to some search images. For example, when the user wants to search for an image of “car”, as shown in FIG.
Of the image of “mountain and lake”, the image of “animal” and the image of “truck”, the user sets the degree of relevance to the image of “passenger car” to 80%, 0% for images of "and lakes", 0% for images of "animals", and 0% for images of "tracks" The height is set to 60%.

The evaluation V can be expressed by the following equation (6), where the degree of relevance input to the i-th search image is an evaluation value vi. For example, if the user selects the second, seventeenth, and 53
The high relevance of the 72nd, 72nd, and 103rd search images is 30%, 80%, 40%, 98%, 2
When 0% is set, v2, v17, v5 of the evaluation V
3, v72 and v103 are 30, 80, 40, 98, 2 respectively
0 and all other values are set to 0.

[0013]

(Equation 6)

Next, based on the input evaluation V and the respective characteristic information corresponding to the evaluation target image, a reference position K serving as a reference for a search in a search virtual space in which each item of the characteristic information is assigned to a different axis. Is calculated. Specifically, the reference position K can be expressed as a vector including the same number of values kn as the number N of items of feature information by the following equation (7), and the characteristic of the evaluation target image is calculated by the following equation (8). Sum the value min of each item of information and the evaluation value vi,
It can be calculated by dividing the sum of products by the sum of the evaluation values vi.

[0015]

(Equation 7)

[0016]

(Equation 8)

Next, among the items of the feature information corresponding to the image to be evaluated, an item having the same or similar value is determined to be a highly relevant specific item, and the value mix of the specific item (hereinafter, mix is referred to as mix) , Indicates the value of a specific item.) A predetermined weighting operation is performed on the value min of each item of the search image in the image database so that the degree of contribution of the specific item to the search increases. Specifically, the projection vector W is calculated, and the projection vector W is calculated for the value min of each item of the search image.
Weighting operation is performed. The projection vector W can be expressed as a vector including the number wn of the same number as the number N of items of feature information by the following equation (9). The difference between the value min and the value kn at the reference position is calculated, the calculated difference is multiplied by the evaluation value vi, and the sum of the evaluation values vi is divided by the sum of the products. Can be.

[0018]

(Equation 9)

[0019]

(Equation 10)

Next, each search image is arranged in the search virtual space based on the result of the weighting operation using the projection value vector W, and the distance D from the reference position K in the search virtual space is calculated. The distance D is calculated by the following equation (11) as the square root of the sum of the result of subtracting the value kn of the reference position from the value min of each item of feature information and the value of the projection vector wn. can do.

[0021]

[Equation 11]

Next, the search results are searched in order from the search image at the position where the calculated distance D is small. When the desired image is found, the image search for the image is terminated, but the desired image is not found. At this time, an evaluation for the search result is input, and the above operation is repeated.

[0023]

However, in the conventional image retrieval apparatus, the value of one of the characteristic information items is very similar, but the values of the other items are completely different. Images having similar values may have high relevance for items with similar values, but may be determined to have low relevance overall, and may not be output as a search result. For example, when searching for an image of "car", among the search images, for an image including an image of "car", the values of specific items are very similar, but they are other than "car" such as a background or a person. When the values of other items are completely different,
Despite including the image of "car", it may not be output as a search result. In this case, the image of “car” cannot be efficiently searched.

Therefore, in the conventional image retrieval apparatus, images in which the values of the items of the characteristic information are generally similar are output as retrieval results, so that they can be found relatively easily. However, it is difficult to search for images in which only the values of the specific items of the feature information are similar, because they may not be output as search results.

Therefore, the present invention has been made in view of such unresolved problems of the prior art, and when performing an image search of an image, only the value of a specific item of characteristic information is obtained. It is an object of the present invention to provide a search method, a search device, and an image search device suitable for visually recognizing the relevance of similar images.

[0026]

In order to achieve the above object, a search method according to claim 1 of the present invention is based on feature information obtained by extracting a feature to be searched for each of a plurality of items. In a method of searching for a search target that matches feature information of a search key given as a search key from among a plurality of search targets, the plurality of search targets are stored in target storage means, and the search key A search key input step of inputting a plurality of different search keys given as a search key, and a search target adapted to feature information common to the search keys based on respective feature information corresponding to the search key input in the search key input step A target search step of searching for an object from search targets in the target storage unit; and the feature information based on a value of feature information common to each of the search keys. And a target display step of arranging and displaying on the display for a virtual space according to the feature information of the search target object searched by said target search step assigns items to the axis of the display virtual space.

Here, the display virtual space may be in any form as long as it is a space recognizable by the user, and can be represented by a coordinate system having one or more axes, for example. In this case, some or all of the items of the characteristic information are assigned to each axis. Hereinafter, Claim 3
The same applies to the retrieval device described above and the image retrieval device according to claims 4, 6 or 7.

The search targets include, for example, images, documents, music, moving images and other search targets. That is, when performing image search for such an object, the application of the present invention is preferable. Hereinafter, the same applies to the search device according to claim 3. Further, the feature information of the search target may be calculated in advance, may be calculated when the search target is arranged in the display virtual space, or may be arranged in the display virtual space. The time and the time of the search may be respectively calculated. The same applies to the feature information to be searched. Here, from the viewpoint of speeding up the search process, it is preferable to calculate the feature information in advance, and from the viewpoint of reducing the data volume, the feature information is not calculated in advance but is calculated at the time of arrangement or search. Is preferred. Hereinafter, the same applies to the search device according to claim 3.

The search key object may be input from among the search objects in the object storage means, or may be an object different from the search object in the object storage means. In the former case,
For example, when input is made from among the search targets arranged in the display virtual space, the input is easy. Hereinafter, Claim 3
The same applies to the described search device. The search method according to claim 2 according to the present invention, in the search method according to claim 1, further comprising: an axis reassigning step of reassigning the item of the feature information assigned to the axis of the display virtual space. And

[0030] On the other hand, in order to achieve the above object, a search device according to claim 3 of the present invention, based on feature information obtained by extracting a feature to be searched for each of a plurality of items,
In an apparatus for searching for a search target that matches feature information of a search key given as a search key from among a plurality of search targets, a target storage unit storing the plurality of search targets and the search key is provided. A search key input unit for inputting a plurality of different search keys; and a search target adapted to feature information common to the search keys based on respective feature information corresponding to the search key input by the search key input unit. A target search means for searching among search targets in a target storage means, and assigning an item of the characteristic information to an axis of a display virtual space based on a value of characteristic information common to each of the search keys; Target display means for arranging and displaying in the display virtual space in accordance with the characteristic information of the search target searched by the above.

