JPH07160725A - Picture retrieval device - Google Patents

Abstract

PURPOSE:To eliminate a need for giving a keyboard to each picture at the time of registering the picture and to shorten the time required for retrieval processing at the time of retrieving the picture. CONSTITUTION:When attribute conditions of the picture showing the features at a noticed position of the picture is inputted as picture retrieval conditions from an input device 101, a condition formula synthesizer 103 synthesizes a retrieval condition formula matched to the actual retrieval processing in accordance with the condition formula grammer in a directionary 102. The picture in a data base 104 is longitudinally and transversely divided into 5X5 areas by a divider 105. Attribute information of the picture in each of these areas is calculated by a feature extracting device 106. A feature quantity comparator 107 evaluates attribute information in the position on the picture of each area in accordance with the retrieval condition formula and obtains the degree of match with the retrieval condition with respect to each area and obtains the degree of comformity to the retrieval condition of the whole of the picture based on degrees of match of individual areas. A selector 108 selects the picture, whose obtained degree of comformity is highest, as the picture most meeting the retrieval condition.

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 an image desired by a user from an image database accumulating a large amount of image data based on the contents of the image.

[0002]

2. Description of the Related Art Conventionally, an image database has been developed which stores a large amount of image information in a database and centrally manages image data. Common operations for such an image database include image registration, search, and deletion. These operations usually require keywords for images. That is, when registering an image, the user first selects a keyword that characterizes the image. Then, the image is registered in the image database together with this keyword. Further, when searching for an image, it is performed based on the keyword added at the time of additional registration. Also, when deleting an image, it is necessary to search for an image to be deleted and then delete the image. Thus, the search operation plays a central role in the operation on the image database.

On the other hand, an apparatus for performing image analysis without performing a keyword directly on an image and performing a search based on the content of the image has been studied (for example, image content search, "information processing" Vol. 33 No. 5). , Pp466-477,
1992). However, in this case, since the search is for an image in a specific field such as a painting, a trademark, or a design figure, it is not possible to search for an unspecified image.

[0004]

As described above, the conventional image retrieval methods in the image database include a keyword selection method in which a user selects (assigns) a keyword when registering an image and an image analysis method. It can be roughly divided into two methods, that is, an image analysis method for performing a search based on the content of an image.

The former keyword selection method has an advantage that an image having a keyword matching a search condition set by a user can be surely searched, but a keyword is selected when registering for all images. However, there is a problem in that it imposes a considerable burden on the user.

Further, there is a problem that even an image intended by the user cannot be searched unless an appropriate keyword is selected (assigned) to the image. on the other hand,
In the latter method of image content retrieval using image analysis (image analysis method), it is necessary to limit the types of target images, so this method is applied to images containing many unspecified types. There was a problem that was not effective. Also,
Since this image analysis method generally performs advanced image analysis, there is a problem in terms of processing time and cost.

The present invention has been made in view of the above circumstances, and an object thereof is to reduce the burden on the user without the need for keywords when registering images and to require advanced image analysis when searching for images. It is another object of the present invention to provide a low-cost image search device that does not require search processing time.

[0008]

SUMMARY OF THE INVENTION According to the present invention, in order to retrieve an image held in an image holding means, an attribute condition indicating a feature of an image and a position condition for deciding which region in the image to pay attention to. Search condition input means for inputting a search condition consisting of a combination of, and image dividing means for dividing the image held in the image holding means into a plurality of areas,
Image feature extraction means for obtaining the attribute information indicating the image feature for each of the divided regions, and the image feature indicated by the image attribute information for each region and the attribute condition in the input search condition Degree, and the degree of coincidence between the position in the image of each area obtained and the position condition in the input search condition,
It is characterized in that image feature comparison means for obtaining the matching degree between the search condition and the image and image selecting means for obtaining the image that most matches the input search condition based on the obtained matching degree are provided. .

Further, according to the present invention, a search condition defined in a user-specific notation format (user-defined search condition) so that the search condition can be given in a user-specific notation format.
When a dictionary for registering and linking with the search conditions defined in the system and user-defined search conditions registered in this dictionary are input, the user-defined search conditions are registered in the system. A search condition formula synthesizing means for replacing the defined search condition is further provided.

[0010]

In the above structure, the user inputs the search condition for a desired image using the search condition input means. The search condition is composed of a combination of an attribute condition indicating a feature of an image to be searched and a position condition that determines which region in the image is focused on.

