JP2004094379A - Similar image retrieval device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image retrieval device highly precisely detecting whether images having similar compositions are similar or not. <P>SOLUTION: This similar image retrieval device selects an image out of a plurality of images, divides it into a plurality of image regions based on a prescribed division number, produces a regional value, or information on color in the divided image region, produces difference value data by calculating a difference between a regional value produced on a reference image region, or the image region to be a reference out of the divided image regions with a regional value produced in an object image region, or the image region different from the reference image region, and compares the difference value data on the reference image to be the reference for detecting the similarity out of the plurality of images, with the difference value data produced on a comparison image to be an object of detecting the similarity to the image to be the reference image so as to detect whether the images are coincident or not. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a similar image search device that searches for similar images.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, when searching for similarity between multimedia contents such as images using a computer, a method of making a determination using color information of an image is generally performed. When a human recognizes an image, the image is recognized using the color itself, a change in color, the shape of the image, and the like.
When judging the similarity of the images, comparing arbitrary parts in the images, or comparing the whole image of one image with an arbitrary part of the other image, comparing the whole two images There are cases.
[0003]
[Problems to be solved by the invention]
When detecting whether or not there is similarity by comparing the whole images of the images, unlike similarity search for individual objects in the images, it is considered that similarity in the entire composition is similar. Become a reference.
However, when the conventional method is used, there are cases where the compositions are similar but cannot be regarded as similar, or where the compositions are completely different but are regarded as similar.
[0004]
For example, there is an absolute value comparison method for comparing color information itself of two images. In this method, two images are compared one pixel at a time for color values to detect whether or not the images have commonality. However, in this method, since the color values are compared, there is a problem that an image having a different color tone cannot be detected as a similar content even if the content has a visually similar composition.
[0005]
On the other hand, as shown in FIG. 6, there is a conversion comparison method in which features are extracted from two images by a common method and the features are compared. In this method, there is a method of expressing a color value of an image by a spectrum or a histogram feature and comparing them (codes (a) and (b)). In addition, there is a method in which a keyword is set in the image itself, which is a feature, and the keyword is compared (code (c)).
[0006]
However, in the method using a spectrum or a histogram, spatial distribution information of pixel values, which is indispensable for identifying an image, is excluded. Even if an image has almost the same composition, similar content is obtained when the color tone is different. Although there is a case where the image is not detected as an image and the color tone is common, an image having a completely different composition may be detected as a similar content, and there is a problem that the detection accuracy is reduced. Also, in the method using keywords, when setting a keyword, it is difficult for a worker to set the keyword by looking at the image, and it is difficult to fully automate the keyword. Is reduced.
[0007]
The present invention has been made in view of such circumstances, and an object of the present invention is to provide an image search device capable of detecting, with higher accuracy, whether images having similar compositions are similar or not. It is in.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, the present invention provides a similar image search device for comparing a plurality of images, wherein one of the plurality of images is selected, and a plurality of images are selected based on a predetermined number of divisions. A division unit that divides the image region into a plurality of image regions; a region value calculation unit that generates a region value that is information on a color in the image region divided by the division unit; A region value generated by the region value calculation unit for a reference image region that is an image region to be, and a region value generated by the region value calculation unit for a target image region that is an image region different from the reference image region , A difference value calculation unit that calculates difference between the two images, and a difference image calculation unit for a reference image serving as a reference for detecting whether or not the images are similar among the plurality of images. The difference value data generated is compared with the difference value data generated by the difference value calculation unit for a comparison image to be detected as to whether or not the image is similar to the reference image. And a output unit for outputting a detection result detected by the difference value comparing unit.
[0009]
Further, according to the present invention, in the above-described similar image search device, the division unit determines an image size detection unit that detects a size of the selected image and a size of the image detected by the image size detection unit in advance. And an image size normalizing unit for changing the image size to the predetermined size, and dividing the image after being changed by the image size normalizing unit.
[0010]
The present invention also provides the above-described similar image search device, further comprising a difference value database that stores difference value data generated by the difference value calculation unit for each image, wherein the difference value comparison unit compares the images. In response to the instruction, the difference value data stored in the difference value database is read, and whether or not the images match is detected.
[0011]
Further, the present invention is a similar image search method in a similar image search device for comparing a plurality of images, wherein one image is selected from the plurality of images and a plurality of images are selected based on a predetermined number of divisions. Dividing into image regions, generating region values that are information on colors in the divided image regions, and generating region values for a reference image region that is a reference image region among the divided image regions. And a difference value between the reference image region and the region value generated for the target image region, which is an image region different from the reference image region, to generate difference value data. The difference value data for a reference image serving as a reference for detecting the difference image data, and the difference value data generated for a comparison image to be detected whether or not the image is similar to the reference image are compared. And detecting whether or not they match each other.