With such a configuration, when performing an image search for a search target, the user inputs a plurality of different search targets similar to the search target to be imaged as search keys from the search key input means, and the target search is performed. By means,
Based on each piece of feature information corresponding to the input search key, a search target that matches feature information common to the search keys is searched from among the search targets in the target storage unit. Then, based on the value of the characteristic information common to each of the search keys, the item of the characteristic information is assigned to the axis of the display virtual space by the target display means, and the characteristic of the search target object searched by the target search means. It is arranged and displayed on the display virtual space according to the information.

On the other hand, in order to achieve the above object, an image retrieval apparatus according to a fourth aspect of the present invention provides an image retrieval apparatus based on feature information obtained by extracting features of an image for each of a plurality of items.
In an apparatus for searching for a search image that matches feature information of a search key image given as a search key from among a plurality of search target images, an image storage storing the plurality of search target images in association with the feature information Means, a search key image input means for inputting a plurality of different images as the search key, and a common search key image based on respective characteristic information corresponding to the search key image input by the search key image input means. An image search means for searching a search image in the image storage means for a search image suitable for the feature information; and a virtual space for displaying the items of the feature information based on feature information common to each of the search key images. And arranged on the display virtual space according to the feature information of the search image searched by the image search means. And an image display means for.

With such a configuration, when performing an image search for an image, the user inputs a plurality of different images similar to the image to be imaged from the search key image input means as search keys, and the image search means Based on the respective pieces of feature information corresponding to the input search key images, a search image matching the feature information common to the search key images is searched from the search target images in the image storage unit. The image display means assigns the item of the characteristic information to the axis of the display virtual space based on the characteristic information common to each of the search key images, and the characteristic information of the search image searched by the image search means. Is displayed in the display virtual space in accordance with.

According to a fifth aspect of the present invention, in the image retrieval apparatus according to the fourth aspect, further, the item of the characteristic information assigned to the axis of the display virtual space is reassigned. Axis reassignment means is provided. With such a configuration, when a desired image is found as a result of the search by the image search means, the image search for the image is terminated, but when the desired image is not found, the axis search is performed. The item of the characteristic information assigned to the axis of the display virtual space is reassigned by the assigning means.

Further, according to a sixth aspect of the present invention, there is provided the image retrieval apparatus according to the first aspect, wherein the characteristic of the retrieval key image given as a retrieval key is determined based on characteristic information obtained by extracting the characteristic of the image for each of a plurality of items. An apparatus for searching for a search image suitable for information from among a plurality of search target images, an image storage unit storing the plurality of search target images in association with the feature information, and displaying the feature information items. An image display unit that allocates the plurality of search target images to the axis of the virtual space and arranges and displays the plurality of search target images on the display virtual space according to the value of the item of the feature information corresponding to the plurality of search target images; Image input means for inputting an image as the search key image from the arranged search target images, based on respective characteristic information corresponding to the search key image input by the image input means; Image search means for searching a search target image in the image storage means for a search image suitable for characteristic information common to the search key images of the plurality of search key images, wherein the image display means corresponds to the plurality of search key images. The item of the characteristic information is reassigned to the axis of the display virtual space based on the value of each characteristic information.

With such a configuration, when performing an image search for an image, first, the item of the characteristic information is assigned to the axis of the display virtual space by the image display means, and a plurality of search target images are assigned to the axis. It is arranged and displayed on the display virtual space according to the value of the item of the corresponding feature information. Here, when the user inputs a search target image similar to the image to be imaged from the image input unit as a search key image from among the search target images arranged in the display virtual space, the input is performed by the image search unit. Based on the respective feature information corresponding to the plurality of different search key images, a search image matching the feature information common to the search key images is searched from the search target images of the image storage unit.

As a result, when the desired image has been found, the image search for the image is terminated. When the desired image has not been found, a plurality of different search data inputted by the image display means are displayed. The feature information items are reassigned to the axes of the display virtual space based on the values of the respective feature information items corresponding to the key images. Further, the image search device according to claim 7 of the present invention matches feature information of a search key image given as a search key based on feature information obtained by extracting features of an image for each of a plurality of items. An image storage unit that stores the plurality of search target images in association with the feature information, and a number of dimensions smaller than the number of items of the feature information. All or some of the items of the characteristic information are assigned to the axis of the display virtual space, and the plurality of images are arranged and displayed on the display virtual space according to the values of the corresponding items of the characteristic information. Image display means,
Image input means for inputting an image as the search key image from among the search target images arranged on the display virtual space, based on respective characteristic information corresponding to the search key image input by the image input means. Image search means for searching a search image suitable for the characteristic information common to the search key image from among the search target images in the image storage means, wherein the image search means corresponds to the plurality of search key images. Based on each feature information, a reference position serving as a search reference is calculated in a search virtual space in which each item of the feature information is assigned to a different axis, and each feature corresponding to the plurality of search key images is calculated. The search target of the image storage means is set such that the value of a specific item having the same or similar value among the information items increases the degree of contribution to the search. Performing a predetermined weighting operation on the value of the item of the image, arranging each of the search images on the search virtual space based on the result of the weight calculation, and comparing the calculated reference position with the calculated reference position in the search virtual space. The search image is searched in order from the search image at the position where the distance is small,
The image display means is configured to reassign the item of the characteristic information to the axis of the display virtual space based on the value of each item of the characteristic information corresponding to the plurality of search key images.