On the other hand, in response to the input of the search condition, the image registered in the image holding means is divided by the image dividing means into a predetermined area, for example, a vertical and horizontal 5 × 5 area.

Then, for each of the divided areas, the image attribute information is calculated by the image feature extracting means. Here, the image attribute information is image feature information that can be easily calculated from image data such as hue, brightness, saturation, and complexity.

The image feature comparison means is input by the user based on the input search condition (attribute condition and position condition) and the attribute information of the image indicated by the feature amount of the image calculated by the image feature extraction means. The matching degree between the input search condition and the image is obtained by calculating how much the obtained search condition matches the actual image feature amount.

For each image in the image holding means, when the matching degree with the inputted search condition is obtained by the image feature comparing means, the image having the highest matching degree is selected by the image selecting means as the search result. The selected image is presented to the user.

As described above, in the above configuration, the image can be efficiently searched for even if the user does not add the keyword to the image. Further, when the search condition dictionary is used, the search condition uniquely defined by the user (user uniquely defined search condition) can be registered in advance in the search condition dictionary. The user-defined search condition is an intuitive expression of the image attribute condition and position condition.

If the desired image feature is "bright at the center", for example, the attribute condition "bright" in the image and the position condition "center" in the image having the same attribute condition "bright" are system-defined. It may be registered in the search condition dictionary in advance in association with the condition of this form (attribute condition and position condition).

In the configuration using such a search condition dictionary, when an attribute condition in an image such as "image with a bright center" and a position condition in the image are input as the search condition, the condition is According to the registered contents of the dictionary, the search condition formula synthesizing means replaces the search condition with a combination of the attribute condition and the position condition of the system-defined format, and this is compared with the image feature amount obtained by the image feature extracting means. Used for.

As described above, by using the search condition dictionary and the search condition expression synthesizing means, the user can define a unique image feature condition such as "a bright image in the center", thereby simplifying the configuration. A desired image can be searched by specifying conditions.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of the image search apparatus according to the embodiment. The image search device shown in FIG. 1 includes an image search condition input device 101, a feature attribute definition user dictionary 102,
Internal search conditional expression synthesizer 103, image database 10
4, an image division device 105, an image feature extraction device 106, an image feature amount comparison device 107, an optimum image selection device 108, and an image output device 109.

The image search condition input device 101 is an input device for the user to input search conditions for images (image information in the image database 104) and to define characteristic conditions unique to the user. , Keyboard, mouse, etc.

The characteristic attribute definition user dictionary 102 is a database for registering search conditions (user uniquely defined conditions) uniquely defined by the user by the image search condition input device 101. The dictionary 102 also stores a conditional expression grammar for converting the search condition input by the image search condition input device 101 into a format suitable for internal processing.

The internal search condition expression synthesizing device 103 replaces the user-specified search condition input by the input device 101 with the uniquely defined search condition defined in the characteristic attribute definition user dictionary 102, and further, The uniquely defined search condition is combined with a search condition expression used inside the system (image search device).

The image database 104 is a database that can store a lot of image information. The image dividing device 105 extracts one image from the image database 104, and divides the extracted image into, for example, 5 × 5 vertical and horizontal sections.

The image feature extraction device 106 calculates image attribute information (hue, lightness, saturation, complexity) for each region of the image divided into 5 × 5 sections by the image division device 105.

The image feature amount comparison device 107 calculates the degree of conformity between the attribute information of each area of the image calculated by the image feature extraction device 106 and the internal search condition formula synthesized by the internal search condition formula synthesis device 103. To do.

The optimum image selection device 108 rearranges the images in descending order of suitability based on the suitability calculated by the image feature amount comparison device 107. The image output device 109 is for presenting the images to the user, and outputs the images rearranged by the optimum image selection device 108 in the arrangement order.

Next, the search condition (user input search condition) input by the user, the characteristic attribute definition user dictionary 102, and the internal search condition expression will be described in detail. FIG. 2 shows an example of search conditions input by the user. As shown in the figure, the user can select attribute conditions such as “bright” / “blue” / “complex” that indicate the characteristics of the image and “center” / “surrounding” / “above” in the image. A search condition as indicated by reference numeral 201 or 202 is given in combination with the position condition indicating the position to be set. Further, the user gives a search condition as indicated by reference numeral 203 by using the expressions combined with logical conditions such as “and” / “or”.