[0012]
The present invention is also a similar image search program for comparing a plurality of images, wherein one of the plurality of images is selected and divided into a plurality of image regions based on a predetermined number of divisions. Generating an area value that is information about a color in the divided image area; and, among the divided image areas, an area value generated for a reference image area that is a reference image area. Calculating a difference between an area value generated for a target image area that is an image area different from the reference image area to generate difference value data; and determining whether or not the plurality of images are similar. The difference value data for a reference image serving as a reference for detecting the difference between the reference image and the difference value data generated for a comparison image serving as a target for detecting whether or not the image is similar to the reference image. When, wherein the comparison image to execute the steps of detecting whether matches, to computer.
[0013]
Further, the present invention is a computer-readable recording medium recording a similar image search program for comparing a plurality of images, wherein one of the plurality of images is selected and the number of divisions is specified in advance. Dividing the image area into a plurality of image areas, generating an area value that is information on a color in the divided image area, and setting a criterion that is a reference image area among the divided image areas. Calculating a difference between an area value generated for the image area and an area value generated for the target image area, which is an image area different from the reference image area, to generate difference value data; Among the images, difference value data for a reference image serving as a reference for detecting whether or not the images are similar to each other, and a pair for detecting whether or not the images are similar to the image serving as the reference image. Wherein the comparison image and the step of detecting whether matching was recorded similar image retrieval program for causing the computer to execute, and the difference value data generated for comparison image becomes.
[0014]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, a similar image search device according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a schematic block diagram showing a configuration of a similar image search device 1 according to an embodiment of the present invention.
In this figure, an image data DB (database) 10 stores various image data. The dividing unit 11 selects one image from among a plurality of images to be compared among the image data stored in the image data DB 10, and selects a plurality of image regions (hereinafter, “divided blocks”) designated in advance. ). In this division processing, images to be compared are sequentially selected, and division processing is performed for each of the selected images. The division unit 11 includes an image size detection unit 12 that detects the size of the selected image, and an image size normalization unit that changes the size of the image detected by the image size detection unit 12 to a predetermined size. And an image division unit that divides the image after being changed by the image size normalization unit.
[0015]
The area value calculation unit 14 generates, for each of the divided blocks, an area value that is information on a color in the divided block. The information about the color is, specifically, an average color value which is an average value of color values in the divided block, an average luminance value which is an average value of luminance values in the divided block, and a color value of a representative pixel in the divided block. Alternatively, a luminance value or the like is included. The divided block arrangement unit 15 arranges the divided blocks divided by the division unit 11 in a predetermined order.
[0016]
The difference value calculation unit 16 is configured to output, from among the divided blocks divided by the division unit 11, a region value generated by the region value calculation unit 14 for a reference divided block that is a reference divided block, and a divided block different from the divided block. Then, a difference between the target divided block and the region value generated by the region value calculation unit 14 is calculated to generate difference value data. The difference value calculation unit 16 creates a difference distribution file from the calculated difference value data, and stores the created difference distribution file in the difference distribution file database 20. The difference distribution file database 20 stores the difference distribution file output from the difference value calculation unit 16 for each image.
[0017]
The difference distribution file reading unit 32 reads a difference distribution file of an image to be compared from the difference distribution file database 20 based on an instruction from the difference value comparison unit 33. The input unit 31 indicates image data to be searched.
The difference value comparison unit 33 compares the difference value data generated by the difference value calculation unit 16 for a reference image serving as a reference for detecting whether or not the images are similar to each other, and an image serving as the reference image. The difference value data generated by the difference value calculation unit 16 is compared with a comparison image to be detected as to whether or not they are similar to each other, and whether or not the images match is detected. The output unit 35 is an output device such as a liquid crystal display device or a CRT (Cathode Ray Tube), and outputs a detection result detected by the difference value comparison unit 33. Note that data indicating the detection result may be output to another device.
[0018]
Next, the processing operation of the similar image search device 1 in FIG. 1 will be described with reference to the flowchart in FIG. 2 and the explanatory diagram in FIG.
1) The difference distribution file generation processing division unit 11 selects image data for which a difference distribution file is to be created from the image data stored in the image data DB 10 (step S1 in FIG. 2), and selects the selected image data The size of the image in the vertical and horizontal directions is detected by the image size detection unit 12 (step S2 in FIG. 2). Then, the dividing unit 11 detects whether the detected image size is a predetermined size by the image size normalizing unit 13, and when the detected image size is not the predetermined size, the image size normalizing unit 13 The image size is changed to a predetermined size and normalized (step S3 in FIG. 2). On the other hand, if the size is a predetermined size, the image size is not changed in step S3.