With this configuration, when performing an image search for an image, first, a part or all of the items of the characteristic information are assigned to the axis of the display virtual space by the image display means, and a plurality of search operations are performed. The target image is arranged and displayed on the display virtual space according to the value of the item of the corresponding feature information. Here, when the user inputs a search target image similar to the image to be imaged from the image input unit as a search key image from among the search target images arranged in the display virtual space, the input is performed by the image search unit. A reference position is calculated in the search virtual space based on the respective pieces of feature information corresponding to the plurality of different search key images. Then, among the items of the characteristic information corresponding to the plurality of search key images, the value of the specific item having the same or similar value contributes to the search so that the degree of the item of the search image of the image storage means is increased. A predetermined weighting operation is performed on the value, each search image is arranged in the search virtual space based on the result of the weight calculation, and the search image having a small distance from the calculated reference position in the search virtual space. Are retrieved in order from

As a result, when the desired image is found, the image search for the image is terminated. When the desired image is not found, a plurality of different search data inputted by the image display means is displayed. The feature information items are reassigned to the axes of the display virtual space based on the values of the respective feature information items corresponding to the key images. Here, the search virtual space may be in any form as long as each item of feature information is assigned to a different axis, and may be, for example, a one-dimensional space, a two-dimensional space, There is a three-dimensional space. Note that the search virtual space is a space formed for performing a search, and is not displayed unlike the display virtual space.

Further, according to an image search device of the present invention, in the image search device of any one of claims 4 to 7, the image display means corresponds to the plurality of search key images. A specific item having the same or similar value among the items of the characteristic information is reassigned to the axis of the display virtual space. With such a configuration, when a desired image is not found as a result of the search, the image display means sets a value among the items of the characteristic information corresponding to the plurality of different search key images input. Are assigned to the axes of the display virtual space.

Further, according to a ninth aspect of the present invention, in the image retrieval apparatus according to the eighth aspect, the image display means reproduces a plurality of items on one axis of the display virtual space. When assigning, the number of assigned items of the axis to which the specific item is assigned is made smaller than the number of assigned items of other axes. With such a configuration, when assigning a plurality of items to one axis of the display virtual space,
The image display means is re-assigned so that the number of assigned items of the axis to which the specific item is assigned is smaller than the number of assigned items of the other axes.

Further, according to a tenth aspect of the present invention, in the image retrieval apparatus according to the fourth aspect, the weight of the reference position or the predetermined weighting operation can be changed. Has become. With such a configuration, the user determines the weight of the reference position or the predetermined weighting calculation so that the features of the image to be imaged appear most, while observing the arrangement status of the display virtual space during the search. If changed, images having similar values of the specific item are arranged close to each other in the display virtual space. Therefore, since the feature of the image that the user images in the display virtual space can be visually distinguished, a desired image can be found relatively easily.

In the above, a search method, a search device, and an image search device for achieving the above object have been proposed. However, the present invention is not limited to this, and the following storage media may be proposed for achieving the above object. it can. This storage medium, based on the feature information obtained by extracting the features of the image for each of a plurality of items, a search image that matches the feature information of the search key image given as a search key, A storage medium storing an image search program for searching from among the plurality of search target images, the image storage means storing the plurality of search target images in association with the feature information, and a search target image arranged in the display virtual space. A plurality of image information input means for inputting an image as the search key image from among the plurality of search target images by assigning the item of the characteristic information to an axis of a display virtual space. Image display means for arranging and displaying in the display virtual space according to the value of the item, and respective features corresponding to the search key image input by the image input means A program for executing a process realized by an image search means for searching a search target image in the image storage means for a search image suitable for the characteristic information common to the search key image based on the information. A computer-readable medium configured to reassign the item of the feature information to an axis of the display virtual space based on a value of each feature information corresponding to the plurality of search key images. It is a possible storage medium.

With such a configuration, the program stored in the storage medium is read by the computer,
When the computer executes the program in accordance with the read program, the same operations and effects as those of the image search device according to claim 6 can be obtained.

[0045]

Embodiments of the present invention will be described below with reference to the drawings. 1 to 12 are diagrams showing a search method, a search device, and an image search device according to an embodiment of the present invention. In the present embodiment, a search method, a search device, and an image search device according to the present invention are implemented by a computer 100 based on a plurality of different search key images given as search keys by a user, as shown in FIG. This is applied to a case where a search image suitable for an image common to the search key images is searched from a plurality of search images.

First, the computer 10 to which the present invention is applied
0 will be described with reference to FIG. FIG. 1 is a block diagram showing a configuration of a computer 100 to which the present invention is applied. The computer 100, as shown in FIG.
A CPU 30 for controlling calculations and the entire system based on a control program;
ROM 32 storing the control program of the
It comprises a RAM 34 for storing data read from the M32 or the like and a calculation result required in a calculation process of the CPU 30, and an I / F 38 for mediating input / output of data to and from an external device. Are connected to each other via a bus 39 which is a signal line for transferring data.

The I / F 38 includes, as external devices, an input device 40 such as a keyboard and a mouse capable of inputting data as a human interface, a display device 42 for displaying a screen based on image signals, and a search target. Database 4 storing a plurality of search image data
4 are connected. Next, the data structure of the image database 44 will be described in detail with reference to FIGS. FIG. 2 is a diagram showing search image data, and FIG.
FIG. 4 is a diagram showing a data structure of a feature information registration table 70, and FIG. 4 is a diagram showing a data structure of an image correspondence table 75.

As shown in FIGS. 2 to 4, a plurality of search image data jpg01 to jpg09 are stored in the image database 44.
And a feature information registration table 70 in which feature information obtained by extracting features of the image for each of the plurality of search image data jpg01 to jpg09 is registered.
Image correspondence table 75 that associates pg09 with each feature information
And are stored.

As shown in FIG. 3, one record is registered in the characteristic information registration table 70 for each characteristic information. Each record includes a field 74 for registering feature information and a field 72 for registering a corresponding number for specifying the feature information of the field 74. The feature information is obtained by extracting the features of the image as a feature amount for each of a plurality of items such as a color distribution, a grayscale distribution, and a frequency distribution, and is executed by the CPU 30 when the image is stored in the image database 44. It is generated by characteristic information generation processing. The contents of the feature information will be described later in detail.

