JP2013109662A - Similar image determination device and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a similar image determination device capable of easily performing similar determination between an original image and a retrieval object image.SOLUTION: A similar image determination device 1 includes: a retrieval object image processing unit 22 that extracts a brightness level of a retrieval object image; an original image processing unit 13 that extracts a brightness level of an original image; and an image comparison unit 23 that determines whether the retrieval object image is similar to the original image or not. The similar image determination device 1 prepares a normalization concentration value histogram and determines whether the retrieval object image is similar to the original image or not by comparison between occurrence probability of the brightness level in the retrieval object image and occurrence probability of the brightness level in the original image.

Description

  The present invention provides similar image determination for determining whether a search target image is similar to the original image (including the same) or whether the search target image is similar to a part of the original image (including the same). The present invention relates to a device and a program.

  In recent years, it has become easy to convert images into digital data and store digitalized images as computer performance increases, network communication speeds up, and man-machine interface functions increase. ing. In addition, image duplication technology has been improved, and there are many problems that illegally duplicated images can be transferred to the network. Under such circumstances, many image searches are performed for the purpose of searching for inappropriate images.

  When searching for a desired image from images stored on a database or a network, a method using a keyword representing the feature of the image is generally used. However, in this method, if a keyword is not registered in an image to be searched or a different keyword is registered, the search cannot be performed.

Thus, a method for searching for a desired image without using a keyword has been proposed. For example, Patent Document 1 describes a method using an autocorrelation function.
In this method, first, one image (hereinafter referred to as “search target image”) is selected from images stored on a database or network. Then, the autocorrelation function obtained from the original image and the search target image is divided into a plurality of sections. Then, the distribution of pixel values in each section is represented by an exponential function, a sine function, and a cubic polynomial, and it is determined whether the original image and the search target image are similar based on the correlation of these functions or expressions. ing.

Japanese Patent No. 4134024

  By the way, as in Patent Document 1, the autocorrelation function obtained from the search target image is divided into a plurality of sections, and a similar function represented by an exponential function, a sine function, and a cubic polynomial is obtained in each section. If the image similarity determination is performed by determining the respective correlations, the number of processing steps increases and the processing becomes complicated.

  The present invention has been made to solve the above-described problems of the prior art, and is similar image determination that can suppress the complexity of processing when determining similarity between an original image and a search target image. An object is to provide a device and a program.

  A first aspect of the present invention is made to solve the above-described problem, and is a similar image determination device that determines whether or not a search target image and an original image are similar to each other. Search target image processing means for extracting the feature amount of the original image, original image processing means for extracting the feature amount of the original image, and similarity determination processing for determining whether the search target image is similar to the original image And the similarity determination process includes a frequency distribution of feature quantities of the search target image obtained through the search target image processing unit and a frequency distribution of the original image obtained through the original image processing unit. It is characterized in that it is determined whether the search target image and the original image are similar by comparing with a frequency distribution of feature amounts.

According to this configuration, the similarity determination is performed by comparing the frequency distribution of the feature amount, so that the number of processing steps is reduced as compared with the case where a function or the like is applied to display the feature of the image. . For this reason, it becomes possible to easily determine whether or not the original image and the search target image are similar (similarity determination).
Note that, as the feature amount in the present invention, a luminance signal of a pixel in an image, a density of a color signal (color signal), a difference value between adjacent pixels, or the like can be used. Alternatively, a statistic such as a density value histogram can be calculated based on the density of the luminance signal and the color signal (color signal) and used as a feature quantity.

  A second aspect of the present invention is the similar image determination device according to the first aspect, wherein the similarity determination processing includes the occurrence probability of the feature amount in the search target image and the occurrence of the feature amount in the original image. A correlation coefficient with the probability is calculated, and it is determined that the search target image and the original image are similar when the correlation coefficient is equal to or greater than a predetermined value.