[0019]
Next, the dividing unit 11 divides the normalized image data into a predetermined number (step S4 in FIG. 2, reference numeral (a) in FIG. 3), and assigns an area identification code to the divided blocks. (Reference numeral (b) in FIG. 3). When the image is divided, the region value calculation unit 14 calculates the region value of the region for each of the divided blocks (Step S5 in FIG. 2), and associates the calculation result with the region identification code. The calculation result of the region value calculated here is an average value of the luminance values in the region.
The divided block arrangement unit 15 arranges the divided blocks in a predetermined order. This array is arranged in the order of the area identification codes (FIG. 3C). At this time, the area value of the divided block also remains associated with the area identification code of the arranged divided block.
[0020]
When the divided blocks are arranged, the difference value calculation unit 16 calculates difference between the area values between adjacent divided blocks in the divided blocks of the array to generate difference value data (Step S7 in FIG. 2). Then, a difference distribution file is created from the generated difference value data (Step S8 in FIG. 2), and the created difference distribution file is stored in the difference distribution file database 20 (Step S9 in FIG. 2). After the difference distribution file data is stored in the difference distribution file database 20, the division unit 11 detects whether or not the division processing has been performed on all the image data among the image data stored in the image data DB 10 ( Step S10 in FIG. 2) is repeated to perform the division process on all the image data (reference numeral (d) in FIG. 3). Thereby, the difference distribution file is created for all the images in steps S2 to S10.
[0021]
2) Image Search Process Next, the image search process will be described with reference to the flowchart of FIG.
When the image data serving as a reference for searching for similarity is specified from the input unit 36, the difference distribution file reading unit 32 reads the difference distribution file data of the reference image data from the difference distribution file database 20 (step S21). If there is no difference distribution file data of the reference image data, an instruction to execute steps S2 to S9 in FIG. 2 is output to the dividing unit 11, and a difference distribution file is created.
[0022]
Next, when the difference distribution file data of the reference image data is read, the difference value comparison unit 33 determines an allowable error value (Step S22). Initially, “0” is set.
[0023]
Next, the difference distribution file reading unit 32 reads a difference distribution file of the comparison target image data to be compared with the reference image data from the difference distribution file database 20 (Step S23).
The difference value comparison unit 33 compares the difference distribution file of the reference image data with the difference distribution file of the comparison target image data, and detects whether or not the allowable error is within a range of “0” (Step S24). That is, it is detected whether or not the difference value data matches.
[0024]
When the difference is outside the allowable error range (here, when they do not match), the difference value comparison unit 33 ends the comparison of all the image data stored in the difference distribution file database 20 with the difference distribution file. It is detected whether or not the process has been completed (step S26). If the process has not been completed, the process proceeds to step S23, and the difference distribution file reading unit 32 sets the difference distribution file of the other image data as the difference distribution file of the comparison target image data. ,read out.
On the other hand, if it is determined in step S24 that the difference is within the allowable error range (here, the values match), the difference value comparison unit 33 registers the image data to be compared as a match candidate in a predetermined memory (step S24). S25), and proceed to step S26.
[0025]
When the comparison between the difference distribution file data of the reference image data and the difference distribution file data of all the image data is completed, the difference value comparison unit 33 detects whether or not the number of matching candidates is one or more. (Step S27). If the number of matching candidates is one or more, the search ends (step S28), and a list of matching candidates is output to the output unit 35 (step S29).
[0026]
On the other hand, if there is no match candidate in step S27, the difference value comparison unit 33 expands the range of the allowable error within the range of the allowable error set from the allowable error storage unit 31 (step S30). Here, since the permissible error is increased by 1, “0” is set to “1”, and the process proceeds to step S22.
[0027]
Next, the difference distribution file generation processing and the image search processing described in FIGS. 2 and 4 will be further described with reference to FIG. The image A (code (a)) serving as the reference image data and the image B (code (b)) serving as the comparison target image data are divided into a plurality of divided blocks by the division unit 11 (code (c)). Then, the area values of the divided blocks are calculated by the area value calculation unit 14 (code (d)), arranged in a predetermined order by the divided block arrangement unit 15, and then divided by the difference value calculation unit 16. Is calculated for each adjacent divided block (code (e)), a difference distribution file as a calculation result is generated (code (f)), and the generated difference distribution file is compared with the difference value comparison unit. 33 (sign (g)). Here, when the error range is “0”, the difference value (code (f)) calculated based on the area value of the divided block of 1 row and 5 columns matches in the image A and the image B, respectively. Are detected as similar images and output to the output unit 35.
