JPH05242160A - Image feature extraction device, image collation device, and image retrieval device - Google Patents

Abstract

PURPOSE:To provide the image retrieval device which performs retrieval by semiautomatically extracting feature quantities from objects in images and collating the images even when the extracted feature quantities and the objects in the images are different in number, and can adaptable for various retrieval requests. CONSTITUTION:When an image data base is generated, the areas and names of constituent elements in the images are inputted and the image feature quantities in the specified areas are automatically extracted. When the images are collated, a corresponding constituent element extraction part 9 extracts the corresponding constituent elements in the images and a constituent element similarity calculation part 10 finds the similarities between the elements, element by element, from the image feature quantities and couples them to find the degrees of matching between the elements. The similar image is retrieved according to the found degrees of matching between the images and then the retrieval can be performed even if the numbers of the constituent elements in the images are different; and diverse retrieval is provided by changing the form of the function for finding the similarities.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image retrieving apparatus for retrieving a feature amount from an image to create an image database and retrieving an image similar to an exemplary image from the created image database.

[0002]

2. Description of the Related Art As a conventional image retrieval method, for example, there is one shown in IE86-117, Technical Report of IEICE.

In this method, a mountain is recognized as an object from a landscape image and automatically extracted, and the color, size, position and shape are obtained as the attribute characteristics of the mountain. The accumulated image is represented as a point in a four-dimensional space of color, size, position, and shape, and matching with the search request image is performed in this feature space. That is, the distance between the search request image and the stored image is obtained in the feature space, and the stored image closest to the search request image is output as the search result.
Alternatively, a plurality of images are output in ascending order of distance.

[0004]

In the conventional example, a mountain is recognized as an object, but there are various kinds of natural images other than the mountain, and if they overlap each other, the object in the image It is very difficult to automatically extract and recognize objects.

On the other hand, when a similar image is searched by collating an image composed of a plurality of objects, the objects included in the image are different depending on the accumulated image, or the feature amount obtained from the object is different for each image. Sometimes. In this case, even if an attempt is made to obtain the distance in the feature space as in the conventional example, there is a problem that the distance cannot be obtained because the number of objects and feature amounts differ for each image.

Further, it is not possible to realize various searches such as lowering the search order when the degree of search request differs for each object as a search request or when objects other than the objects indicated by the search request are included. Have difficulty.

In view of the above point, the present invention semi-automatically extracts a feature amount from an object in an image and collates the images even when the extracted feature amount or the number of objects included in the image is different. It is an object of the present invention to provide an image search device capable of searching and further responding to various search requests.

[0008]

In order to solve the above-mentioned problems, the following constitution is adopted.

According to a first aspect of the invention, an image input section for inputting an image for extracting image features, an image display section for displaying the input image, and an area designating section for designating areas of constituent elements in the image. An element name input unit for inputting an element name of a component designated by the region designation unit, an image feature extraction unit for extracting an image feature for each component from the region of the designated component, an element name and an image feature An image feature quantity storage unit that stores quantities in association with each other is provided.

According to a second aspect of the present invention, a matching image input unit for inputting an image to be matched, a matched image input unit for inputting an image to be matched, and a matching image and a matched image have the same component name. A corresponding component extraction unit that extracts the image similarity, an inter-component similarity calculation unit that calculates the image feature similarity for each corresponding component, and a similarity for each corresponding component of the matching image and the matched image And an inter-image matching degree calculation unit that obtains the matching degree between the images.

According to a third aspect of the present invention, a collation image input section for inputting an image to be collated, a collation image input section for inputting an image to be collated, and a collation image and a collation image have the same element name. Corresponding component extraction unit that extracts the image feature, an inter-component similarity calculation unit that calculates the similarity of image features for each corresponding component, and a component that is present in the collation image but not in the collation image A non-corresponding component extraction unit, a dissimilarity calculation unit that calculates a dissimilarity based on the components extracted by the non-corresponding component extraction unit, and each corresponding component of the collation image and the collated image And an inter-image matching degree calculation unit that obtains a matching degree between images by integrating the dissimilarities of the constituent elements that do not correspond to the above-mentioned similarity degrees.