FIG. 3 shows an example of a dictionary registered in the characteristic attribute definition user dictionary 102. Here, the user uniquely defines the characteristic condition of “sky” as “upward (very blue and uncomplicated)”. Further, the user uniquely defines the characteristic condition of “tree” as “green and complex”.

By utilizing the user-specific characteristic conditions defined in this way, it is possible to specify an intuitive search condition as indicated by reference numeral 204 in FIG. The search condition (user input search condition) 204 in FIG. 2 is
User-defined conditions 301 and 30 as shown in FIG.
According to 2, the search condition has the same meaning as "the upper part (very blue and uncomplicated) and the center part (green and complex)".

In the system of this embodiment, the above-mentioned search condition expression is used after being converted into a format in which the grammar is strictly defined. FIG. 4 shows an internal search condition expression corresponding to the search condition of FIG. In the figure, the internal search condition expressions 401 to 404 correspond to the search conditions 201 to 204 shown in FIG. 2, respectively.

FIG. 5 shows the grammar of the internal search conditional expression as BNF.
(Backus-Naur form) notation. The image attribute conditions are “Bright” indicating lightness and “Co” indicating hue.
6, “Contrast” indicating saturation, “Complex” indicating complexity, and 13 types of position conditions shown in FIG. There is something.

As this grammar shows, the internal conditional search expression is given by a logical combination of "&(and)","*(or)", and "# (negation)" in the basic form A @ B. To be In this basic formula A @ B, A corresponds to the above-mentioned attribute condition and B corresponds to the position condition. In addition, the conditions of A and B are "VeryHig" depending on the degree.
h ”,“ High ”,“ Normal ”,“ Low ”,
And "VeryLow" can be modified at five levels.

For example, "Colo" regarding the attribute condition
The description “r: red [Low]” means “the hue is close to red and the degree may be slightly apart from red.” Also, regarding the position condition, “Up [Ver]
The description "yHigh]" means "well above the image".
The description "r: red [Low] @Up [VeryHigh]" means that "a considerable upper area in the image is a reddish image".

Next, the mechanism for calculating the attribute information of the image stored in the image database 104 of FIG. 1 will be described in detail. First, the image is the image database 10
4 is stored as a color image having RGB (red, green, blue) values in pixel units.

When an image search condition is input from the outside via the image search condition input device 101, the image is extracted from the image database 104 by the image dividing device 105. The extracted image is divided by the dividing device 105 into vertical and horizontal 5 × 5 regions as shown in FIG. 7, and then output to the image feature extracting device 106. The respective areas divided here are referred to as S11 to S55.

Next, for each of the areas S11 to S55, the image feature extraction device 106 calculates the attribute information of the image. Here, the attribute information of the image includes hue, lightness, saturation, and complexity. The image feature extraction device 106 first determines the hue, lightness, and saturation of the attribute information in each of the areas S11 to S55.
The hue, lightness, and saturation are calculated for each pixel. Next,
The image feature extraction device 106 takes an average for each of the hue, lightness, and saturation obtained in pixel units.

Regarding the complexity which is the remaining attribute information,
The image feature extraction device 106 uses the value of the high frequency component of the DCT transform (discrete cosine transform) or performs image processing such as region division by labeling.

In the present embodiment, the above hue, lightness,
For both the attribute information of saturation and complexity, a value that is normalized as a continuous numerical value of 0 to 1 is used as the attribute value. FIG. 8 shows an example of the hue attribute information (attribute value) calculated as described above for a certain image. Other attribute information is also expressed in the same format as the hue attribute information shown in FIG.

The attribute information of the image of hue, lightness, saturation, and complexity calculated in this way is used by the image feature extracting apparatus 10.
6 is output to the image feature amount comparison device 107. Next, a method of calculating the matching degree between the internal search conditional expression synthesized by the internal search conditional expression synthesizing device 103 and the image attribute information calculated by the image feature extraction device 106 will be described in detail.

First, the internal search conditional expression synthesized by the internal search conditional expression synthesizer 103 is defined as P & Q, P * Q, # according to the definition of the internal search conditional expression grammar shown in FIG.
It is classified into either P or A @ B format. Where P
And Q are conditional expressions each having the basic form A @ B.