In the example shown in FIG. 3, in the first record in the figure, "(1)" is assigned as the corresponding number in the field 72, and "F1 = (m11, m12,
m13, ...) "is registered. In this field 74, feature information of the search image data jpg01 is registered. In the record in the fifth row in the figure, “(5)” is registered as the corresponding number in the field 72 and “F5 = (m51, m52, m53,...)” As the characteristic information in the field 74. In this field 74, search image data jpg05
The feature information of is registered.

In the image correspondence table 75, as shown in FIG. 4, one record is registered for each of the search image data jpg01 to jpg09. Each record is configured to include a field 76 for registering the file name of the search image data and a field 78 for registering the corresponding number of the feature information corresponding to the search image data in the field 76. In the example of FIG. 4, the first record in the figure includes the field 7
6, the file name “jpg01” of the search image data jpg01 is registered as the file name, and “(1)” is registered as the corresponding number in the field 78. This is the search image data
jpg01 is identified by the feature number “F1” specified by the corresponding number (1).
= (M11, m12, m13, ...) ". In the fifth record in the figure, the field 7
6, the file name "jpg05" of the search image data jpg05 is registered as the file name, and "(5)" is registered as the corresponding number in the field 78. This is the search image data
jpg05 is identified by the corresponding number (5).
= (M51, m52, m53,...) ".

Next, the configuration of the CPU 30 and the CPU 30
Will be described with reference to FIGS. 5 to 12. The CPU 30 includes a microprocessing unit MPU and the like, activates a predetermined program stored in a predetermined area of the ROM 32, and executes the characteristic information generation processing and the image search processing illustrated in the flowcharts of FIGS. It is supposed to.

First, the characteristic information generation processing will be described in detail with reference to FIG. FIG. 5 is a flowchart illustrating the feature information generation processing. The feature information generation process is a process of generating feature information of an image when storing the image in the image database 44. When the feature information generation process is executed by the CPU 30, the process first proceeds to step S80 as shown in FIG. It is supposed to.

In step S80, the image database 4
4, feature information is extracted as a feature amount for each of a plurality of items from the search image data, the process proceeds to step S82, the feature information is normalized, and the process proceeds to step S84, where the normalized feature information is stored in the feature information registration table. Is registered in the corresponding record, and a series of processing ends. Step S80, S8
In 2, if the number of items of feature information to be extracted is N, the i-th feature information Fi can be expressed by the above equation (1). The feature information Fi is obtained by normalizing the feature amounts extracted from the search image data. This is because it is necessary to normalize the value of each item in order to treat the value of each item of the feature information with the same weight as the value of the other items in the search.

Here, the value min of the n-th item among the items of the characteristic information will be considered. First, the feature amount extracted from the search image data is expressed as shown in the above equation (2), and when the total number of search images in the image database 44 is Nd, the average value mdn and the variance σ are obtained for the nth item of the feature information.
dn ² can be calculated by the above equations (3) and (4), respectively. Then, the value min of the n-th item of the feature information
Can be calculated by dividing the average value mdn from the feature quantity Pin of the n-th item by the above equation (5), and dividing the result of the division by the standard deviation σdn.

Next, the image retrieval processing will be described in detail with reference to FIG. FIG. 6 is a flowchart showing the image search processing. The image search process is a process of searching for a search image matching an image common to the search key images based on a plurality of different search key images given as search keys. As shown in FIG. 6, first, the process proceeds to step S102.

In step S102, a plurality of search images (for example, some of the search images in the image database 44) are displayed on the display device 42 based on the search image data in the image database 44, and among the displayed search images, Is input from the input device 40 as a search key image, and the process proceeds to step S108. In step S108, the evaluation V for the input plurality of search key images is input. Specifically, in step S108, the user inputs a high degree of relevance to the plurality of input search key images. For example, when the user wants to search for an image of “car”, as shown in FIG. 13, the image of “passenger car”, the image of “mountain and lake”, the image of “animal”, and the image of “truck” are searched. If the user inputs the key image as a key image, the user moves the slide bar of which the degree of relevance can be adjusted in the left-right direction with a mouse or the like, and increases the relevance to 80% for the image of “passenger car”. , The height of the relevance is 0% for the image of “mountain and lake”, the height of the relevance is 0% for the image of “animal”, and the image of “truck” is This is performed in such a manner that the degree of relevancy is set to 60%.

The evaluation V can be expressed by the above equation (6), where the degree of relevance inputted to the i-th search image is an evaluation value vi. For example, if the user searches the second, seventeenth,
The high degrees of relevance of the 53rd, 72nd, and 103rd search images are 30%, 80%, 40%, and 98%, respectively.
% And 20%, v2, v1 of the evaluation V
7, v53, v72, and v103 are 30, 80, 40, and 9, respectively.
8, 20 and all other values are set to zero.

Next, the process proceeds to step S110, and based on the respective pieces of feature information corresponding to the plurality of input search key images, items of the feature information are assigned to different axes in a search virtual space. A reference position K serving as a reference is calculated. Note that the search virtual space is a space formed for performing a search, and is not displayed on the display device 42 unlike the display virtual space.
Specifically, in step S108, the reference position K can be expressed as a vector including the same number of values kn as the number N of feature information items by the above equation (7), and is, for example, arranged in the display virtual space. When the user inputs two images as search key images from among the searched images, as shown in FIG. 7, the value min and the evaluation value vi of each item of the feature information of the search key image are calculated by the above equation (8). The product sum can be calculated by dividing the product sum by the sum of the evaluation values vi. In this case, since the evaluation value vi is 0 for other search images for which the evaluation V has not been input, only the characteristic information of the two input search key images is substantially reflected in the reference position value kn. . Similarly, when three or more search key images are input, the value min of each item of the feature information is weighted and averaged for the search key images, and the average value is set as the reference position K.