According to a fourth aspect of the present invention, a collation image input unit for inputting an image to be collated, a collation image input unit for inputting an image to be collated, and a collation image and a collation image have the same element name. A corresponding component extraction unit that extracts a component, a target element designating unit that designates a target component among the corresponding components, and a similarity of image features for each component corresponding to the component designated as the target component An element-of-interest similarity calculation section, a non-element-of-interest element similarity calculation section that calculates the image feature similarity for each constituent element other than the element of interest among the corresponding constituent elements, and an element of interest of the matching image and the collated image An inter-image matching degree calculation unit that obtains a matching degree between images by integrating the similarity degree for each of the elements and the similarity degree for each of the non-target components.

A fifth aspect of the present invention comprises an image feature extracting device and an image collating device, each of which comprises the above means, an image storage unit for storing an image, and an image input unit for inputting a search request image.

[0014]

According to the present invention, by inputting the region and the component name of the component in the image at the time of creating the image database, the image feature amount such as the position, the area, the color, etc. is automatically generated from the designated region. It is extracted and registered as a database together with the component name.

Further, when performing image matching, the similarity is calculated for each corresponding component between images, and the similarity of the entire image is calculated by combining them. Further, the dissimilarity is obtained for the constituent elements that are not present in the search request image but are present in the accumulated image, and the similarity degree of the entire image is obtained by combining the dissimilarity degree together with the similarity degree of each corresponding constituent element.

Further, the degree of search request different for each component can be expressed by a function for obtaining the degree of similarity.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An image feature extraction apparatus according to the first embodiment of the present invention will be described. The configuration of the image feature extraction apparatus of the first embodiment is shown in FIG. Reference numeral 1 is an image input unit, 2 is a region designation unit for designating regions of constituent elements in an image, 3 is an element name input unit, 4
Is an image feature extraction unit, 5 is an image display unit, and 6 is an image feature amount storage unit.

The operation of the image feature extracting apparatus of the first embodiment configured as described above will be described by taking the case of creating an image database from the images shown in FIG. 2 as an example. In FIG. 2, there are mountains, people, and houses as constituent elements in the image.
When this image is input to the image input unit 1 of FIG. 1, it is displayed on the image display unit 5. While viewing the displayed image, the user specifies the area of each constituent element in the image with the area specifying unit 2 and inputs the element name (mountain, person, house) of each constituent element with the element name input unit 3. .. The designated area is indicated by a wavy line in FIG. The image feature extraction unit 4 extracts the image feature amount in the designated area. As the image feature amount, the center position of the area, the area of the area, the main color in the area, and the like are extracted. The image feature amount storage unit 6 registers the element name input by the element name input unit 3 and the image feature in the designated area extracted by the image feature extraction unit 4.

As described above, in the present embodiment, since the human beings perform the area extraction and recognition of the object in the image, it is possible to deal with any complicated image. On the other hand, after extracting the region of the target object, the image feature amount is automatically extracted from the image in the region, which enables a simple image database creation.

Next, an image collating apparatus according to the second embodiment of the present invention will be described. FIG. 4 shows the configuration of the image collating apparatus according to the second embodiment. In FIG. 4, 7 is a collation image data input unit, 8 is a collation image data input unit, 9 is a corresponding component extraction unit, 10 is an inter-component similarity degree calculation unit, and 11 is an inter-image collation degree calculation unit.

The operation of the image collating apparatus of the second embodiment constructed as above will be described. Image data to be collated is input to the collation image data input unit 7, and image data to be collated is input to the collation target image data input unit 8. Corresponding component extraction unit 9 extracts components having the same component name from the input image data. The inter-element similarity calculation unit 10 calculates the similarity for each corresponding element between images as follows. A plurality of image feature quantities are extracted from the constituent elements, and the similarity degree for each feature quantity is obtained from the membership function of the similarity degree determined based on the distance of the image feature quantity between the two constituent elements. An example of the membership function of similarity is shown in FIG. The degrees of similarity of the respective feature amounts obtained from the membership function of FIG. 5 are combined using the Dempster-Shafer (Dempster-Shafer) connection rule. This will be described in more detail.

First, the Dempster-Shafer theory will be briefly described. Let p ₁ and p ₂ be the basic probabilities obtained based on independent evidence, and let B _1q and B _2r (q, r = 0,1,2, ...) be the focal elements of Dempster-Shafer. The basic probabilities can be integrated by (Equation 1) from the combining rule.