Next, FIG. 9 shows a procedure for calculating the degree of matching between the internal search condition expression and the image attribute information, which is performed by the image feature quantity comparison device 107. First, Expr is an internal search conditional expression, and Img is image attribute information (step S90).
2).

Therefore, the image feature quantity comparison device 107 determines whether the internal search condition expression is described in any of P & Q, P * Q, #P and A @ B formats (step S90).
3 to S906).

Next, the image feature amount comparison device 107 calculates the evaluation value of the internal conditional expression according to the determined internal conditional expression format. The calculation of this evaluation value is performed according to the priority order of the logical symbols. Here, the priority is #, *, &
Since they are set in the order of, the evaluation value is calculated in the order of #, *, &.

The image feature quantity comparison device 107 recursively evaluates P and Q, respectively, when the internal condition search expression is a logical combination of #, *, and &. The internal search condition expression synthesized by the internal search condition synthesizer 103 is P *.
In the case of Q (step S903), the image feature quantity comparison device 1
In 07, first, evaluation values are calculated for P and Q, respectively. Next, the image feature amount comparison device 107 sets the larger one of the evaluation value of P and the evaluation value of Q as the evaluation value of the conditional expression P * Q (step S907).

When the internal search conditional expression synthesized by the internal search conditional expression synthesizer 103 is P * Q (step S90)
4), the image feature amount comparison device 107 first calculates evaluation values for P and Q, respectively. Then, the image feature amount comparison device 107 sets the smaller one of the evaluation value of P and the evaluation value of Q as the evaluation value of the conditional expression P * Q (step S).
908).

When the internal search conditional expression synthesized by the internal search conditional expression synthesizer 103 is #P (step S90)
5), the image feature amount comparison device 107 first calculates the evaluation value of P. Next, the image feature amount comparison device 107 sets the value obtained by subtracting the evaluation value of P from 1 as the evaluation value of #P (step S909).

When the internal search conditional expression synthesized by the internal search conditional expression synthesizer 103 is A @ B (step S90)
6), the image characteristic amount comparison device 107 first determines whether or not A indicating the attribute conditional expression of the image is a conditional expression indicating the hue attribute (step S910).

When A is other than the hue attribute condition, the image feature quantity comparing device 107 causes the image attribute value Img to indicate the value (0
(The range of to 1) itself is the evaluation value of A (step S
912).

On the other hand, when A is the hue attribute condition, the image feature quantity comparing device 107 evaluates the closeness between the specific hue given by the retrieval condition expression and the hue of the actual image, and the evaluation value. (Steps S911 and S913).

The evaluation value of A thus obtained is xa
And Now, when the evaluation value xa of the attribute condition A is obtained as described above, the image feature quantity comparison device 107 next obtains the evaluation value xb of the position condition B.

The evaluation value xb of the position condition B is a numerical value between 0 and 1 in the area corresponding to FIG. The position attribute condition “U” shown in FIG.
10 shows the evaluation value corresponding to p ". Other position attribute conditions" Whole "," Center "," Arou "
d ”,“ Vcenter ”,“ Hcenter ”,“ L
eft ”,“ Right ”,“ Down ”,“ Uple ”
ft ”,“ Upright ”,“ Downleft ”,
The same can be said for "Downright".

Next, the image feature quantity comparison device 107 evaluates the evaluation value xb of the position condition B and the evaluation value xa of the attribute condition A obtained in step S911 or step S913.
Using and, the evaluation value of conditional expression A @ B is obtained (step S914).

At the time of evaluation of the conditional expression A @ B, the image feature quantity comparing device 107 modifies the evaluation values xa and xb of the attribute condition A and the position condition B, respectively, by a five-step modification ", which indicates the degree of the condition. "Veryhigh", "High",
"Normal", "Low", and "VeryLo"
It is re-evaluated according to w ".

That is, the re-evaluation is performed by using the function f (x) for the re-evaluation according to the degree of the above five levels (where x = xa, xb). This function f (x) is
For example, Veryhigh: f (x) = x ^2.0 (1) High: f (x) = x ^1.5 (2) Normal: f (x) = x (3) Low: f (x) = x ^0.7 ... ( 4) VeryLow: f (x) = x ^0.5 (5)

As described above, the image feature quantity comparison device 107
Is the evaluation value x of each of the attribute condition A and the position condition B.
After re-evaluating a and xb, the evaluation value of the conditional expression A @ B is obtained. The following equation (1) is used for this evaluation calculation.

