JP3812161B2 - Image comparison apparatus, image comparison method, and recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a technique for comparing one image with another image, and more particularly to a technique for searching for an image similar to a specific image (hereinafter referred to as “key image”) from a plurality of search target images.
[0002]
[Prior art]
In recent years, the development of the information processing field has been remarkable, and with the improvement in performance of personal computers and the like, image databases and electronic filing techniques have become widespread not only in the business field and specific applications, but also in general users and consumers. For example, an image or document is captured by an input device, stored and managed together with additional information such as a keyword, and printed by an output device as necessary.
[0003]
However, when searching for an image in a large amount of image databases, it is inferior in speed and inconvenient in practice unless additional information such as an appropriate keyword is added to the image.
[0004]
In such an image search, as a technique for searching for an image similar to a key image from a plurality of images, there is a technique that employs a method of comparing feature amounts of images. For example, saturation and edge strength are obtained as feature quantities of color and texture, respectively, and images having values close to the key image are searched from a plurality of images. As such technology, for example, Visual RetrievalWare, which is a search software of Exculibur, and VIR Image Engine, which is a search software of Virage, are known.
[0005]
[Problems to be solved by the invention]
However, in these known similar image retrieval techniques, a fixed default value or a value set by the user is adopted as a weighting (hereinafter referred to as “weight”) for each feature amount, and comparison is performed based on this weight. I was going. Therefore, after performing a search with a higher or lower weight for a certain feature amount, if a satisfactory search result is not obtained, another search is required by increasing or decreasing the weight of another feature amount.
[0006]
For this reason, in order for a user to search for a desired image, it is desirable to know in advance which feature amount of the key image is to be searched (that is, which feature amount should be weighted).
[0007]
In view of the above circumstances, the present application provides a technique for automatically setting a weight of a feature amount of a key image and improving operability at the time of a user's search.
[0008]
[Means for Solving the Problems]
  According to a first aspect of the present invention, there is provided an image comparison device for comparing a plurality of search target images with a reference image, wherein a plurality of feature amounts are obtained for the search target image and the reference image. An amount calculation unit; a weight setting unit configured to set a weight for each of the plurality of feature amounts from the plurality of feature amounts of the reference image; and a degree of similarity of the search target image with respect to the reference image. Similarity calculation means for each feature quantity, and total similarity calculation means for obtaining a total similarity of the search target image with respect to the reference image from the similarity and the weighting.The weight setting unit sets the weight equally for any of the plurality of feature amounts when all of the plurality of feature amounts are lower than a predetermined value.
[0009]
A second aspect of the present invention is the image comparison apparatus according to the first aspect, wherein the total similarity is compared with a predetermined threshold value, and a similar image similar to the reference image is determined. Comparing means obtained from the search target image is further provided.
[0011]
  Claim of the invention3According to claim 1Or 2The image comparison apparatus according to claim 1, further comprising weight display means for displaying the weight set by the weight setting means, and weight correction means for correcting the weight from outside.
[0012]
  Claim of the invention4The thing concerningBy computerAn image comparison method for comparing a plurality of search target images with a reference image, comprising: (a)The computer isA plurality of feature amounts are obtained for the search target image and the reference image.Save to the feature storageAnd (b)The computer isThe plurality of feature amounts of the reference imageIs loaded from the feature quantity storage unit.Setting a weight for each of the plurality of feature quantities; and (c)The computer isObtaining a similarity to the reference image for each of the plurality of feature amounts; and (d)The computer isObtaining a total similarity of the search target image with respect to the reference image from the similarity and the weighting.In the step (b), when all of the plurality of feature amounts are lower than a predetermined value, the computer sets the weighting equally for any of the plurality of feature amounts.
[0013]
  Claim of the invention5The claim is4A method for comparing images as described in (e)The computer isThe method further includes the step of comparing the total similarity with a predetermined threshold value to obtain a similar image similar to the reference image from the search target image.
[0015]
  Claim of the invention6The claim is4 or5InAn image comparison method as described in (f)The computer isThe weight set by the step (b) isOn the displayA step of displaying; (g)The computer isWeightingSetting change from userTheAcceptProcess and,Further provided between the step (b) and the step (d).