[0023]

[Equation 1]

Two images are similar as focal elements
A for things, A for things that are not similar ^cAnd a certain configuration is required
Raw image feature x_iThe similarities in p
_i(A), the dissimilarity factor is p_i(A^C).
In addition, the ignorance amount is p_i(A, A^C). At this time the image features
X as a trait_iConfidence p when_i(A), p
_i(A, A^C) Is explained.

Now, consider the positions and areas of the constituents as the image feature quantities. It is assumed that a difference in position between corresponding components between two images is x ₁ and a difference in area is x ₂ . Similarities s ₁ and s ₂ are obtained from x ₁ and x ₂ using a membership function. For s ₁ and s ₂ obtained from the membership function, these values are divided by the number of feature quantities (2 in this case), and this is p ₁
(A) and p ₂ (A). Furthermore, the ignorance amount p ₁ (A,
A _C ) and p ₂ (A, A _C ) are calculated from (Equation 2).

[0026]

[Equation 2]

From the obtained p _i (A) and p _i (A, A _C ), the similarity p (A) of the constituent elements is obtained by (Equation 1).
When applied to (Equation 1), it becomes like (Equation 3).

[0028]

[Equation 3]

The degree of similarity between components can be obtained as described above. In addition, here, the degree of similarity is calculated for each feature amount for a plurality of feature amounts of the constituent elements, and the similarity is calculated.
We calculated the similarity between constituents using the Dempster-Shafer associative rule, and calculated the distances in the multidimensional feature space for multiple features of the constituents, and then calculated the similarity membership function. The similarity between components may be obtained.

The inter-image collation degree calculation unit 11 integrates the similarity degrees between the corresponding components of the collation image and the collated image obtained by the inter-element similarity calculation unit 10 to obtain the collation degree between the images. To calculate. Dempster-Shafer for similarity integration
The associative rule of is used. The details will be described below.

Now, it is assumed that there are L constituent elements of the collation image, and there are N pairs of corresponding constituent elements between the collation image and the collation image. At this time, p ¹ ′ (A) to p ^N ′ (A) are calculated from the similarities p ¹ (A) to p ^N (A) for each component obtained by the inter-component similarity calculation unit 10 in (Equation 4). Ask.

[0032]

[Equation 4]

Further, from (Equation 5), the ignorance amount p ¹ '(A,
A _C ) -p ^N '(A, A _C ) is calculated.

[0034]

[Equation 5]

P ⁱ ′ (A) and p ⁱ ′ of the obtained N constituent elements
From (Equation 1) using (A, _AC ), the inter-image matching degree M
Find (A).

As described above, the matching degree M (A) between images
Can be asked. As in the present embodiment, matching is performed for each constituent element in the image to obtain the degree of similarity, and the similarity is calculated as Demp.
By using the ster-Shafer associative rule and integrating to obtain the matching degree between images, it is possible to perform image matching even if the number of components is different between images.

Next, an image collating apparatus according to the third embodiment of the present invention will be described. FIG. 6 shows the configuration of the image collating apparatus according to the third embodiment. The same parts as those in FIG. 4 are designated by the same reference numerals and the description thereof will be omitted. In FIG. 6, 12 is a non-corresponding component extraction section, 13 is a dissimilarity calculation section, and 14 is an inter-image matching degree calculation section.

The operation of the image collating apparatus of the third embodiment constructed as above will be described. The process up to the step of obtaining the similarity between the corresponding components of the collation image and the collated image in the component similarity calculation unit 10 is the same as in the second embodiment of the present invention, and a description thereof will be omitted. The non-corresponding component extraction unit 12 extracts components that are present in the collation image but not in the collation image. The dissimilarity calculation unit 13 calculates the dissimilarity based on the components extracted by the non-corresponding component extraction unit 12.

The dissimilarity pd (A _C ) is calculated from the sum Sd of the areas of the element regions extracted as the non-corresponding constituents and the area S of the entire image by (Equation 6).

[0040]

[Equation 6]

Similar to the second embodiment of the present invention, the similarity calculation unit 10 between the constituent elements calculates the similarity p for each constituent element.
¹ (A) to p ^N (A) are obtained, and p ¹ '(A) to p in (Equation 4)
^N '(A), and from (Equation 5), ignorance p ¹ ' (A, A _C ).
_{^{~p N '(A, A C}} ) Request.