1- | f (xa) -f (xb) | (6) As is apparent from the equation (1), the re-evaluation value f of the attribute condition A is
The closer the value (xa) is to the re-evaluation value f (xb) of the position condition B, the closer the evaluation value of the conditional expression A @ B is to "1", and the closer it is, the closer it is to "0".

Finally, as a simple concrete example, the internal search condition expression is "Color: Blue [High] @Up [L
ow] ”and the hue attribute information of the image calculated by the image feature extraction device 106 is as shown in FIG. 8, the degree of matching between the internal search condition expression and the hue attribute information is obtained.

First, the retrieval condition expression "Color: Blue
[High] @Up [Low] ”is a basic expression, and thus“ Color: Blue [High] ”corresponds to the attribute condition A and“ Up [Low] ”corresponds to the position condition B.

Further, the specific hue attribute information of the hue "Blue" is set to "0.3". Here, since A is the hue attribute condition (step S910), the image feature amount comparison device 107 calculates the specific hue attribute information “0.3” of the hue “Blue” by the image feature extraction device 106. The hue attribute information of the obtained image is compared. The image feature amount comparison device 107 obtains the closeness of each area S11 to S55 in FIG. 8 to the hue attribute information for each area S11 to S55 using the following equation (7) (steps S911 and S913). {Cos (2π | a−b |) +1} / 2 (7) In the above equation, a is preset hue attribute information,
b is the hue attribute information obtained for each of the areas S11 to S55.

Thereafter, the image feature quantity comparison device 107
"High" in "Color: Blue [High]"
The function f (x) = x ^1.5 corresponding to, each region S11~S55
Apply to the evaluation value for the hue attribute information obtained for each
Re-evaluate the hue attribute information. The result of this calculation is shown in FIG.

The image feature quantity comparison device 107 determines the position condition B.
Regarding the evaluation of (“Up [Low]”), the function f (x) = x corresponding to “Low” for each area shown in FIG.
Apply ^0.7 . As a result, the position condition “Up [Lo
The position attribute information in the case of evaluating “w]” is shown in FIG.
become that way.

The image feature quantity comparison device 107 subtracts the absolute value of the difference between them (the difference between the re-evaluation value of the hue attribute condition A and the re-evaluation value of the position condition B) from 1 to obtain the evaluation value of the attribute information. It is calculated for each of the areas S11 to S55 (step S914). This is shown in FIG.

The image feature quantity comparison device 107 is shown in FIG.
The average value of the evaluation values for each of the areas S11 to S55 shown in 3 is obtained (step S915). This average value is the goodness of fit. In the example shown in FIG. 13, the goodness of fit is “0.81”.

When the goodness of fit is obtained in this way, the optimum image selection device 108 rearranges the images in the descending order of goodness of fit obtained by the image feature quantity comparison device 107. Then, the selection device 108 sends the image having the highest matching degree to the image output device 109.

The image output device 109 presents the image sent from the optimum image selection device 108 to the user. At this time, if the presented image is not the image desired by the user, the optimum image selection device 108, for example, responds to the next candidate output request operation by the user and the next candidate image, that is, the next matching degree. The high image is sent to the image output device 109.

Thereafter, the optimum image selection device 108 sends the images to the image output device 109 in descending order of suitability until the image desired by the user is presented. In the above embodiment, the hue, saturation, brightness, and complexity are used as the image attribute information, and the 13 types of information shown in FIG. 6 are used as the position attribute information, but the present invention is not limited to this. It is not something that will be done. For example, information indicating the edge change of the image may be used as the image attribute information.

[0067]

As described above, according to the present invention, the condition indicating the image attributes such as hue, brightness, saturation, and complexity, and the condition indicating the position of the image attribute of interest in the image. If a combination is given as a search condition, the registered image is divided into a plurality of areas, and attribute information such as hue, brightness, saturation, and complexity of the image is obtained for each divided area, and this is obtained. An image that meets the search condition without the need for the user to add a keyword is obtained by finding the matching level between the search condition and the image based on the attribute information and selecting the image with the highest matching level. Can be searched.