[0016]
  Claim of the invention7The thing concerningA computer-readable recording medium that records a program that causes a computer to compare a plurality of search target images with a reference image, wherein the program includes (a) a plurality of feature amounts, Obtaining the search target image and the reference image and storing them in a feature amount storage unit; and (b) loading the plurality of feature amounts of the reference image from the feature amount storage unit, and for each of the plurality of feature amounts (C) a step of obtaining the similarity of the search target image with respect to the reference image for each of the plurality of feature amounts; and (d) from the similarity and the weighting to the reference image. Obtaining a total similarity of the search target images, and in the step (b), the program causes the computer to execute the plurality of special features. If none of the quantity is lower than a predetermined value, equal to also set for the plurality of feature quantities of any of the weighting.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
FIG. 1 is a conceptual diagram showing a configuration capable of realizing a similar image detection system according to an embodiment of the present invention.
[0019]
The system is composed of a CPU and controls the entire system 1, a display 2 for displaying images including characters, a keyboard 3 and a mouse 4 for inputting instructions and operations, and data are stored. A flexible disk driver 5 that drives a flexible disk as a recording medium, a hard disk driver 6, a printer 7 that outputs image data or a drawing obtained by organizing an image, a scanner 8 that captures image data, and data are stored. A CD-ROM driver 9 that drives a CD-ROM that is a medium to be recorded, a speaker 10 that outputs sound including sound, and a microphone 11 that inputs sound. These are connected to each other with the computer 1 as a center, and exchange data.
[0020]
FIG. 2 is a block diagram showing an outline of the configuration of the computer 1 and the connection between the peripheral devices 2 to 11. The computer 1 includes, for example, a CPU 201 such as an Intel product number i80486DX, a clock generation circuit 202 that supplies clock signals of various control circuits described later, and a data bus 220. Further, all are connected to a data bus 220, and a ROM 203 for storing a program for controlling the system, a RAM 204 for storing data and programs, a display control circuit 205 for controlling the operation of the display 2, and an input transfer from the keyboard 3 A keyboard control circuit 206 for controlling the input, a mouse control circuit 207 for controlling the transfer of input from the mouse 4, a flexible disk driver control circuit 208 for controlling the operation of the flexible disk driver 5, and a hard disk driver control for controlling the operation of the hard disk driver 6. A circuit 209, a printer control circuit 210 that controls output to the printer 7, a scanner control circuit 211 that controls the operation of the scanner 8, a CD-ROM driver control circuit 212 that controls the operation of the CD-ROM driver 9, and a speaker Speaker control circuit 213 for controlling the output to 10, a microphone control circuit for controlling the input from the microphone 11 214 is also provided. Further, an expansion slot 215 for connecting an expansion board is also provided connected to the data bus 220. It is also possible to connect the SCSII board to the expansion slot 215 and connect the flexible disk driver 5, the hard disk driver 6, the scanner 8, and the CD-ROM driver 9 through this.
[0021]
A plurality of images and key images to be searched are stored on a flexible disk or a hard disk. Alternatively, an information storage device such as an MO medium may be further used to store an image. Of course, a new image can be captured from the scanner 8 or CD-ROM and stored in these media, or a digital camera (not shown) can be employed to capture the image.
[0022]
FIG. 3 is a flowchart showing a main routine of the information processing program in the system shown in FIGS. This program may include steps that are not related to the similar image search described in this embodiment.
[0023]
Initial setting is performed in step S1. Specifically, for example, initialization of flags necessary for information processing, initialization of screen display (for example, displaying a menu indicating the type of information processing), and the like can be mentioned. Thereafter, the process proceeds to step S2 and waits for an input for selecting a menu for what kind of information processing is performed. If no selection is made, step S2 is repeatedly executed.
[0024]
Three routes are illustrated here depending on the input selection. The route # 1 proceeds to a process of calculating the feature amount of the image, the route # 2 proceeds to a process of searching for a similar image, and the route # 3 proceeds to other information processing. When the information processing for calculating the feature amount of the image is selected, the process proceeds to step S3 via the route # 1, and the feature amount is calculated. When the information processing called the similar image search is selected, the process proceeds to step S4 via the route # 2, and the similar image search is performed. When other information processing is selected, the process proceeds to step S5 via route # 3, and the information processing is performed. Then, even when any route # 1 to # 3 is routed, other information processing is performed in step S6. As described above, steps S5 and S6 may be processing unrelated to the similar image search described in this embodiment.
[0025]
FIG. 4 is a block diagram illustrating the concept of functions realized by the computer 1. It goes without saying that the present embodiment can be realized without using the computer 1 and using an apparatus that exhibits the functions shown as blocks in this figure.