Then, the process proceeds to step S112, and among the items of the characteristic information corresponding to the plurality of input search key images, an item having the same or similar value is determined to be a highly relevant specific item. A predetermined weighting operation is performed on the value min of each item of the search image in the image database 44 so that the value mix of the specific item contributes to the search increases. Specifically, step S112
Then, a projection vector W is calculated, and a weighting operation using the projection vector W is performed on the value min of each item of the search image. The projection vector W can be represented as a vector including the same number of values wn as the number N of feature information items by the above equation (9). For example, the user can select 2 from the images arranged in the display virtual space. When the two images are input as the search key images, as shown in FIG. 8, the difference between the value min of each item of the feature information of the search key image and the value kn of the reference position is calculated by the above equation (10). It can be calculated by multiplying and summing the difference obtained and the evaluation value vi, and dividing the sum of the evaluation values vi by the sum of the products. In this case, since the evaluation value vi is 0 for other search images for which the evaluation V has not been input, the projection vector value wn includes:
Only the characteristic information of the two input search key images is substantially reflected. Similarly, when three or more search key images are input, the difference between the value min of each item of the characteristic information and the value kn of the reference position is weighted and averaged for the search key images, and the average value is calculated. Let it be a projection vector W.

Next, the process proceeds to step S114, where each search image is arranged in the search virtual space based on the result of the weighting operation using the projection value vector W, and the distance D from the reference position K in the search virtual space is calculated. I do. The distance D is calculated as the square root of the sum of the square of the subtraction result and the value wn of the projection vector, by subtracting the value kn of the reference position from the value min of each item of the feature information by the above equation (11). can do. As described above, when the respective search images are arranged in the search virtual space based on the result of the weighting calculation, for example, as shown in FIG.
It can be seen that the relevance is higher than that of M3, M5, and M6, so that search images with high relevance can be brought close to each other, and search images that are not so can be separated from each other.

Next, the process proceeds to step S116, where a predetermined number of search images are searched out as search results in order from the search image at the position where the calculated distance D is small, as shown in FIG. Is assigned to the axis of the display virtual space based on the value min of each item of feature information corresponding to
It is arranged in the display virtual space according to the value min of each item of the corresponding feature information and displayed on the display device 42. Specifically, in step S116, a specific item having the same or similar value among the items of the characteristic information corresponding to the plurality of search key images is assigned to the axis of the display virtual space, When a plurality of items are assigned to one axis, the number of assigned items of an axis to which a specific item is assigned is set to be smaller than the number of assigned items of another axis.

For example, items M4, M1, M6,
Assuming that the value wi of the projection vector calculated for the search key image is small in the order of M3, M2, and M5, first, the item M4
Is selected as a specific item. Next, as shown in FIG. 11, on the x-axis of the display virtual space, the value wi of the projection vector is given.
Of M5, M2, M3, M6 selected in descending order of
M4 and M4 are assigned to the y-axis of the display virtual space in the descending order of the projection vector value wi.
Assign the two items of 1. Accordingly, the search images having high relevance are arranged close to each other in the display virtual space, so that the characteristics of the search image visually stand out.

Then, the search image is arranged in the display virtual space as follows. For example, the x-axis of the display virtual space is M
When four items of 6, M3, M2 and M5 are assigned,
The coordinate Dsi on the x-axis of the ith search image is given by the following equation (12)
By subtracting the value kn of the reference position from the value min of the items M6, M3, M2, and M5 of the characteristic information, the result is squared, and the square root of the sum of the squared value and the value wi of the projection vector, It can be calculated by multiplying Psi. The sign Psi is given by the following equations (13) and (14).
For the items M6, M3, M2, and M5, the reference position values k6, k3, and k3 are obtained from the integration result of the vector composed of the value min of the item and the vector composed of the reference position values k6, k3, k2, and k5.
The coefficient H is calculated by subtracting the square of the magnitude of the vector consisting of k2 and k5, taking +1 when the value of the coefficient H is 0 or more, and-when the value of the coefficient H is negative. Take one.

[0065]

(Equation 12)

[0066]

(Equation 13)

[0067]

[Equation 14]

Then, the process shifts to step S118 to determine whether or not the desired image has been found by the input from the input device 40. When it is determined that the desired image has been found (Yes), A series of processing is ended. If it is determined that the processing is not performed (No), the process proceeds to step S122. In step S122, it is determined whether the magnification of the display virtual space has been changed. If it is determined that the magnification of the display virtual space has been changed (Yes), the process proceeds to step S124,
The projection vector W is changed according to the input from the user, and the process proceeds to step S114. The change of the projection vector W is
As shown in FIG. 12, while viewing the display virtual space, the user sets a slide bar whose change amount can be adjusted in the horizontal direction on the x-axis and a slide bar whose change amount can be adjusted in the vertical direction on the y-axis. The projection vector W is changed while moving with a mouse or the like. Here, assuming that the change amount of the projection vector value wn is Δw, the changed projection vector value wn ′ is calculated by the following equation (15).
The value w of the projected vector is obtained by factoring 2 with w as the exponent.
It can be calculated by multiplying n.

[0069]

(Equation 15)

On the other hand, if it is determined in step S122 that the magnification of the display virtual space is not to be changed (No), the flow shifts to step S126 to determine whether or not to shift the display virtual space. When it is determined that the virtual space is shifted (Y
es), the process proceeds to step S128, the reference position K is changed according to the input from the user, and the process proceeds to step S114. As shown in FIG. 12, the user changes the reference position K by using a slide bar whose change amount can be adjusted in the left-right direction for the x-axis and changing the change amount in the vertical direction for the y-axis while viewing the display virtual space. The reference position K is changed while moving the slide bar which can be adjusted by a mouse or the like. Here, assuming that the change amount of the reference position value kn is Δk, the changed reference position value kn ′ is given by the following equation (16).
Thus, it can be calculated by adding the change amount Δk to the value kn of the reference position. As a result, the coordinate Dsi on the x-axis of the i-th search image can be calculated by the following equation (17) in a manner similar to the above equation (12).

[0071]

(Equation 16)

[0072]

[Equation 17]

On the other hand, if it is determined in step S126 that the display virtual space is not shifted (No), the process proceeds to step S1.
Then, it is determined whether or not to change the configuration of the axis of the display virtual space. If it is determined that the configuration of the axis of the display virtual space is to be changed (Yes), the process proceeds to step S132. Then, the configuration of the axis of the display virtual space is changed according to the input from the user, and the process proceeds to step S112.

