JP2001319232A - Device and method for retrieving similar image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the retrieval accuracy by effectively retrieving an image having a feature in a gradation, pattern or texture like a photograph image. SOLUTION: This device is provided with an image input part 2 for inputting image data, an image data storage part 6 for storing the image data, a feature amount extracting part 3 for extracting plural kinds of feature amounts from the image data and a feature amount retrieving part 4 for calculating the similarity of the image data inputted by the image input part 2 and the image data stored in the image data storage part 6 by respectively comparing plural kinds of feature amounts extracted by the feature amount extracting part 3 in both the image data and for retrieving image data similar to the image data inputted by the image input part 2 out of the image data storage part 6 on the basis of that similarity.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a similar image search device and a similar image search method for searching an image database or the like for an image similar to a designated image.

[0002]

2. Description of the Related Art Conventionally, as a method of searching for a specific image from an image database in which a large number of image data are stored, when registering image data in an image database, a search for a keyword or the like is performed on the image data. It has been a general method to add accompanying data and to search for and specify image data using information of the accompanying data for search.

[0003] However, the added keyword is
The content differs depending on the registered user and the content of the keyword required for the search, and effective search results may not be obtained even when the search is actually performed using the keyword.

Therefore, a method has been proposed in which a search is not performed based on information added at the time of registration, such as a keyword, but is performed based on characteristics of an image itself.

[0005] For example, as disclosed in Japanese Patent Application Laid-Open No. 11-212993, a method for searching based on the color characteristics of an image, or displayed on an image as disclosed in Japanese Patent Application Laid-Open No. 9-35058. This is a search method based on the features of the outline of the object.

When a similar image is searched for based on the feature amount of an image, the feature amount of the image is expressed as a vector in a feature amount space as shown in FIG. Distance K1 between (A →) and a feature amount vector (B →) of image data registered in advance.
It is a general method to compare and determine that the closer the distance, the more similar.

[0007]

In such a search for similar images, if the content of the image is a simple image such as a mark such as a trademark or a clip art, it is based on the color feature or the feature of the outline of the object. It is possible to search to some extent.

[0008] However, in the case where an image contains many gradations such as computer graphics and landscape photographs, and the density gradient of the gradations is a large feature of the image, the above-mentioned color feature is used. The image retrieval method based on the image and the contour feature is not effective.

In other words, a drawn object can be easily specified in the case of a mark such as a trademark or clip art, so that its outline can be extracted and its outline characteristic can be obtained. It is difficult to extract contour features from a photographic image because it is difficult to extract.

In a photographic image, a fine pattern or texture such as a leaf or a meadow, a cloth or a rock surface of a tree often has the characteristic of the image. In the conventional method, the characteristic of the pattern or texture is such. There is a problem that can not be used effectively.

Furthermore, when searching for similar images, an effective search result may not be obtained by comparing the distance between the feature vector of the search key image and the feature vector of the registered image.

For example, a feature vector (C →) of the search key image and a feature vector (D
When the absolute value of the feature amount vector is small as in (→), there is a case where the directionality (see θ1 and θ2) of the vector greatly differs even for the feature amount vector of the registered image with a short distance K2. Therefore, when a human actually compares those images, it is determined that the similarity is low, and the image cannot sufficiently serve as a search system.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to effectively search for an image such as a photographic image having a characteristic gradation, pattern, or texture. It is an object of the present invention to provide a similar image search device and a similar image search method that improve accuracy.

[0014]

SUMMARY OF THE INVENTION The present invention has been made to achieve the above object. That is, the similar image search device of the present invention includes an image input unit that inputs image data, an image storage unit that stores image data, a plurality of feature amount extraction units that extracts a plurality of types of feature amounts from the image data, A plurality of types of feature amounts extracted by a plurality of feature amount extracting means in the image data input by the image input means;
A similarity calculating unit for comparing each of the plurality of types of feature amounts extracted by the plurality of feature amount extracting units in the image data stored in the image storing unit to calculate a similarity between the two image data; Based on the similarity calculated by the means,
An image search unit that searches the image storage unit for image data similar to the image data input by the image input unit.

Further, the similar image retrieval method of the present invention includes a step of extracting a plurality of types of feature amounts from accumulated image data, a step of extracting a plurality of types of feature amounts from the input image data, and Comparing each of the plurality of types of feature amounts extracted from the stored image data with the plurality of types of feature amounts extracted from the stored image data, and calculating a similarity between the two image data. Based on the similarity,
Searching the stored image data for image data similar to the input image data.