The inter-image matching degree calculation unit 14 uses the Dempster-Sha
Based on fer's associativity (Equation 1), p ⁱ '(A),
From p ⁱ '(A, A _C ), pd (A _C ) to the inter-image matching degree M
Find (A).

As described above, the matching degree M (A) between images
Can be asked. As in the present embodiment, the matching is performed for each corresponding component between images to obtain the similarity, and the dissimilarity is determined for the non-corresponding elements.
By using the Dempster-Shafer associative rule to integrate and obtain the degree of matching between images, image matching is performed even when the number of constituent elements differs between images or when there are extra constituent elements in the collated image. It is possible to

Next, an image collating apparatus according to the fourth embodiment of the present invention will be described. FIG. 7 shows the configuration of the image collating apparatus according to the fourth embodiment. The same parts as those in FIG. 4 are designated by the same reference numerals and the description thereof will be omitted. In FIG. 7, reference numeral 15 is a corresponding component extraction unit, 16 is a target element designation unit, 17 is a target element similarity calculation unit, 18 is a non-target element similarity calculation unit, 19
Is an inter-image matching degree calculation unit.

The operation of the image collating apparatus of the fourth embodiment constructed as above will be described. Corresponding component extraction unit 15 extracts components having the same component name among the input image data. The attention element designating unit 16 designates the attention constituent element among the constituent elements of the image input to the collation image input unit 7. In the similarity calculation unit 17 between the elements of interest,
The similarity between images is calculated for the target element among the constituent elements extracted by the corresponding constituent element extraction unit 15. The non-interest element similarity calculation unit 18 calculates the inter-image similarity for components other than the target element among the components extracted by the corresponding component extraction unit 15. The inter-image matching degree calculation unit 19 integrates the similarities of the respective constituent elements obtained by the target element similarity calculation section 17 and the non-target element similarity calculation section 18 to calculate the image matching level.

The process of calculating the matching degree will be described in more detail below. It should be noted that, in the following, for the sake of simplicity of explanation, the image feature amount in each component is only the position, but the same applies to a plurality of image feature amounts.

The difference between the attention element and the non-attention element is the similarity.
This is the shape of the desired membership function. Here is the image feature
Since the amount is the position, the element of interest is the position of the component.
It is to pay attention. The position where the focused component exists
The area is limited, and the one that exists in the limited area is the similarity.
Highly dissimilar to those outside the limited area
Try to be. On the other hand, non-attentional elements are limited in location
The similarity will decrease as the position difference increases.
I will Membership for finding the similarity to the element of interest
Figure 8 shows the B-function and finds the similarity to non-target elements.
The membership function that The solid line in FIG.
Similarity and wavy lines indicate dissimilarity, and limited areas are similar
This is where the degree becomes 0. The membership function in Figure 8
Therefore, the similarity p for the element of interest in the limited areaⁱ
(A) is obtained, and the degree of dissimilarity for those outside the limited area
p^j(A_C) Is required. Here, the element of interest exists in the collation image.
If not present, the dissimilarity is 1. For non-attention elements
Then, from the membership function of FIG.^k(A)
, But when the non-attention element does not exist in the matching image
Has a similarity of 0. Divide these by the number of components N and p
ⁱ'(A), p^j'(A _C), P^k'(A) asked,
Ignorance p as in (Equation 5)ⁱ'(A, A_C), P^j'
(A, A_C), P^k'(A, A_C).

The inter-image matching degree calculation unit 19 uses the Dempster-Sha
Based on fer's associativity (Equation 1), p ⁱ '(A),
p ⁱ '(A, A _C ), p ^j ' (A _C ), p ^j '(A, A _C ),
From p ^k '(A), p ^k ' (A, A _C ) the inter-image matching degree M
Find (A).

As described above, the matching degree M (A) between the images
Can be asked. In the present embodiment, when matching is performed for each corresponding component between images and the similarity is obtained, the similarity is obtained between the focused element and the non-focused element, and the Dempster is calculated.
-The degree of matching between images is obtained by integrating using Shafer's connection rule. This makes it possible to perform image matching with a strict similarity measure for the element of interest.