On the other hand, when it is determined in step S130 that the configuration of the axis of the display virtual space is not to be changed (No), the flow shifts to step S134 to input a new key image from the displayed image. It is determined whether or not a new key image is to be input (Yes).
The flow shifts to 36, where a key image is newly input from the displayed images, and the flow shifts to step S108.

On the other hand, if it is determined in step S134 that a new key image is not to be input (No), the process proceeds to step S1.
Move to 02. Next, the operation of the above embodiment will be described. When performing an image search for an image, first, after step S102, a plurality of search images are displayed based on the search image data in the image database 44.

Here, from among the displayed search images, the user inputs an image similar to the image to be imaged from the input device 40 as a search key image, and inputs evaluations for those search key images. S1
12, the reference position K is calculated based on the respective pieces of feature information corresponding to the plurality of input different search key images, and the values of the items of the feature information corresponding to the plurality of search key images are the same or different. The value m of the specific item to be approximated
A weighting operation using the projection vector W is performed on the value min of each item of the search image in the image database 44 so that the degree of ix contributing to the search increases.

Next, through step S114, each search image is arranged in the search virtual space based on the weighted calculation result, and the distance D from the reference position K in the search virtual space is calculated. Then, after step S116, the items of the feature information are assigned to the axes of the display virtual space based on the values min of the items of the feature information corresponding to the input plurality of search key images, and are searched out. The search image is arranged and displayed on the display virtual space according to the value min of each item of the feature information corresponding to the search image.

As a result, when the desired image is found, the image search for the image is terminated. However, when the desired image is not found, through steps S122 to S136, the user enters the projection vector. While changing the configuration of W, the reference position K, and the axis of the display virtual space, or inputting a new search key image from among the search images arranged in the display virtual space, the image that the user self-images. To narrow down.

Therefore, if the user successively inputs search images similar to the image to be imaged, a search image highly relevant to the image to be imaged becomes a virtual display image every time the search is performed. It is expected to be placed close to any axial direction in space. in this way,
In addition to assisting the search operation by assigning the axes of the display virtual space, the user must change the axes of the display virtual space independently so that the characteristics of the image to be imaged in the display virtual space stand out. Can also.

That is, the user changes the reference position K and the projection vector W so that the characteristics of the image to be imaged appear most while watching the arrangement state of the display virtual space during the search. Images having similar item values mix are arranged close to each other in the display virtual space. Therefore, since the feature of the image that the user images in the display virtual space can be visually distinguished, a desired image can be found relatively easily.

As described above, in the present embodiment, among the items of the characteristic information corresponding to a plurality of different search key images input as search keys, the specific items having the same or similar values are displayed in the display virtual space. Is assigned to the axis. Thereby, the value min of each item of the characteristic information
In addition to the fact that the images that are approximately similar to each other are arranged close to each other in the display virtual space, the images in which only the value mix of the specific item of the characteristic information is close to each other are also close to each other in the display virtual space. Placed. Therefore, even if the images are similar only in the value “mix” of the specific item of the feature information, it is easier to visually grasp the relation between the images than in the related art.

Further, in this embodiment, when a plurality of items are assigned to one axis of the display virtual space, the number of assigned items of an axis to which a specific item is assigned is smaller than the number of assigned items of another axis. I made it. Accordingly, images in which only the value of the specific item of the characteristic information is approximated are arranged closer to each other in the display virtual space, and images in which only the value of the specific item of the characteristic information is approximated are mixed. Even so,
It becomes even easier to visually grasp their relevance.

Further, in this embodiment, the user can change the reference position K and the projection vector W.
Accordingly, the user can change the reference position K and the projection vector W so that the features of the image to be imaged appear most while watching the arrangement state of the display virtual space while proceeding with the search. Since the features of the image that the user imagines in the space can be visually distinguished, a desired image can be found relatively easily.

Further, in the above-described embodiment, a reference position K serving as a search reference in a search virtual space is calculated based on each feature information corresponding to a plurality of different search key images input as search keys, Among the items of the feature information corresponding to the plurality of search key images, the value mix of the specific item having the same value or an approximate value of each item of the search image in the image database 44 is increased so that the degree of contribution to the search is increased. A weighting operation is performed on the value min with the projection vector W, and each search image is arranged in the search virtual space based on the result of the weight calculation, and a distance D from the reference position K in the search virtual space is calculated. In the search virtual space, images are searched in order from the image at a position where the distance D from the reference position K is small.

As a result, not only are the images in which the value min of each item of the characteristic information is comprehensively close to each other arranged in the search virtual space but also only the value mix of the specific item of the characteristic information Are similar to each other in the search virtual space. Therefore, it is possible to relatively easily search for images in which only the value mix of the specific item of the feature information is similar.

In the above embodiment, the search image corresponds to the search object described in claim 1 or 3, the search key image corresponds to the search key object described in claim 1 or 3, and the image database 44 It corresponds to the object storage means according to claim 1 or 3, or the image storage means according to claim 6 or 7. Step S102 corresponds to the object display step described in claim 1, the object display means described in claim 3, or the image display means described in claim 4, 6, 7, 8 or 9. Step S102 and the input device 40 correspond to the search key input step described in the item 1.
Alternatively, it corresponds to the image input means according to claim 6 or 7. Steps S108 to S116 correspond to the object search step described in claim 1, the object search means described in claim 3, or the image search means described in claim 4, 6, or 7, and step S116 is described in claim 2. It corresponds to the axis reassignment step described above or the axis reassignment means according to claim 5.

In the above embodiment, the retrieval key image is inputted from among the retrieval images arranged in the display virtual space. However, the present invention is not limited to this. May be input as a search key image. In this case, for example, the input of the user image, the search of the search image, the arrangement in the display virtual space are performed in this order, and when performing the re-search, the axis of the display virtual space is reassigned, the input of the search key image, The evaluation is performed in the order of the search key image and the search of the search image.

In the above embodiment, a plurality of search key images are input for each search. However, the present invention is not limited to this, and one search key image is input for each search. At the time of the search, the search may be performed based on the feature information corresponding to the search key image input in the current search and the feature information corresponding to the search key image input in the previous search. .