According to the present invention, a plurality of types of features are extracted from the input image data, and various features of the image data are comprehensively compared with the plurality of types of features extracted from the stored image data. The similarity calculated in consideration of the above is calculated. As a result, even if the image has characteristics such as gradation, pattern, and texture such as a photographic image, image data similar to the input image data can be effectively searched from the stored image data.

[0017]

Embodiments of the present invention will be described below in detail with reference to the drawings. The similar image search device described in detail in the present embodiment is a system that searches for an image similar to an image specified by a user from a plurality of images registered in advance, and presents the result to the user.

The image specified by the user includes, for example, a photograph printed on photographic paper, image data photographed by a digital camera, an image recorded on a CD-ROM such as a photo CD, or via a network. Image data obtained by performing the above operation.

FIG. 1 is a block diagram illustrating a similar image retrieval apparatus according to the present embodiment. This similar image search device 1
Represents an image input unit 2, a feature amount extraction unit 3, a feature amount search unit 4,
It comprises a search result display unit 5, an image data storage unit 6, and a feature data storage unit 7, all of which are connected by a data bus B. On the data bus B, image data, feature amount data, and control signals are exchanged.

The image input unit 2 has a function of reading a photograph or the like and converting it into image data, like a scanner. Furthermore, a memory card on which image data taken by a digital camera is recorded, and a CD-RO on which image data is recorded
It has a function of reading image data from a storage medium such as M.

The image data storage unit 6 is a device such as a hard disk (HDD) that can store a large amount of data, and can store a large amount of image data.
The feature data storage unit 7 is also a device such as an HDD that can store a large amount of data. The image data storage unit 6 and the feature data storage unit 7 can be used by dividing the same HDD.

The feature amount search unit 4 searches the image data stored in the feature amount data storage unit 7 for data having a feature amount similar to the feature amount of the image serving as a search key, and stores the stored data. Are ranked according to the similarity. Details of this operation will be described later.

The search result display unit 5 can display not only characters but also image data and the like like a computer display. In this search result display section 5,
The search result is displayed based on the ranking result according to the similarity obtained from the feature amount search unit 4.

FIG. 2 shows a block diagram of the feature amount extraction unit.
The feature amount extraction unit 3 has two main functions.
A normalization processing unit and a feature amount extraction processing unit; The feature amount extraction processing unit performs a normalization process as a first stage on the received image data, and performs a feature amount extraction process as a second stage.

The normalization processing section includes a resolution conversion processing circuit 31
And a color space conversion processing circuit 32. First, in the resolution conversion processing circuit 31, the resolution of the image data is converted.

The vertical and horizontal sizes of image data to be registered or image data serving as search keys generally take various values depending on the image data. The purpose of the resolution conversion processing circuit 31 is to adjust the vertical and horizontal sizes of the image to a certain value before extracting the feature amount.

The constant value is (horizontal × vertical) = (512 pixels × 512 pixels) or (horizontal × vertical) = (1024 pixels × 1)
024 pixels).

In this embodiment, a value of (horizontal × vertical) = (512 pixels × 512 pixels) is employed. For example, the size of the image data read by the image input unit 2 is (horizontal × vertical) =
In the case of (4200 pixels × 5500 pixels), the image size is (horizontal × vertical) by multiplying the horizontal direction by 512/4200 = 0.1219 times and the vertical direction by 512/5500 = 0.0931 times. ) = (512 pixels × 512 pixels).

The color space conversion processing circuit 32 converts the color space of the image data whose resolution has been converted by the resolution conversion processing circuit 31. The color space of the image data to be registered or the image data serving as a search key takes various values. For example, RG
B, YMCK, YUV, YMCK, L ^* a ^* b ^* , and the like. The color space conversion processing circuit 32 aims to convert various color spaces into a predetermined color space.

[0030] The color space in this embodiment into a ^{L * a * b *, L} * / a * / b * is expressed by the data of 8bil / pixel for each.

The feature quantity extraction processing section comprises a sub-block division circuit 33, a contour feature extraction circuit 34, a frequency distribution feature extraction circuit 35, a color count circuit 36, and an image attribute information data collection circuit 37.