  A third aspect of the present invention is the similar image determination device according to the first aspect or the second aspect, wherein the search target image processing means performs a smoothing process on the frequency distribution of the feature amount of the search target image. The original image processing means executes a smoothing process on the frequency distribution of the feature amount of the original image.

When the search target image is a compressed original image, the search target image and the original image are similar in appearance, but the frequency distribution of the feature amount of the search target image and the frequency distribution of the feature amount of the original image May be different. Then, due to such a difference in frequency distribution, it may be determined that the search target image and the original image are not similar.
In this respect, according to the same configuration, the frequency distribution change caused by the compression process can be reduced by performing the smoothing process on the frequency distribution of the feature amount of the search target image and the original image. The influence of the compression process on the similarity determination process can be reduced.

  A fourth aspect of the present invention is the similar image determination device according to any one of the first aspect to the third aspect, wherein the search target image processing means detects a dimension of the search target image, and the determination When the dimension of the search target image is smaller than the dimension of the original image, the means sets a first search area for the search target image and is identical to the first search area for the original image. A second search area having a dimension is set, and the similarity determination process is executed between the first search area and the second search area. The image of the first search area and the second search When it is determined that the image of the area is similar, an area where the search target image is estimated to be displayed is extracted from the original image based on the first search area and the second search area, The area and the search target And executes the class determining process with the image.

  When the search target image is similar to only a part of the original image (for example, when the search target image is a trimmed image of the original image), the original image includes images other than the images constituting the search target image. It will be. For this reason, the frequency distribution of the feature amount of the search target image is different from the frequency distribution of the feature amount of the original image, and it may be determined that the search target image and the original image are not similar. In this regard, according to the same configuration, an area in which the search target image is estimated to be displayed is extracted from the original image, and the similarity determination process is executed between the area and the search target image. . For this reason, even when the search target image is similar to only a part of the original image, the region estimated to display an image similar to the search target image in the original image, the search target image, and Similarity determination processing can be executed between the two. For this reason, when the search target image is an image similar to only a part of the original image, it is possible to suppress the determination that the search target image and the original image are not similar.

  According to a fifth aspect of the present invention, there is provided the similar image determination device according to any one of the first aspect to the fourth aspect, wherein the determination unit may include the original image obtained by rotating the search target image by a predetermined angle. A rotation determination process for determining whether or not there is a possibility, and when it is determined through the rotation determination process that the search target image may include the original image rotated by a predetermined angle, the original image is A search rotation image including a rotation original image rotated by a predetermined angle is created, and the similarity determination process is performed between the original image and a region corresponding to the rotation original image in the search target image. It is characterized by that.

  When the search target image is obtained by adding another image to the image obtained by rotating the original image (rotated original image), the frequency distribution of the feature amount of the search target image and the frequency distribution of the feature amount of the original image Will be different. For this reason, it may be determined that the search target image and the original image are not similar. In this regard, according to the same configuration, when it is determined through the rotation determination process that the search target image may include the rotation original image, an area corresponding to the rotation original image in the search target image and the original image Since the similarity determination process is performed between the regions, the frequency distribution of the feature amount can be compared between the region corresponding to the original image in the search target image and the original image. For this reason, even if the search target image is an image obtained by adding another image to the rotated original image, it is determined that the search target image and the original image are not similar due to the other image. It becomes possible to suppress this.

  A sixth aspect of the present invention is a program for causing a computer to execute a similar image determination process for determining whether or not a search target image and an original image are similar to each other. A search target image processing step for extracting a feature amount; an original image processing step for extracting a feature amount of the original image; a frequency distribution of the feature amount of the search target image obtained in the search target image processing step; In order to execute the similarity determination step of comparing the frequency distribution of the feature amount of the original image obtained in the original image processing step to determine whether the search target image is similar to the original image It is the program of this.