In the above embodiment, the difference between the values min of the items of the feature information is calculated for two or more search key images in two different combinations, and the sum of the absolute values of the calculated differences is the smallest. Is selected as a specific item, but the present invention is not limited to this. For a plurality of search key images, the weighted variance of the value min of each item of the feature information is calculated in two different combinations, and the reciprocal of the calculated weighted variance is The smallest item may be selected as the specific item.

Further, in the above embodiment, the color distribution, the grayscale distribution, the frequency distribution, and the like are employed as the items of the characteristic information. In addition, the average value or the variance of the pixel values of the image data can be considered. . As the average value of the pixel values, a statistic of RGB values in each pixel or a statistic of values of another color coordinate system (HSV, YIQ, or the like) is used. Depending on the color coordinate system used, the features of each color coordinate can be reflected in the search. Further, if attached information including a photographing position, a photographing time, and the like is attached to the search image data, the position, time, season, and the like can be adopted as the characteristic information. As the characteristic information obtained from the attached information, it is conceivable to use information recorded using a GPS system. Specifically, the photographing position is developed three-dimensionally (where the ground axis is the Z axis and longitude and latitude are declinations), and the three values are used as feature information. In addition, the elapsed time from a certain point in time according to date and time, the season, the time in one day, the week, and the like can be used as the characteristic information.

In the above embodiment, when assigning the items of the characteristic information to the axes of the display virtual space in step S116, the specific items of the characteristic information are arranged in ascending order of the value wn of the projection vector, and the top two items are assigned one of them. , And the lower four are assigned to the other axis. However, the present invention is not limited to this, and the feature information items may be equally assigned to each axis, or only specific items may be assigned to one axis. Assignment, other items may be assigned to the other axis, or the feature information item may be assigned to the projection vector value w.
Allocate the upper half to one axis in ascending order of n,
The lower half may be assigned to the other axis.

Further, in the above embodiment, the display virtual space is formed in two dimensions, but is not limited to this, and may be formed in one or three dimensions. As a method of allocating the items of the feature information to the axes of the display virtual space when the display virtual space is three-dimensional, for example, the specific items of the feature information are arranged in ascending order of the projection vector value wn, and the top three items are assigned The method of assigning to the axis and the value w of the projection vector
A method of arranging in ascending order of n and assigning to each axis in that order (that is, when there are six feature information items, 1, 4
The second item on the x-axis, the second and fifth items on the y-axis,
The sixth item is assigned to the z-axis), and the top first item is arranged on the x-axis by
A method is conceivable in which the second highest item is assigned to the y-axis and other items are assigned to the z-axis.

In the above embodiment, the feature information is normalized before the image search of the image. However, the present invention is not limited to this. You may comprise so that it may normalize. Further, in the above embodiment, when executing the processes shown in the flowcharts of FIG. 5 and FIG.
The case where the control program stored in the ROM 32 is executed in advance has been described. However, the present invention is not limited to this, and the program is read from the storage medium storing the program indicating the procedure into the RAM 34 and executed. Is also good.

Here, the storage medium includes a semiconductor storage medium such as a RAM and a ROM, a magnetic storage type storage medium such as an FD and an HD, an optical read type storage medium such as a CD, CDV, LD, and DVD, and a storage medium such as an MO. A magnetic storage type / optical readout type storage medium includes any storage medium that can be read by a computer regardless of an electronic, magnetic, optical, or other read method.

In the above embodiment, the search method, the search device and the image search device according to the present invention are the same as those shown in FIG.
As shown in FIG. 5, the computer 100 uses the plurality of different search key images given as search keys by the user to select a search image that matches the image common to the search key images from among the plurality of search images. Although applied to the case of searching, the present invention is not limited to this, and may be applied to other cases without departing from the gist of the present invention. For example, the present invention is also applicable to a case where an image search is performed on a homepage on the Internet. in this case,
The image of the home page and the document of the home page are obtained from the Internet by robot search or the like, and the obtained image is stored in the image database 44.

[0096]

As described above, according to the retrieval method according to the first or second aspect of the present invention, objects in which the values of the items of the characteristic information are totally similar are displayed in the display virtual space. In addition to being placed close to each other on the top, objects whose only specific item values of the feature information are similar are also placed close to each other in the display virtual space. Therefore, even in the case where only the values of the specific items of the characteristic information are similar to each other, it is possible to obtain an effect that it is easier to visually grasp the relevance than in the related art.

On the other hand, according to the search device of the third aspect of the present invention, the same effect as the search method of the first aspect can be obtained. On the other hand, according to the image retrieval apparatus according to claims 4 to 10 of the present invention, images in which the values of the items of the characteristic information are totally similar are arranged close to each other in the display virtual space. Needless to say, images in which only the value of the specific item of the feature information is similar are arranged close to each other in the display virtual space. Therefore, even in the case of images in which only the values of the specific items of the characteristic information are similar, an effect is obtained that it is easier to visually grasp the relevance of the images than in the related art.

Further, according to the image search device of the ninth aspect of the present invention, images in which only the value of the specific item of the characteristic information is approximate are arranged closer to each other in the display virtual space. In addition, even in the case of images in which only the value of the specific item of the characteristic information is similar, an effect is obtained that it is easier to visually grasp the relevance between the images. Furthermore, according to the image search device of the tenth aspect of the present invention, the user looks at the arrangement status of the display virtual space while proceeding with the search so that the feature of the image to be imaged appears most. If the weight of the reference position or the predetermined weighting calculation is changed, the features of the image imaged by the user in the display virtual space can be visually distinguished, so that the desired image can be found relatively easily. The effect that can be obtained is also obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a computer 100 to which the present invention is applied.

FIG. 2 is a diagram showing search image data.

FIG. 3 is a diagram showing a data structure of a feature information registration table 70.

FIG. 4 is a diagram showing a data structure of an image correspondence table 75.

FIG. 5 is a flowchart illustrating a feature information generation process.

FIG. 6 is a flowchart illustrating an image search process.