The feature quantity extraction processing section receives the image data normalized by the normalization processing section, extracts feature quantities from the normalized image, and collects attribute information data from the image data before normalization. The main purpose is.

In the feature amount extraction processing section, the attribute information data of the image data before normalization is extracted by the image attribute information data collecting circuit 37. The attribute information data of the image data includes the file name of the image data, the date and time when the image data was created, the vertical and horizontal sizes (the number of pixels) of the image data, the number of bytes of the image data, and the like. In addition, information of an identification number that allows the image data to be distinguished from already stored image data is also collected.

The normalized image data is first input to the sub-block dividing circuit 33. FIG. 3 shows a conceptual diagram of sub-block division. In the sub-block division circuit 33,
Divide the normalized image into multiple blocks.

Various numbers can be considered for the number of blocks to be divided. In this embodiment, one piece of image data is
Is divided into a total of 16 × 16 = 256 blocks.

The image size is set to 51 by normalizing the image.
Resolution conversion to 2 pixels x 512 pixels, and 16 pixels x 1
Since the sub-block is divided into six sub-blocks, the pixel size of the sub-block is 32 pixels × 32 pixels.

The image data divided into sub-blocks is
The contour line feature extraction unit 34, the frequency distribution feature extraction unit 35, and the number-of-colors count unit 36 are input to each sub-block, and feature amounts such as contour lines, the number of colors, and frequency components are extracted. By dividing into sub-blocks and extracting each feature amount for each sub-block, a feature amount reflecting the distribution of image data can be extracted.

FIG. 4 is a block diagram of the contour line feature extraction unit. Image data for each sub-block is represented by a color space L ^* a ^* b ^* . Therefore, in order to extract contour features for each sub-block, a contour image creating circuit 34a and a contour feature extracting circuit 34b are provided for each L ^* / a ^* / b ^* , and each contour feature is extracted.

Where L^*Contour image creation circuit for images
34a, L as shown in FIG.^*Image contours
Create an image. FIG.^*Original image of Fig. 5
(B) is L ^*FIG.

In this embodiment, the L ^* image is 8 bits / p
pixel, and is represented as a 256-level grayscale image. The contour image creating circuit 34a divides the 256 shades of light and shade at equal intervals to create a contour image. In this embodiment, the contour image is created by dividing the image at 16-value intervals.

FIG. 6 shows an example of the contour line creation processing. Here, an example in which a contour image is created from a gradation image as shown in FIG. For example, the gradation image is cut along the axis AA ′, the density is plotted on the vertical axis, and the axis A− is plotted on the horizontal axis.
When the position on A ′ is plotted, a graph as shown in FIG. 6B is obtained. The vertical axis of this graph is divided at 16 value intervals,
The position where the intersection with the plot of the graph is lowered is the position where the contour line is drawn, and as a result, a contour image as shown in FIG. 6C is obtained.

The contour image obtained by the contour image creating circuit 34a is sent to a contour feature extracting circuit 34b. The contour feature extraction circuit 34b extracts the features of the drawn contour from the contour image created for each sub-block. The features of the contour lines to be extracted include the density of the contour lines occupying the sub-blocks, the drawing direction of the contour lines, and the like.

The density of the contour lines is obtained, for example, by the ratio of the number of pixels of the pixel values representing the contour lines in the block, and the drawing direction of the contour lines is obtained, for example, by the ratio of the number of inversions of the pixel values in the vertical and horizontal directions in the block.

[0044] The contour line feature extraction processing a ^* image, b ^* is also carried out similarly to the L ^* image to image, respectively extracts a feature quantity.

FIG. 7 shows a block diagram of the frequency distribution feature extraction unit for the image. Image data for each sub-block is represented by a color space L ^* a ^* b ^* . Therefore, to extract the characteristics of the frequency components of the image for each sub-block, L ^* / a ^*
For each / b ^* , a orthogonal transform circuit 35a and a frequency distribution analysis circuit 35b are provided, and each of them extracts a feature.

Hereinafter, the feature extraction from the L ^* image will be described in the same manner as the contour feature extraction unit. In the present embodiment, the size of the sub-block image is 32 pixels × 32 pixels. The orthogonal conversion circuit 35a performs orthogonal conversion processing on the image having the size of 32 pixels × 32 pixels to generate data obtained by converting the sub-block image into frequency components. DCT transformation or the like is used for orthogonal transformation.