Next, an image search apparatus according to the fifth embodiment of the present invention will be described. In the present embodiment, the image feature extraction device of the first embodiment and the second, third, or fourth embodiment.
The image matching device is used to configure an image search device. FIG. 10 shows the configuration of the image search apparatus of the fifth embodiment. In FIG. 10, 20 is an image input unit that captures images to be stored as an image database, 21 is a search request image input unit, 22 is an image feature extraction device, and 23 is the image input unit 2.
An image data storage unit that stores image data input from 0, a feature amount storage unit 24 that stores the image features extracted by the image feature extraction device 22, and a feature amount storage unit 25 and a feature amount storage unit 24 of the search request image. An image collation device that obtains a candidate image by comparing the stored image features of the stored images, and an image output unit 26 that extracts the candidate image obtained by the image collation device from the image data storage unit 23 and outputs it. The operation of the image search device configured as described above will be described below.

At the time of image storage, the image input unit 20 takes in an image to be stored as an image database, and the image data storage unit 23 stores the image. The image feature extraction device 22 extracts image features from the input image. The extraction of image features is as described in the first embodiment.

At the time of image retrieval, the retrieval request image input section 21
Captures the image requested by the user for the search. The search request image is given as a handwritten image of the image to be searched by the user, a similar image, or an image showing the region of the constituent element. From the captured image,
The image feature extraction device 22 extracts image features. In the image matching device 25, as in the second, third, or fourth embodiment, the feature amount of the accumulated image and the feature amount of the search request image stored in the feature amount storage unit 24 created at the time of image accumulation are stored. The comparison is performed, and the degree of similarity between the accumulated image and the search request image is obtained. Among the calculated degrees of similarity, the images having a certain degree or more are taken out from the image data storage unit 23 in order from the one having the highest degree of similarity, and output to the image output unit 26.

In the image retrieval apparatus of the present embodiment described above, when extracting the image feature amount, a human performs the extraction and recognition of the object area in the image as described in the first embodiment. It can also handle complex images. On the other hand, after extracting the region of the object, the image feature amount is automatically extracted from the image in the region, so that a simple image database can be created. In the image matching device, the second
As described in the embodiment, since the matching is performed for each constituent element in the image, it is possible to perform the matching even if the number of corresponding constituent elements in the search request image and the accumulated image are different. Further, by using the image collating apparatus according to the third embodiment, when a component other than the component designated in the search request image is present in the stored image, the search can be realized by lowering the search priority. it can. Further, by using the image collating apparatus of the fourth embodiment, it is possible to realize a search focusing on specific constituent elements. As described above, the image search apparatus according to the present embodiment can retrieve images corresponding to various search requests intended by the searcher from the stored images.

[0054]

As described above, in the image feature extraction apparatus of the first invention, since the human performs the area extraction and recognition of the object in the image, it can handle any complicated image. After the area is extracted, the image feature amount is automatically extracted from the image in the area, so that a simple image database can be created.

Further, in the image collating apparatus of the second invention, the constituent elements in the image are collated to obtain the degree of similarity, which are integrated by using the Dempster-Shafer associative rule to obtain the degree of collation between the images. By determining, it is possible to perform image matching even if the number of constituent elements differs between images.

In the image collating apparatus according to the third aspect of the present invention, the components corresponding to each other between the images are collated to obtain the degree of similarity, the dissimilarities of the non-corresponding elements are obtained, and the dissimilarities are calculated by the Dempster-Shafer By combining using the associative rule and determining the matching degree between images, image matching can be performed even when the number of constituent elements differs between images or when there are extra constituent elements in the matching image. It is possible.

Further, in the image collating apparatus according to the fourth aspect of the invention, when collating each corresponding component between images to obtain the similarity, the similarity between the attention element and the non-attention element is obtained, and the similarity is determined by the Dempster- The degree of matching between images is obtained by integrating using Shafer's associative rule. This makes it possible to perform image matching with a strict similarity measure for the element of interest.

Further, in the image retrieval apparatus of the fifth invention, it is possible to semi-automatically extract the feature amount from the object in the image to construct the image database, and when performing the image retrieval,
It is possible to perform a search even when the feature amount and the number of objects included in the images differ between the images to be collated, and it is possible to specify unnecessary components or focus on a specific component to perform a search.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an image feature extraction device according to a first embodiment.

FIG. 2 is a diagram showing an input image of the first embodiment.

FIG. 3 is a diagram showing a designated area of the first embodiment.

FIG. 4 is a block diagram showing a configuration of an image matching device according to a second embodiment.