[0026]
The data storage unit V0 includes an image data storage unit V1 that stores image data and a feature amount storage unit V2 that stores image feature amounts. The image and the feature amount obtained from the image are associated with each other. Stored. The data storage unit V0 can be realized by the flexible disk driver 5 and the hard disk driver 6, for example. The feature amount calculation unit V3 obtains image data from the image data storage unit V1, and executes the process of step S3. The similar image search unit V4 designates an image, obtains image data and the feature amount of the image from the data storage unit V0, and executes the process of step S4. Both the feature amount calculation unit V3 and the similar image search unit V4 can be realized by the function of the computer 1.
[0027]
FIG. 5 is a flowchart showing details of step S3. First, in step S301, image data is loaded. In step S302, the size of the loaded image data is normalized. After that, the feature quantity that is the center of step S3 is calculated in step S303. This feature amount is stored in step S304.
[0028]
FIG. 6 is a block diagram showing details of the feature quantity calculation unit V3, which is a block showing the concept of functions. The image data load unit V31, the image size normalization unit V32, and the feature amount calculation unit V33 execute steps S301, S302, and S303, respectively. Step S304 is executed by the data storage unit V0 shown in FIG.
[0029]
As the feature amount calculated in step S303, for example, a color feature amount and a texture feature are known. An example of the color feature amount is an average value C of saturation. The average value C of the saturation can be calculated by converting the RGB value of the pixel of the image indicated by the image data into an HSV value and obtaining a saturation histogram. The average value C is generally normalized to a value between 0 and 1. Further, examples of the feature amount of the texture include a peak value P of the spectrum image and an edge strength E. The peak value P of the spectrum image corresponds to the strength of the periodicity of the texture, and is obtained as the peak of the spectrum image obtained by performing discrete Fourier transform on the image data. The peak value P is obtained as a value between 0 and 255, for example. Also, the edge strength E is obtained by making an image monochrome and calculating the sum of differences from neighboring pixels in each pixel. The edge strength E is generally normalized to a value between 0 and 1.
[0030]
FIG. 7 is a flowchart showing details of step S4, and FIG. 8 is a block diagram showing details of the similar image search unit V4 and connections with peripheral blocks. The similar image search unit V4 includes a weight setting unit V41, a similarity calculation unit V42, an overall similarity calculation unit V43, and a comparison unit V44. The weight setting unit V41 includes, for example, a display unit V5 that can be realized by the display 2, an input unit V6 that can be realized by the keyboard 3 and the mouse 4, a key image feature amount input unit V41a connected to the data storage unit V0, and a data storage unit V0. And a weight calculation unit V41b connected to the. Further, the comparison unit V44 receives a threshold value TH as a comparison reference for the total similarity Ra (described later in detail) output from the total similarity calculation unit. The threshold value TH can be stored in the data storage unit V0, for example.
[0031]
First, in step S401, a key image is designated. This can be realized by causing the key image feature input unit V41a to give the key image designation command A1 to the data storage unit V0 by the user operating the input unit V6. When selecting a key image, a procedure may be used in which a plurality of images are displayed on the display unit V5, and the user designates from these images. In this case, it is also desirable to cause the display unit V5 to display a catalog of a plurality of images, for example, reduced images, in advance from the data storage unit V0 before the key image designation command A1 is given to the data storage unit V0. One.
[0032]
In step S402, the feature amount for the specified key image is loaded, and in step S403, the weight for each feature amount is calculated from the feature amount for the key image. This can be realized by inputting the feature amount A2 for the key image from the data storage unit V0 to the weight calculation unit V41b, and the weight calculation unit V41b outputs, for example, a color weight Wc and a texture weight Wt. In view of performing step S402, it is desirable to execute step S3 in advance before executing step S4 in the flowchart shown in FIG.
[0033]
For example, the feature value A2 for the key image is the average value C of the saturation.₀, Peak value P of spectrum image₀, Edge strength E₀For example, the color weight Wc and the texture weight Wt are, for example, Wc = 1 + Kc · C.₀, Wt = 1 + Kp · P₀+ Ke · E₀As determined. Therefore, the average value C of the saturation of the key image₀The color weight Wc increases as the value of C increases.₀, Edge strength E₀The larger the is, the larger the texture weight Wt is set. In other words, if the weight ratio Wc / Wt is taken, the periodicity of the texture and the strength of the edge are not so strong but increase in the key image with high saturation, and conversely, the saturation is not high, but the periodicity of the texture and the edge The key image with high strength becomes small. The coefficients Kc, Kp, and Ke are set as appropriate to adjust the dynamic range of the feature value.