The obtained frequency component data is input to the frequency component analysis circuit 35b, and the characteristics of the distribution state of the frequency component data are extracted.

The frequency component analysis circuit 35b scans zigzag as shown in FIG. 8 and converts it into a one-dimensional frequency data string. Further, since the frequency component data takes positive and negative values, the value is converted to the absolute value of the component value.

FIG. 9 is a graph showing an example of the frequency component values obtained in this manner. The frequency distribution analysis circuit 35b extracts such a graph-shaped feature.

The characteristics of this frequency distribution graph include, for example, the total value of the entire frequency distribution as shown in FIG. 10A and the total value of only the high frequency components of the frequency distribution as shown in FIG. And a total value of low frequency components among the frequency components as shown in FIG. Further, the ratio of the total value of the high frequency components to the total value of the entire frequency distribution can be extracted as the feature amount.

This frequency component feature extraction processing is performed by using an a ^* image,
The same process is performed for the b ^* image as for the L ^* image, and the feature amount is extracted for each.

The color count section 36 counts the number of colors included in the sub-block image. For example, the number of colors of the sub-block image as shown in FIG.
The number of colors is "256" because the image of (b) is a gradation including a very large number of colors.

The number of colors of the image shown in FIG. 11C is "3". In this way, the number of colors in the sub-block is counted, and the value of the number of colors is a feature of the sub-block image.

The plurality of feature quantities extracted as described above are combined into one data pack in the feature quantity packing circuit 38. This data packing is sent to and stored in the feature data storage unit 7, and is extracted from the feature data storage unit 7 as needed.

The similar image retrieval apparatus of this embodiment has a function of registering and storing a large number of images. FIG. 12 shows a flowchart of the operation for registering this image. First, an image to be registered is set in the image input unit 2 and image data is read (step S101). The read image data is transferred to the image data storage unit 6 via the data bus B.

The image data storage section 6 stores the received image data (step S102). Here, when storing the image data, the image data is stored with an identification number that can be distinguished from the already stored image data.

The image input unit 2 sends out the image data to the image data storage unit 6 and also sends out the image data to the feature amount extraction unit 3 via the data bus B.

In the feature quantity extraction unit 3, first, the image input unit 2
A normalization process is performed on the image data received from (step S103), and attribute information data of the image data is collected (step S104). This attribute information data also includes the identification number described above.

Next, a process of extracting a feature value from the normalized image is performed (step S105). The extracted feature value and attribute information data are stored together in the feature value data storage unit 7 as one data pack (step S106).

With the above operation, the image registration operation in the similar image search device 1 of the present embodiment is completed.

Next, a similar image search method using the similar image search device 1 of the present embodiment will be described. FIG. 13 shows a flowchart of the similar image search method.

First, an image serving as a search key (hereinafter, referred to as a "search key image") is set in the image input unit 2 and image data is read (step S201).

The read image data is sent to the feature extracting unit 3 via the data bus B. The feature amount extraction unit 3 first performs a normalization process on the image data received from the image input unit 2 (step S20).
2) Collect attribute information data of the image data (step S203).

Next, a process of extracting a characteristic amount from the normalized image is performed (step S204). The extracted feature amounts are combined with the attribute information data into one data pack and sent to the feature amount search unit 4.

The feature amount search unit 4 holds the feature amount data of the received search key image, and stores image data (hereinafter referred to as “registered image”) registered in advance from the feature amount data storage unit 7. The feature data is read (step S205).

Next, the feature amount search unit 4 calculates the similarity between the individual feature amount data of the registered image read from the feature amount data storage unit 7 and the feature amount data of the search key image (step S206). ). Then, the stored sorted images are sorted in the descending order of the similarity of the feature amount data, and the result is output to the search result display unit 5 via the data bus B (step S207).

The search result display unit 5 displays a registered image similar to the search key image based on the information of the feature amount data received from the feature amount search unit 4 and sorted in descending order of similarity. The value of the similarity calculated at the time of display, the attribute information data, and the like are displayed. As a result, registered images similar to the search key image are searched, and displayed in the order of similarity.

Next, a method of calculating the similarity performed by the feature value search unit 4 will be described in detail. The similar image search device of the present embodiment searches for a similar image based on the feature amount of image data. As described above, a plurality of feature values such as contour features, frequency component features, and the number of colors included in a sub-block are used for each sub-block image as described above. The similar image search device of the present embodiment calculates the similarity based on the plurality of feature amounts.