FIG. 7 is a diagram illustrating a case where a reference position K is calculated in a search virtual space.

FIG. 8 is a diagram illustrating a case where a projection vector W is calculated in a search virtual space.

FIG. 9 is a diagram for explaining a case where a search image is arranged in a search virtual space based on a result of a weighting operation using a projection vector W;

FIG. 10 is a diagram illustrating a case where a distance from a reference position K is calculated in a search virtual space.

FIG. 11 is a diagram illustrating a case where an item of feature information is assigned to an axis of a display virtual space.

FIG. 12 is a diagram for explaining a case where a reference position K and a projection vector W are changed.

FIG. 13 is a diagram illustrating a case where a search key image is input by giving an evaluation V to a search image.

[Explanation of symbols]

 REFERENCE SIGNS LIST 100 computer 30 CPU 32 ROM 34 RAM 35 VRAM 36 CRTC 38 I / F 40 input device 42 display device 44 image database 70 feature information registration table 75 image correspondence table

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04N 1/21 H04N 1/21

Claims (10)

[Claims] A search object that matches feature information of a search key given as a search key is extracted based on feature information obtained by extracting features of the search target for each of a plurality of items. In the method of searching from among them, a plurality of search targets are stored in a target storage unit, and a plurality of different search keys given as the search keys are input in a search key input step; A target search step of searching a search target in the target storage means for a search target matching the characteristic information common to the search keys based on the respective characteristic information corresponding to the search keys; and The items of the characteristic information are assigned to the axes of the display virtual space based on the value of the common characteristic information, and are searched by the object search step. Search method characterized by including the target display step of arranging and displaying on the display for a virtual space according to the feature information of the search target object. 2. The search method according to claim 1, further comprising an axis reassignment step of reassigning an item of the feature information assigned to an axis of the display virtual space. 3. A search target that matches feature information of a search key given as a search key is extracted based on feature information obtained by extracting features of the search target for each of a plurality of items. An apparatus for performing a search from among them, a target storage unit storing the plurality of search targets, a search key input unit for inputting a plurality of different search keys given as the search keys, and a search key input by the search key input unit A search target that searches for a search target matching the characteristic information common to the search keys from among the search targets in the target storage unit based on the respective feature information corresponding to the search keys; The items of the characteristic information are assigned to the axes of the display virtual space based on the value of the characteristic information, and the characteristics of the search target searched by the target search unit are assigned. Search device being characterized in that a target display means for displaying arranged on the display for a virtual space in accordance with the information. 4. A search image that matches feature information of a search key image given as a search key based on feature information obtained by extracting image features for each of a plurality of items. An apparatus for searching among them, an image storage unit storing the plurality of search target images in association with the feature information, a search key image input unit inputting a plurality of different images as the search key, and the search key image Image search means for searching a search target image in the image storage means for a search image suitable for feature information common to the search key images based on respective feature information corresponding to the search key images input by the input means; Assigning an item of the characteristic information to an axis of a display virtual space based on characteristic information common to each of the search key images, and Image retrieval apparatus comprising the image display means for displaying arranged on the display for a virtual space according to the feature information of the searched search image by. 5. The image retrieval apparatus according to claim 4, further comprising an axis re-assignment unit that is capable of re-assigning the item of the feature information assigned to the axis of the display virtual space. 6. Based on feature information obtained by extracting features of an image for each of a plurality of items, a search image suitable for feature information of a search key image given as a search key is extracted from a plurality of search target images. An apparatus for searching from among the plurality of search target images, the image storage means storing the plurality of search target images in association with the feature information, and allocating items of the feature information to axes of a display virtual space to store the plurality of search target images. Image display means for arranging and displaying the image in the display virtual space in accordance with the value of the item of the corresponding feature information; and displaying the image from the search target images arranged in the display virtual space by the search key. Image input means for inputting as an image, and a search adapted to characteristic information common to the search key image based on respective characteristic information corresponding to the search key image input by the image input means Image search means for searching for an image from the search target images in the image storage means, wherein the image display means is configured to output the feature information based on a value of each feature information corresponding to the plurality of search key images. The image retrieval device is adapted to reassign the item of (1) to the axis of the display virtual space. 7. Based on feature information obtained by extracting features of an image for each of a plurality of items, a search image that matches feature information of a search key image given as a search key is extracted from a plurality of search target images. An image storage unit that stores the plurality of search target images in association with the feature information, and wherein the feature information is set on an axis of a display virtual space having a smaller number of dimensions than the number of items of the feature information. Image display means for allocating a part or all of the items and arranging and displaying the plurality of images on the display virtual space according to the value of the item of the corresponding feature information; and displaying the plurality of images on the display virtual space. Image input means for inputting an image as the search key image from among the search target images arranged in the image processing apparatus, and based on respective characteristic information corresponding to the search key image input by the image input means. Image search means for searching a search target image in the image storage means for a search image suitable for feature information common to the search key images, wherein the image search means corresponds to each of the plurality of search key images. Based on the feature information, a reference position serving as a search reference is calculated in a search virtual space in which each item of the feature information is assigned to a different axis, and the respective feature information corresponding to the plurality of search key images is calculated. A predetermined weighting operation is performed on the value of the item of the search target image in the image storage means so that the value of the specific item having the same or similar value among the items of the item contributes to the search is increased. The search images are arranged in the search virtual space based on the weighting calculation result, and the distance from the calculated reference position in the search virtual space is small. The image display means displays the item of the characteristic information based on the value of each item of the characteristic information corresponding to the plurality of search key images. An image search device, wherein the image search device is reassigned to an axis of a virtual space for use. 8. The image display device according to claim 4, wherein the image display unit displays the specific item having the same or similar value among the items of the characteristic information corresponding to the plurality of search key images. An image search device, wherein the image search device is reassigned to an axis of a virtual space for use. 9. The image display unit according to claim 8, wherein the image display unit, when re-assigning a plurality of items to one axis of the display virtual space, assigns the number of items assigned to the specific item to another axis. Characterized in that the number of assigned items is smaller than the number of assigned items. 10. The image search device according to claim 4, wherein a weight of the reference position or the predetermined weighting calculation can be changed.