[0034]
In step S404, weights are displayed. The user who visually recognizes this has the option of adopting the weight or resetting it to an arbitrary value with reference to this weight. This can be realized by transmitting the weights Wc and Wt to the display unit V5 and accepting the weights Wc and Wt or changing the settings by the input unit V6.
[0035]
In step S405, the record pointer of the storage area storing the image to be searched is moved to the top. Thereafter, the similarity is sequentially obtained for the search target images, and the similarity with the key image is checked. In FIG. 8, this corresponds to the initialization of the record pointer in the data storage unit V0.
[0036]
In step S406, the feature amount of the search target image is loaded. This can be realized by inputting the feature amount A3 of the search target image from the data storage unit V0 to the similarity calculation unit V42. In step S407, the similarity is calculated from the feature amount A2 of the key image and the feature amount A3 of the search target image. For this reason, the feature amount A2 of the key image is also input to the similarity calculation unit V42.
[0037]
For the similarity, for example, a color similarity Rc and a texture similarity Rt are obtained separately, for example. The color similarity Rc is, for example, the average value C of the saturation of the key image.₀And the average value C of the saturation of the image to be searched_j (J = 1, 2,...), And the texture similarity Rt is, for example, the peak value P of the spectrum image of the key image.₀, Edge strength E₀And the peak value P of the spectrum image of the image to be searched_j, Edge strength E_jIt is demanded from.
[0038]
In step S408, the total similarity is calculated. The total similarity is a single index obtained on the basis of a plurality of similarities such as a color similarity and a texture similarity. In this case, the weight obtained in step S403 is considered. That is, the higher the weight similarity, the higher the contribution rate to the overall similarity. For example, the total similarity is defined as Ra = Σ (Wk · Rk) / ΣWk. Here, k represents the type of similarity, Wk represents the weight for the k-th feature amount for the key image, Rk represents the similarity for the k-th feature amount between the key image and the search target image, Σ represents the sum for parameter k.
[0039]
This is realized by obtaining the total similarity Ra as (WcRc + WtRt) / (Wc + Wt) from the color weight Wc and the texture weight Wt and the color similarity Rc and the texture similarity Rt in the total similarity calculation unit V43. it can.
[0040]
In step S409, it is determined whether or not the overall similarity is larger than a predetermined threshold value. If the total similarity is large, it is determined that the key image is similar to the search target image, and the search target is determined in step S410. The image of is displayed. This can be realized by the comparison unit V44 comparing the threshold value TH with the overall similarity Ra, and giving a display command D1 to the display unit V5 when Ra> TH. However, in FIG. 8, in order to avoid complication, a path through which image data is transmitted from the data storage unit V0 to the display unit V5 is omitted.
[0041]
If the total similarity is less than or equal to the predetermined threshold value in step S409, if step 3 or step S410 is executed, the process proceeds to step S411 to determine whether there is an image to be searched next. In step S412, the record pointer is incremented. This is because the comparison unit V44 compares the threshold value TH with the total similarity Ra, and when Ra ≦ TH or when the display command D1 is given to the display unit V5, the search command D2 is given to the data storage unit V0. This can be realized by causing the similarity calculation unit V42 to input the feature amount A3 for new image data from the data storage unit V0.
[0042]
If the search for all the images to be searched has already been finished in step S411, the process of step S4 is finished.
[0043]
As described above, according to the present invention, the weight of the feature amount is uniquely obtained for the key image based on the feature amount of the key image, while the similarity with the key image is obtained for each feature amount for each image to be searched. . Then, the total similarity is obtained based on the weight specific to the feature amount of the key image and the similarity of the feature amount with respect to the key image of the search target image. Therefore, it is not necessary for the user to set the weight by trial and error, and a desired similar image can be searched with a small number of searches.
[0044]
In the above description, step S403 may be omitted and weight setting may be performed automatically, or a step may be added in which the user can set a threshold before proceeding to step S406 and subsequent steps.
[0045]
Further, the weight may be set as follows. That is,
Wc = Kc · C₀                (Kc · C₀> 1),
Wc = 1 (Kc · C₀<1),
Wt = Kp · P₀+ Ke · E₀    (Kp · P₀+ Ke · E₀> 1),
Wt = 1 (Kp · P₀+ Ke · E₀<1)
It is.