Here, it is assumed that the search key image is IMGk and the line image is IMGn. n is a number for identifying an image registered in the similar image search device of the present embodiment. When the total number of registered images is N, n can take a value from 0 to N-1.

The value of the feature value extracted from each image data is defined as Fm. m is a number for identifying a plurality of feature amounts. If the total number of feature amounts to be extracted from one image is M, m can take a value from 0 to M-1. here,
The feature value Fm of the image IMGn

IMGnFm

Are expressed as follows. For example, the m-th feature amount of the search key image is

IMGkFm

Is expressed as follows. In the following description, the following values are used as the feature values.

Contour density of L ^* component: F0 A Contour density of A ^* component: F1 B ^* Contour density of component: F2 Ratio of high frequency component of L ^* component: F3 Ratio of high frequency component of A ^* component : F4 B ^* Ratio of high frequency component: F5 Number of colors included in sub-block: F6

One value is obtained from one sub-block for each feature amount from F0 to F6. One image is 16
× 16 = 256 subblocks,
For example, if the feature value is F0 (the density of contour lines of the L ^* component), the feature value F0 of one image is composed of 256 values. Therefore, the feature value F0 can be regarded as a 256-dimensional vector.

The same applies to the other feature values F1 to F6, which can be regarded as 256-dimensional vectors. The feature values F0 and I of the registered n-th image
It can be said that MGnF0 is a 256-dimensional vector.

When calculating the similarity between the search key image and the registered image, the similarity is calculated for each feature, and the sum of the similarities for each feature is added to the registered image. The similarity with the search key image is used.

Search key image IMGk and registered image IMGn
And the similarity in the feature value Fm is IMGnRm,
Assuming that the final similarity between the search key image IMGk and the registered image IMGn is IMGnT,
The similarity between the search key image IMGk and the registered image IMGn can be calculated.

[0080]

(Equation 1)

The search key image IMGk and the registered image I
The method of calculating the similarity IMGnRm in the feature value Fm with MGn is based on the following Expression 2.

[0082]

(Equation 2)

Here, IMGnDm shown in Expression 2 is the distance between the feature value Fm of the search key image IMGk and the feature value of the registered image IMGn. Since IMGkFm and IMGnFm are 256-dimensional vectors, the absolute value of the distance between these two vectors is IMGnDm.

Assuming that these two vectors are IMGkFm = (a0, a1,..., A255) IMGnFm = (b0, bl,..., B255), the distance IMGnDm can be expressed by the following equation (3).

[0085]

(Equation 3)

IMGDmax shown in Expression 2 is the maximum value of the distance between the feature value Fm of the search key image and Fm of all the line images. N is the number of registered images. AR is a coefficient corresponding to the angle β formed by the 256-dimensional vectors IMGkFm and IMGnFm.

When the angle β is smaller than the threshold value th set in advance, β = 1, and when the angle β is larger than the threshold value th, the constant value β = C (c> 1) is taken.

Thus, even if the absolute value and the distance of the vector are small, if the direction is different, the calculated distance can be increased, and the similarity can be reduced to match the apparent similarity determination. become able to.

Wm is a weighting factor that can be set in advance for each feature value. For example, when it is desired to exclude the influence of the number of colors for each sub-block from the search result, a small value is set to the value of W6, which is the weight coefficient of the feature amount of the number of colors.

In this embodiment, "100" is set to all weighting factors as default values. To eliminate the influence of the feature, a value smaller than “100” is set. To emphasize the influence of the feature, “100” is set.
Set a value larger than.

By the above method, it is possible to determine the similarity between the search key image and the registered image from various angles and calculate the similarity that matches the apparent similarity.

[0092]

As described above in detail, according to the present invention, a similar image retrieval apparatus which effectively retrieves an image having a characteristic in gradation, pattern, and texture such as a photographic image, and improves retrieval accuracy, It is possible to provide a similar image search method.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a similar image search device according to an embodiment.

FIG. 2 is a block diagram illustrating a feature amount extraction unit.

FIG. 3 is a conceptual diagram of sub-block division.

FIG. 4 is a block diagram illustrating a contour line feature extraction unit.