FIG. 5 is a diagram showing a similarity membership function of the second embodiment.

FIG. 6 is a block diagram showing a configuration of an image matching device according to a third embodiment.

FIG. 7 is a block diagram showing a configuration of an image matching device according to a fourth embodiment.

FIG. 8 is a diagram showing a membership function of similarity of a target element according to the fourth embodiment.

FIG. 9 is a diagram showing a membership function of similarity of non-attention elements according to the fourth embodiment.

FIG. 10 is a block diagram showing the configuration of an image search device according to a fifth embodiment.

[Explanation of symbols]

 1, 20, 21 Image input unit 2 Area designation unit 3 Element name input unit 4 Image feature extraction unit 5 Image display unit 6 Image feature amount storage unit 7 Matching image data input unit 8 Matched image data input unit 9, 15 Corresponding configuration Element extraction unit 10 Inter-element similarity calculation unit 11, 14, 19 Inter-image matching degree calculation unit 12 Non-corresponding component extraction unit 13 Dissimilarity calculation unit 16 Target element designation unit 17 Target element similarity calculation unit 18 Non Element-to-element similarity calculation unit 22 Image feature extraction device 23 Image data storage unit 24 Feature amount storage unit 25 Image matching device 26 Image output unit

Claims (5)

[Claims] 1. An image input unit for inputting an image for extracting image features, an image display unit for displaying the input image,
A region designating section that designates a region of a constituent element in an image, an element name input section that inputs the element name of the constituent element designated by the region designating section, and an image feature for each constituent element from the designated component area An image feature extraction unit that extracts the image feature amount and an image feature amount storage unit that associates and stores the element name and the image feature amount, the user inputs the regions and names of the constituent elements in the image while looking at the input image, An image feature extraction device characterized in that an image feature amount in an area is automatically extracted from an area of each component and stored in an image feature amount storage unit together with a component name. 2. A collation image input unit for inputting an image to be collated, a collation image input unit for inputting an image to be collated, and a unit for extracting a collation image and a collation image having the same component name. The component extraction unit, the inter-component similarity calculation unit that calculates the similarity of image features for each corresponding component, and the similarity between the corresponding components of the matching image and the collated image An image collation apparatus comprising an inter-image collation degree calculation unit that obtains the degree of collation, and obtains the image collation degree by integrating the degree of similarity obtained for each corresponding component between two images. 3. A collation image input unit for inputting an image to be collated, a collation image input unit for inputting an image to be collated, and a unit for extracting a collation image and a collation image having the same component name. A component extraction unit, an inter-component similarity calculation unit that calculates the image feature similarity for each corresponding component, and a non-corresponding configuration that extracts components that are present in the collated image but not in the collation image An element extraction unit, a dissimilarity calculation unit that calculates a dissimilarity based on the components extracted by the non-corresponding component extraction unit, and a similarity for each corresponding component of the collation image and the collation image An inter-image matching degree calculation unit that integrates dissimilarities of non-corresponding constituent elements to obtain a matching degree between images is provided, and from the constituent elements that do not correspond to the similarity degree obtained for each corresponding constituent element between two images. Image matching by integrating the dissimilarities found Image matching system and obtains a. 4. A collation image input section for inputting an image to be collated, a collation image input section for inputting an image to be collated, and extraction of a collation image and a collation image having the same component name. A component extracting unit, a target element designating unit that designates a target component among corresponding components, and a similarity between target components that calculates the similarity of image features for each corresponding component with respect to the component designated as the target component Degree calculation unit, a non-interest element similarity calculation unit that calculates the similarity of image features for each constituent element other than the target element among the corresponding constituent elements, and a similarity degree for each target constituent element of the matching image and the collated image And an inter-image matching degree calculation unit that calculates the matching degree between the images by integrating the similarity degrees of the respective non-focused constituent elements Characterized by finding the degree of matching Image matching apparatus. 5. An image feature extraction device according to claim 1, an image matching device according to claim 2, claim 3, or claim 4, an image storage unit for storing images, and an image for inputting a search request image. An image feature is extracted by the feature amount extraction device from the search request image input to the image input unit and the image stored in the image storage unit, and the search request image and image storage are performed by the image matching device. An image retrieving apparatus, characterized in that an image similar to a retrieval request image is retrieved from an image storage unit by collating with an image stored in a unit.