[0028]
On the other hand, when the miscalculation range is “1” and the difference value data of the difference distribution file is “11, 29, 46, −24”, the first difference value data of the difference value data of the comparison target image data is In the range of 11 ± 1, the second difference value data is in the range of 29 ± 1, the third difference value data is in the range of 46 ± 1, and the fourth difference value data is in the range of −24 ± 1. ” Is detected as a match candidate. As described above, an allowable error is set for each value of the difference value data, and when all of the difference value data are within the set range, they are detected as match candidates. In this case, in addition to the permissible error, the number of pieces of difference value data that match may be set in the difference value comparison unit 33 in advance. For example, the number of matches is set to “3”, and when three or more of the four difference value data are matched (or within the allowable error range), they are detected as matching candidates.
[0029]
In the embodiment shown in FIG. 5, the case where the difference value data of the divided blocks in one row and five columns are compared has been described. In this way, by setting the number of comparisons of the difference value data in the difference value comparison unit 33 in advance, it is possible to shorten the processing time for detecting a similar image. Note that the difference value data of all the divided blocks may be compared. As described above, by increasing the number of comparisons of the difference value data, the area of the image to be compared can be increased, so that an image having a high degree of similarity to the reference image data can be detected.
[0030]
In FIG. 5, when the area values (d) of the images A and B are compared with each other, the comparison result does not match, so that the image is not detected as a similar image. Since the difference is calculated and compared, if the composition is similar and the color value or the luminance value is different, it was not detected as a similar image in the related art, but according to the above-described embodiment, Since the difference between the region values is calculated and the difference value is compared, it is possible to detect an image having a similar composition even if the color value and the luminance value are different.
[0031]
Through the above-described search processing, it is possible to search for contents of similar images.
[0032]
In the above embodiment, when the number of divisions of the dividing unit 11 is increased, images having a high degree of similarity can be detected with high accuracy. The time required for processing can be reduced. This division may be made for each pixel.
[0033]
In the above-described embodiment, the case where the average value of the luminance value is calculated in step S5 of the above-described embodiment has been described. However, the average value of the color values may be calculated, and a difference distribution file may be created and compared. Alternatively, the average value of the color values and the average value of the luminance, the color value and the luminance value of the representative pixel may be compared.
[0034]
In step S27 described above, the number of matching candidates need not be one or more. That is, the search may be terminated when three or more items are detected.
[0035]
Further, according to the above-described embodiment, the case where the difference value data between the adjacent divided blocks is generated has been described. However, the generation of the difference value data does not necessarily need to be performed between the adjacent divided blocks. The calculation may be performed between the divided blocks.
If the image is divided by the division unit 11 in the vertical and horizontal directions at a predetermined ratio, the image size is detected by the image size detection unit 13, and the image size is normalized by the image size normalization unit. 14 eliminates the need for normalization.
[0036]
In addition, the dividing unit 11, the image size detecting unit 12, the image size normalizing unit 13, the area value calculating unit 14, the divided block array unit 15, the difference value calculating unit 16, the difference distribution file reading unit 32, the difference comparing unit in FIG. 33, a program for realizing the function of the output unit 35 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed to execute a similar image detection process. May go. Here, the “computer system” includes an OS and hardware such as peripheral devices.
[0037]
The “computer system” also includes a homepage providing environment (or a display environment) if a WWW system is used.
The “computer-readable recording medium” refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, and a CD-ROM, and a storage device such as a hard disk built in a computer system. Further, a “computer-readable recording medium” refers to a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line, which dynamically holds the program for a short time. In this case, it is also assumed that a program that holds a program for a certain period of time, such as a volatile memory in a computer system serving as a server or a client in that case, is included. Further, the above-mentioned program may be for realizing a part of the above-mentioned functions, or may be for realizing the above-mentioned functions in combination with a program already recorded in a computer system.
[0038]
As described above, the embodiments of the present invention have been described in detail with reference to the drawings. However, the specific configuration is not limited to the embodiments, and includes a design and the like without departing from the gist of the present invention.
[0039]
【The invention's effect】
As described above, according to the present invention, an image is divided, an area value in an image area is generated, difference value data is generated, and a difference between a reference image serving as a reference for detecting similarity is determined. The value data and the difference value data of the comparison image to be detected whether or not they are similar are compared to detect whether or not the images match. Even if the images have different brightness values, it is possible to detect whether or not the images are similar, so that it is possible to prevent an image that can be visually considered to be coincident from being detected as similar, In addition, it is possible to prevent completely different images from being detected as similar images, and to improve the image detection accuracy.