FIG. 5 is a diagram illustrating an example of a contour image.

FIG. 6 is a diagram illustrating an example of a contour line creation process.

FIG. 7 is a block diagram illustrating a frequency distribution feature extraction unit.

FIG. 8 is a diagram illustrating a zigzag scan.

FIG. 9 is a diagram illustrating an example of a frequency component value.

FIG. 10 is a diagram illustrating an example of frequency component analysis.

FIG. 11 is a diagram illustrating an example of color count.

FIG. 12 is a flowchart illustrating an image registration operation.

FIG. 13 is a flowchart illustrating a similar image search method according to the present embodiment.

FIG. 14 is a diagram illustrating the distance between feature amount vectors.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Similar image search device, 2 ... Image input part, 3 ... Feature amount extraction part, 4 ... Feature amount search part, 5 ... Search result display part, 6 ... Image data storage part, 7 ... Feature amount data storage part, 31 ... resolution conversion processing circuit, 32 ... color space conversion processing circuit, 33 ... sub-block division circuit, 34 ... contour line feature extraction unit, 35 ...
Frequency distribution feature extraction unit, 36 ... Color count unit, 37 ...
Image attribute information data collection circuit, 38 ... feature quantity packing circuit

Claims (21)

[Claims] An image input unit that inputs image data; an image storage unit that stores image data; a feature amount extraction unit that extracts a plurality of types of feature amounts from the image data; A plurality of types of feature amounts extracted by the feature amount extraction unit in the image data are compared with a plurality of types of feature amounts extracted by the feature amount extraction unit in the image data stored in the image storage unit. A similarity calculating unit that calculates a similarity between the two image data; and storing the image data similar to the image data input by the image input unit based on the similarity calculated by the similarity calculating unit. A similar image search device comprising: an image search unit that searches from a unit. 2. The feature amount extracting unit extracts a plurality of types of feature amounts for each block obtained by dividing image data into a plurality of blocks, and the similarity calculating unit includes a plurality of types of extracted feature data for each block. 2. The similar image search device according to claim 1, wherein the similarity between the two image data is calculated based on the characteristic amount of the image data. 3. Image input means for inputting image data; image storage means for storing image data; first feature amount extraction means for extracting a feature of a density gradient from the image data; A second feature value extracting means for converting and extracting a feature of a frequency component; a third feature value extracting means for extracting a color feature of image data; and A first feature amount comparing unit that compares the feature amount extracted by the one feature amount extracting unit with the feature amount extracted by the first feature amount extracting unit from the image data stored in the image storing unit; And a feature value extracted by the second feature value extraction means from the image data input by the image input means; and a second feature value from the image data stored in the image storage means. A second feature value comparison unit that compares the feature value extracted by the extraction unit; a feature value extracted by the third feature value extraction unit from the image data input by the image input unit; A third feature value comparison unit that compares the feature value extracted by the third feature value extraction unit from the image data stored in the storage unit; a result of the first feature value comparison unit; Referring to the result of the second feature value comparison means and the result of the third feature value comparison means, the image data input by the image input means and the image data stored in the image storage means are compared. A similarity calculating unit configured to calculate a similarity; and a search result displaying image data similar to the image data input by the image input unit, based on the similarity calculated by the similarity calculating unit. Similar image retrieval apparatus characterized by comprising a result displaying means. 4. The method according to claim 1, wherein the first feature amount extracting unit creates a contour image from the image data, and extracts a feature of a density gradient of the image from a distribution state of the contour lines in the contour image. 3. The similar image search device according to 3. 5. The apparatus according to claim 1, wherein the second feature quantity extracting means converts the image data into frequency component data using orthogonal transform means, and extracts a feature of the distribution state of the frequency component data. 3. The similar image search device according to 3. 6. The similar image search device according to claim 3, wherein said third feature amount extracting means calculates the number of types of colors included in the image data. 7. The similarity calculation means based on a comparison result of the first feature quantity comparison means and a coefficient indicating importance of the feature quantity extracted by the first feature quantity extraction means. The similar image search device according to claim 3, wherein the degree is calculated. 8. The similarity calculation means based on a comparison result of the second feature quantity comparison means and a coefficient indicating importance of the feature quantity extracted by the second feature quantity extraction means. 4. The similar image search device according to claim 3, wherein the degree is calculated. 9. The similarity calculation means based on a comparison result of the third feature quantity comparison means and a coefficient indicating importance of the feature quantity extracted by the third feature quantity extraction means. The similar image search device according to claim 3, wherein the degree is calculated. 10. The method according to claim 1, wherein the similarity calculating unit calculates a value of a comparison result obtained from the first feature amount comparing unit, a value of a comparison result obtained from the second feature amount comparing unit, 4. A value obtained by adding a value of a comparison result obtained from the feature amount comparing means to a similarity is calculated.