  The program of the present invention is the program according to the sixth aspect of the present invention, wherein, in the similarity determination step, the occurrence probability of the feature amount in the search target image and the feature amount in the original image are stored in the computer. A correlation coefficient with the occurrence probability of the image may be calculated, and it may be determined that the search target image and the original image are similar when the correlation coefficient is equal to or greater than a predetermined value.

  A program according to the present invention is the program according to the sixth aspect of the present invention, wherein the computer executes a smoothing process on the frequency distribution of the feature amount of the search target image in the search target image processing step. In addition, in the original image processing step, smoothing processing may be executed on the frequency distribution of the feature amount of the original image.

  A program according to the present invention is a program according to the sixth aspect of the present invention, wherein the computer detects the size of the search target image in the search target image processing step, and the computer determines the size of the search target image. When determining that the size is smaller than the size of the original image, the computer is caused to set a first search area for the search target image, and the original image has the same size as the first search area. A second search area is set, and the similarity determination step is executed between the first search area and the second search area, and the computer and the image of the first search area and the second search When it is determined that the image in the region is similar, the original image is transmitted to the computer based on the first search region and the second search region. To extract a region where al the search target image is estimated to be displayed, it may be caused to execute the class determination step between the search target image and the region.

  The program of the present invention is a program according to the sixth aspect of the present invention, wherein the computer determines whether or not the search target image may include the original image rotated by a predetermined angle. When the determination step is executed and the computer determines that the search target image may include the original image rotated by a predetermined angle through the rotation determination step, the computer displays the original image by the predetermined angle. A rotation image for search including the rotated original image that has been rotated is created, and the similarity determination step is executed between the original image and an area corresponding to the original image to be rotated in the search target image. Also good.

  According to the present invention, it is possible to provide a similar image determination device and a program capable of suppressing the complexity of processing when determining similarity between an original image and a search target image.

The conceptual diagram which shows the structure of the computer concerning 1st embodiment of this invention, and the electrical structure of a similar image determination apparatus. The flowchart which shows the process sequence of the similarity determination process concerning the embodiment. (A) is a schematic diagram illustrating an example of an original image according to the embodiment, (b) is a schematic diagram illustrating an example of a search target image according to the embodiment, and (c) is based on the original image according to the embodiment. (D) is a normalized density value histogram created based on the search target image according to the embodiment. (A) is a normalized density value histogram created based on the original image according to the second embodiment of the present invention, and (b) is based on an image obtained by compressing the original image according to the embodiment. Created normalized density value histogram. The flowchart which shows the process sequence of the similarity determination process concerning the embodiment. The flowchart which shows the process sequence of the similarity determination process concerning 3rd embodiment of this invention. (A) is a schematic diagram which shows an example of an original image, (b) is a schematic diagram which shows an example of a search object image, (c) is a schematic diagram which shows a part of original image shown to (a). The flowchart which shows the process sequence of the similarity determination process concerning 4th embodiment of this invention. (A) is a schematic diagram illustrating an example of an original image according to the embodiment, (b) is an explanatory diagram for illustrating a method of creating a table Θ (Ro, Co) according to the embodiment, and (c) is a search target. The schematic diagram which shows an example of an image. The flowchart which shows the process sequence of the similarity determination process concerning 5th embodiment of this invention. The flowchart which shows the process sequence of the similarity determination process concerning the embodiment. (A) is a schematic diagram showing an example of an original image and a search target image obtained by geometric conversion of the original image, and (b) shows a calculation result of a correlation coefficient between the original image and the search target image obtained by geometric conversion. table. (A) is a schematic diagram illustrating an example of a search target image including a rotated original image, and (b) is a calculation of a correlation coefficient between the search target image and each image included in the table Θ (Ro, Co). Table showing results.

(First embodiment)
Hereinafter, a first embodiment embodying the present invention will be described with reference to FIGS.
First, a schematic configuration of a similar image determination device 1 according to the present embodiment will be described with reference to FIG.