[0040]
Further, according to the present invention, the size of the selected image is detected, and the image is divided after being changed to a predetermined size, so that images having different sizes are similar to each other. Detection can be performed.
[Brief description of the drawings]
FIG. 1 is a schematic block diagram showing a configuration of a similar image search device 1 according to an embodiment of the present invention.
FIG. 2 is a flowchart for describing a difference distribution file generation processing operation of the similar image search device.
FIG. 3 is an explanatory diagram for describing a difference distribution file generation processing operation of the similar image search device.
FIG. 4 is a flowchart illustrating an image search process.
FIG. 5 is a diagram for describing a difference distribution file generation process and an image search process.
FIG. 6 is a drawing for explaining a conventional technique.
[Explanation of symbols]
REFERENCE SIGNS LIST 1 similar image search device 11 division unit 12 image size detection unit 13 image size normalization unit 14 area value calculation unit 15 divided block array unit 16 difference value calculation unit 20 difference distribution file database 31 input unit 32 difference distribution file read unit 33 difference Value comparison unit 35 Output unit

Claims (6)

A similar image search device that compares a plurality of images,
A division unit that selects one image from the plurality of images and divides the image into a plurality of image regions based on a division number specified in advance;
An area value calculation unit that generates an area value that is information about a color in the image area divided by the division unit;
Among the image regions divided by the division unit, a region value generated by the region value calculation unit for a reference image region that is a reference image region, and a target image region that is an image region different from the reference image region And a region value generated by the region value calculation unit, and a difference value calculation unit that calculates difference between the region values to generate difference value data, and a reference that detects whether or not the images are similar among the plurality of images. The difference value data generated by the difference value calculation unit for the reference image, and the comparison value generated by the difference value calculation unit for a comparison image to be detected whether the image is similar to the reference image. Difference value data, and a difference value comparison unit that detects whether the images match.
An output unit that outputs a detection result detected by the difference value comparison unit,
A similar image search device comprising: The dividing unit includes:
An image size detector for detecting the size of the selected image;
An image size normalization unit that changes the size of the image detected by the image size detection unit to a predetermined size,
2. The similar image search apparatus according to claim 1, wherein the image after being changed by the image size normalizing unit is divided. Having a difference value database that stores difference value data generated by the difference value calculation unit for each image,
The difference value comparison unit reads difference value data stored in the difference value database in response to an instruction to compare images, and detects whether or not the images match. The similar image search device according to claim 2. A similar image search method in a similar image search device that compares a plurality of images,
Selecting one image from the plurality of images, dividing the image into a plurality of image regions based on a predetermined number of divisions,
Generating an area value that is information about a color in the divided image area;
Of the divided image areas, an area value generated for a reference image area that is a reference image area, and an area value generated for a target image area that is an image area different from the reference image area, Calculate the difference to generate difference value data,
Among the plurality of images, difference value data for a reference image serving as a reference for detecting whether or not the images are similar, and a comparison image serving as a target for detecting whether or not the images are similar to the image serving as the reference image A similar image search method characterized by comparing with difference value data generated for (i) and detecting whether images match. A similar image search program for comparing a plurality of images,
Selecting one image from the plurality of images and dividing the image into a plurality of image regions based on a predetermined number of divisions;
Generating an area value that is information about a color in the divided image area;
Of the divided image areas, an area value generated for a reference image area that is a reference image area, and an area value generated for a target image area that is an image area different from the reference image area, Calculating a difference to generate difference value data;
Among the plurality of images, difference value data for a reference image serving as a reference for detecting whether or not the images are similar, and a comparison image serving as a target for detecting whether or not the images are similar to the image serving as the reference image And comparing the difference value data generated with respect to, and detecting whether or not the images match,
Similar image search program for causing a computer to execute. A computer-readable recording medium recording a similar image search program for comparing a plurality of images,
Selecting one image from the plurality of images and dividing the image into a plurality of image regions based on a predetermined number of divisions;
Generating an area value that is information about a color in the divided image area;
Of the divided image areas, an area value generated for a reference image area that is a reference image area, and an area value generated for a target image area that is an image area different from the reference image area, Calculating a difference to generate difference value data;
Among the plurality of images, difference value data for a reference image serving as a reference for detecting whether or not the images are similar, and a comparison image serving as a target for detecting whether or not the images are similar to the image serving as the reference image And comparing the difference value data generated with respect to, and detecting whether or not the images match,
Recording medium storing a similar image search program for causing a computer to execute the program.

Images