A similar image search device as described. 11. The first feature value comparing means calculates a value of the feature value to be compared by the first feature value extracting means based on a statistically obtained maximum value of the feature value. 4. The comparison according to claim 3, wherein
A similar image search device as described. 12. The second feature value comparing means calculates a value of the feature value to be compared by the feature value extracting means of the second measure based on a statistically obtained maximum value of the feature value. 4. The comparison according to claim 3, wherein
A similar image search device as described. 13. The third feature value comparison means calculates a feature value calculated by the third feature value extraction means to be compared based on a maximum value of the feature value statistically obtained. 4. The comparison according to claim 3, wherein
A similar image search device as described. 14. The first feature value comparison means refers to an angle value constituting a vector, which is calculated from a feature value value calculated by the first feature value extraction means to be compared. The similar image search device according to claim 3, wherein the comparison is performed. 15. The second feature value comparison means refers to an angle value constituting a vector, which is calculated from the feature value value calculated by the second feature value extraction means to be compared. The similar image search device according to claim 3, wherein the comparison is performed. 16. The third feature value comparison means refers to an angle value constituting a vector, which is calculated from a feature value value calculated by the third feature value extraction means to be compared. The similar image search device according to claim 3, wherein the comparison is performed. 17. A step of extracting a plurality of types of feature amounts from accumulated image data, a step of extracting a plurality of types of feature amounts from input image data, and a plurality of steps extracted from the input image data. Calculating the similarity between the two types of image data by comparing each of the types of characteristic amounts with the plurality of types of characteristic amounts extracted from the stored image data; and, based on the calculated similarity, Retrieving image data similar to the input image data from the stored image data. 18. The method according to claim 18, wherein in the step of extracting a plurality of feature amounts, a plurality of types of feature amounts are extracted for each block obtained by dividing the image data into a plurality of pieces. 18. The similar image search device according to claim 17, wherein a similarity between the two pieces of image data is calculated based on a plurality of types of feature amounts extracted in (1). 19. A step of extracting a plurality of types of feature amounts from accumulated image data, a step of extracting a plurality of types of feature amounts from input image data, and a plurality of types extracted from the input image data. Calculating the similarity between the two types of image data by comparing the respective types of characteristic amounts with the plurality of types of characteristic amounts extracted from the stored image data, based on the calculated similarity. Retrieving image data similar to the input image data from the stored image data. 20. In the step of extracting a plurality of types of feature amounts, a plurality of types of feature amounts are extracted for each block obtained by dividing the image data into a plurality of pieces. 20. The similar image search method according to claim 19, wherein a similarity between the two image data is calculated based on a plurality of types of feature amounts extracted for each. 21. An image inputting step of inputting image data, an image storing step of storing image data, a first feature amount extracting step of extracting a feature of a density gradient from the image data, and converting the image data into frequency components. Converting, extracting a feature of a frequency component, extracting a feature of a color of the image data, extracting a third feature, and extracting the second feature from the image data input in the image input step. A first feature value comparing step of comparing the feature value extracted in one feature value extracting step with the feature value extracted in the first feature value extracting step from the image data stored in the image storing step; And a feature amount extracted from the image data input in the image input step in the second feature amount extraction step, and the second feature from the image data stored in the image storage step. A second feature value comparison step of comparing the feature value extracted in the quantity extraction step, and a feature value extracted in the plexus 3 feature value extraction step from the image data input in the image input step, A third feature amount comparing step of comparing the feature amount extracted by the third feature amount extracting step from the image data accumulated in the image accumulating step, and a result of the first feature amount comparing step; With reference to the result of the second feature value comparison step and the result of the third feature value comparison step, the image data input in the image input step and the image data stored in the image storage step Displaying a search result of image data similar to the image data input in the image input step based on the similarity calculated in the similarity calculation step. Image retrieval method characterized by comprising: a search result display step.