[0046]
If set as described above, weighting is performed only when the feature amount is large to some extent, while any feature amount C₀, P₀, E₀Is also small, Wc = Wt = 1, and no difference is caused between the weights. Therefore, in the case where the feature amount is small and it cannot be said that the feature amount of the image represents the feature of the image, there is no meaningless weighting, and thus the possibility that an unintended image is searched is also avoided. it can.
[0047]
In the above embodiment, the image is searched by comparing the overall similarity with a predetermined threshold value, but all images may be displayed in order from the image with the highest overall similarity. FIG. 9 is a flowchart showing this operation. Compared to the flowchart shown in FIG. 7, step S413 is provided instead of steps S409 and S410, and step S414 that is executed when “NO” is determined in step S411 is added.
[0048]
In step S413, the total similarity is stored. Therefore, until it is determined as “NO” in step S411, the total similarity obtained in step S408 is stored for all the images to be searched. In step S414, the search target image is preferentially displayed in the order of the overall similarity, for example, those having a higher overall similarity.
[0049]
FIG. 10 is a block diagram showing functions for realizing the flowchart shown in FIG. The similar image search unit V4 gives the overall similarity to the data storage unit V0 in order to execute step S413. The data storage unit V0 stores the total similarity in association with the image data having the feature amount that is the basis for the total similarity. The overall similarity can be stored in, for example, the RAM 204 in addition to the data storage unit V0 that can be realized by the flexible disk driver 5 and the hard disk driver 6.
[0050]
Further, the similar image search unit V4 specifies the overall similarity instead of specifying the image to be displayed in order to execute step S414. The data storage unit V0 provides image data corresponding to the designated overall similarity for display on the display unit V5.
[0051]
FIG. 11 is a block diagram showing details of the similar image search unit V4 shown in FIG. 10 and connections with peripheral blocks. 8 has a configuration in which the comparison unit V44 is replaced with an overall similarity designation unit V45. The overall similarity designation unit V45 gives an instruction D3 for designating the overall similarity to the data storage unit V0. At this time, the total similarity specifying unit V45 itself stores the total similarity for all the search target images in advance in, for example, the RAM 204, and rearranges the images in order from the larger one, and sequentially outputs the command D3. You can also. Alternatively, the total similarity that sequentially decreases in a predetermined step may be output as the command D3, and the image data having the closest total similarity may be extracted from the data storage unit V0. In either case, a display command D1 is given to the display unit V5 in order to display the corresponding image data.
[0052]
Further, the search target images to be displayed may be limited to a predetermined number or all of them may be displayed. Since the process of rearranging and displaying the image data based on a certain index as in the process of step S414 is a known technique, details thereof are not shown here.
[0053]
The flowchart shown in the above description is realized by the computer 1 performing each function based on the program stored in the ROM 203. The program recorded on a recording medium such as a flexible disk is stored in the flexible disk driver. 5 may be realized by being read into the computer 1 via the computer 5 and performing the functions of the computer 1. Furthermore, the program recorded on the recording medium may not be a computer that implements all of the above flowcharts, but may be a program that implements a part of the flowchart. This is because the above-described flowchart can be realized even when operating in cooperation with a program already loaded into the computer 1.
[0054]
【The invention's effect】
  Of the present invention, the image comparison apparatus according to claim 1 and the claim4According to the image comparison method, the weighting of the feature amount is uniquely obtained for the reference image based on the feature amount of the reference image, while the similarity with the reference image is obtained for each feature amount for each search target image. Therefore, since the total similarity is obtained based on the weighting specific to the feature amount of the reference image and the similarity of the feature amount with respect to the reference image of the search target image, the user does not need to set weighting by trial and error, The reference image and the search target image can be compared.
In addition, when there is no distinctive feature amount, all feature amounts can be searched equally.
[0055]
  Of the present invention, an image comparison apparatus according to claim 2 and claim5According to the image comparison method, a desired similar image can be searched with a small number of searches.
[0057]
  Claim of the invention3Image comparison apparatus and claim6According to the image comparison method, since the user can set the weighting again while referring to the automatically set weighting, the degree of freedom of the search condition is increased.
[0058]
  Claim of the invention7According to the recording medium,While the weight of the feature amount is uniquely obtained for the reference image based on the feature amount of the reference image, the computer calculates the similarity with the reference image for each search target image for each feature amount. Therefore, since the computer calculates the total similarity based on the weighting specific to the feature amount of the reference image and the similarity of the feature amount of the search target image to the reference image, the user sets the weight by trial and error. No need,This can contribute to efficient determination of image similarity.