Further, according to the present invention, the retrieval process is simple and fast as compared with the conventional image content retrieval using the image analysis method. Further, according to the present invention, the user can define the condition indicating the feature of the image by himself / herself, so that the user can input the search condition in a more simple and easy-to-understand format. Therefore, the work efficiency at the time of image search can be improved.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an image search device according to an embodiment of the present invention.

FIG. 2 is a diagram showing an example of search conditions input by a user.

FIG. 3 is a diagram showing an example of user-specific attribute information registered in a characteristic attribute definition user dictionary 102 of FIG.

FIG. 4 is a diagram showing an example of an internal search conditional expression synthesized by an internal search conditional expression synthesizer 103 of FIG.

5 is a diagram showing a grammar of an internal search conditional expression synthesized by the internal search conditional expression synthesizer 103 of FIG.

FIG. 6 is a diagram showing position attributes of images in the image database 104 of FIG.

FIG. 7 shows 5 × images in the image database 104 of FIG.
The figure which shows the state divided | segmented into the area | region of 5.

FIG. 8 is a diagram showing an example of a hue attribute of an image.

FIG. 9 is a diagram showing a calculation procedure of a matching degree between an internal search conditional expression and image attribute information.

FIG. 10 is a diagram showing an example of evaluation values of position attribute information “Up”.

FIG. 11 is a diagram showing an example of evaluation values of attribute condition A in conditional expression A @ B.

FIG. 12 is a diagram showing an example of evaluation values of position condition B in conditional expression A @ B.

FIG. 13 is a diagram showing an example of evaluation values of a conditional expression A @ B of a search condition in which an attribute condition A and a position condition B are combined.

[Explanation of symbols]

101 ... Image search condition input device, 102 ... Feature attribute definition user dictionary, 103 ... Internal search condition expression synthesizing device, 104
... image database, 105 ... image dividing device, 106 ...
Image feature extraction device, 107 ... Image feature amount comparison device, 10
8 ... Optimal image selection device, 109 ... Image output device.

Claims (2)

[Claims] 1. An image retrieval apparatus for retrieving an image desired by a user from an image retaining unit that retains a plurality of images, wherein an attribute condition indicating a characteristic of an image to be retrieved and which region in the image is selected. Search condition input means for inputting a search condition consisting of a combination with a position condition that decides whether to focus attention, image dividing means for dividing the image held in the image holding means into a plurality of areas, and the image division Image feature extraction means for obtaining attribute information indicating image features for each area divided by means, image feature indicated by the image attribute information for each area obtained by the image feature extraction means, and the search The degree of coincidence with the attribute condition of the image in the search condition input by the condition input means, the position in the image of each of the divided areas, and the position in the input search condition Calculating the degree of coincidence between matter, based on the image feature comparing means for obtaining a goodness of fit between the search criteria and the image, the fitness obtained by the image feature comparison unit,
An image search apparatus comprising: an image selection unit that obtains an image that most matches the search condition input by the search condition input unit. 2. An image retrieval apparatus for retrieving an image desired by a user from an image retaining unit that retains a plurality of images, wherein an attribute condition indicating characteristics of an image to be retrieved and which region in the image A search condition input means for inputting a search condition consisting of a combination with a position condition that determines whether or not to pay attention, and an attribute condition and a position condition that are uniquely defined by the user and that indicate the characteristics of the image are searched. A search condition dictionary for registering attribute conditions and position conditions in a format suitable for internal processing in association with each other, and a user unique definition consisting of a combination of user-defined attribute conditions and position conditions registered in the dictionary. When a search condition is input by the search condition input means, the user-defined search condition is converted into a format suitable for a predetermined internal search process by using the dictionary. A search condition formula synthesizing unit that replaces a search condition that is a combination of an attribute condition and a position condition, an image dividing unit that divides the image held in the image holding unit into a plurality of regions, and a image dividing unit that divides the image. Image feature extraction means for obtaining attribute information indicating image characteristics for each region, image feature indicated by the image attribute information for each region obtained by the image feature extraction means, and the search condition formula combining means The degree of coincidence with the attribute condition of the image in the search condition replaced by the format suitable for the internal processing by, and the position condition in the previous search condition replaced with the position in the image of each of the divided regions, And the image feature comparison means for calculating the matching degree between the search condition and the image, and the matching degree calculated by the image feature comparison means,
An image search apparatus further comprising: an image selection unit that obtains an image that most matches the search condition input by the search condition input unit.