Further, when there are no distinctive features in the feature amount, the computer can be made to search equally for all feature amounts.
[Brief description of the drawings]
FIG. 1 is a conceptual diagram showing a configuration of a similar image detection system according to the present invention.
FIG. 2 is a block diagram illustrating the configuration of the embodiment according to the present invention.
FIG. 3 is a flowchart showing the operation of the embodiment according to the present invention.
FIG. 4 is a block diagram illustrating the configuration of the embodiment according to the present invention.
FIG. 5 is a flowchart showing the operation of the embodiment according to the present invention.
FIG. 6 is a block diagram illustrating the configuration of the embodiment according to the present invention.
FIG. 7 is a flowchart showing the operation of the embodiment according to the present invention.
FIG. 8 is a block diagram illustrating the configuration of the embodiment according to the present invention.
FIG. 9 is a flowchart showing a modification operation of the embodiment according to the present invention.
FIG. 10 is a block diagram illustrating a modified configuration of the embodiment according to the present invention.
FIG. 11 is a block diagram illustrating a modified configuration of the embodiment according to the present invention.
[Explanation of symbols]
V0 data storage unit, V3 feature quantity calculation unit, V33 feature quantity calculation unit, V4 similar image search unit, V41 weight setting unit, V42 similarity calculation unit, V43 total similarity calculation unit, V44 comparison unit, V45 total similarity designation Part, V5 display part, V6 input part, Wc color weight, Wt texture weight, Rc color similarity, Rt texture similarity, Ra total similarity, TH threshold.

Claims (7)

An image comparison device that compares a plurality of search target images with a reference image,
A feature amount calculating means for obtaining a plurality of feature amounts for the search target image and the reference image;
Weight setting means for setting a weight for each of the plurality of feature amounts from the plurality of feature amounts of the reference image;
Similarity calculating means for determining the similarity of the search target image with respect to the reference image for each of the plurality of feature amounts;
Comprehensive similarity calculation means for obtaining an overall similarity of the search target image with respect to the reference image from the similarity and the weighting ,
The weighting setting unit is an image comparison device that sets the weighting equally for any of the plurality of feature amounts when any of the plurality of feature amounts is lower than a predetermined value . The image comparison apparatus according to claim 1, further comprising a comparison unit that compares the total similarity with a predetermined threshold and obtains a similar image similar to the reference image from the search target image.   Weight display means for displaying the weight set by the weight setting means;
  Weight correction means for correcting the weight from outside;
The image comparison apparatus according to claim 1, further comprising:   An image comparison method for comparing a plurality of search target images with a reference image by a computer,
  (A) The computer obtains a plurality of feature amounts for the search target image and the reference image and stores them in a feature amount storage unit;
  (B) the computer loading the plurality of feature amounts of the reference image from the feature amount storage unit, and setting a weight for each of the plurality of feature amounts;
  (C) the computer obtaining a similarity of the search target image with respect to the reference image for each of the plurality of feature amounts;
  (D) The computer includes a step of obtaining a total similarity of the search target image with respect to the reference image from the similarity and the weighting,
  In the step (b), when all of the plurality of feature amounts are lower than a predetermined value, the computer sets the weighting equally for any of the plurality of feature amounts. .   (E) a step in which the computer compares the total similarity with a predetermined threshold and obtains a similar image similar to the reference image from the search target image
The image comparison method according to claim 4, further comprising:   (F) the computer displaying the weight set in the step (b) on a display unit;
  (G) the computer receiving a setting change from the user for the weighting;
The image comparison method according to claim 4, further comprising between step (b) and step (d).   Write a program that allows a computer to compare multiple search target images with a reference image. A recorded computer-readable recording medium,
  The program is stored in the computer.
  (A) obtaining a plurality of feature quantities for the search target image and the reference image and storing them in a feature quantity storage unit;
  (B) loading the plurality of feature quantities of the reference image from the feature quantity storage unit, and setting weights for each of the plurality of feature quantities;
  (C) obtaining a similarity of the search target image with respect to the reference image for each of the plurality of feature amounts;
  (D) performing a step of obtaining an overall similarity of the search target image with respect to the reference image from the similarity and the weighting;
  In the step (b), the program causes the computer to set the weighting equally for any of the plurality of feature amounts when any of the plurality of feature amounts is lower than a predetermined value. ,
Computer-readable recording medium.

Images