FIG. 1 shows the configuration of a computer according to the present embodiment and the electrical functional configuration of a similar image determination apparatus 1. Although not shown, the computer according to this embodiment includes a central processing unit (CPU), a hard disk drive (HDD), a RAM, and the like. The HDD stores a program for executing similarity determination processing, and the similar image determination device 1 is configured by the CPU executing this program.
The computer incorporating the similar image determination device 1 is not limited to a personal computer, and may be a PDA (Personal Digital Assistant), a mobile phone, a portable game machine, an AV device, or the like.

  Such a program is provided in a state stored in a portable medium such as a flexible disk or a CD-ROM, or from a storage device of another computer connected to a network via a communication device (not shown). Provided.

  As shown in FIG. 1, the computer is a similar image determination device 1 and a similar image which is an output device (such as a display or a printer) that outputs the result of similarity determination processing executed by the similar image determination device 1. A confirmation unit 25 is provided, and is connected to one or both of an image database (image DB) 31 and the Internet 41. The HDD of the computer stores image data relating to the original image 11 (hereinafter “original image”).

  When the user causes the CPU in the computer to execute the above-described program and activates the similar image determination apparatus 1, the computer first converts the original image captured via the interface unit 12 into an original image processing unit (“ Equivalent to “original image processing means”) 13 and converted into a predetermined file format (for example, bmp, png, etc.). Similarly, a search target image captured via the interface unit 21 is stored in a predetermined file format (for example, bmp, png, etc.) in a search target image processing unit (corresponding to “search target image processing means” of the present invention) 22. Convert to

  Then, a luminance signal of each pixel constituting the original image is extracted, and a normalized density value histogram (corresponding to “frequency distribution of feature amount” of the present invention) is created based on the extracted luminance signal. Similarly, the luminance signal of each pixel constituting the search target image is extracted, and a normalized density value histogram (corresponding to “frequency distribution of feature amount” of the present invention) is created based on the extracted luminance signal.

Thereafter, the normalized density value histogram created based on the original image and the normalized density value histogram created based on the search target image are respectively image comparison units 23 (corresponding to “determination means” of the present invention). Is output. In the image comparison unit 23, the computer compares the normalized density value histogram created based on the original image with the normalized density value histogram created based on the search target image. Then, in the similar image determination unit 24 (corresponding to “determination means” of the present invention), it is determined whether or not the original image and the search target image are similar (including the same).
The determination result in the similar image determination unit 24 is output to the similar image confirmation unit 25 which is an output device.

Hereinafter, with reference to FIG. 2 and FIG. 3, the processing procedure of the similarity determination process for determining whether or not the original image and the search target image are similar will be described in detail.
When this process is started, first, the computer acquires an original image (step S101). Next, a search target image is acquired (step S102).
Then, a normalized density value histogram is created based on the original image and the search target image (step S103).

  The normalized density value histogram will be described with reference to FIG. FIG. 3A shows an example of the original image, and FIG. 3B shows an example of the search target image. 3C shows an example of the normalized density value histogram created based on the original image, and FIG. 3D shows an example of the normalized density value histogram created based on the search target image. Indicates.

As shown in FIGS. 3C and 3D, the histogram created in the present embodiment divides the extracted luminance signal into a plurality of blocks, which are set as luminance levels on the horizontal axis. The luminance levels are arranged so as to increase toward the right side.
The vertical axis represents a value obtained by normalizing the detected luminance level by the total number of pixels included in the image, that is, a relative value (hereinafter referred to as “occurrence probability”) with the sum being 100%.

なお、「ｘ_ｉ」は原画像のｉ番目の輝度レベルの生起確率を示し、「ｘ（上バー）」は原画像における各輝度レベルの生起確率の平均値を示し、「ｙ_ｉ」は検索対象画像のｉ番目の輝度レベルの生起確率を示し、「ｙ（上バー）」は検索対象画像における各輝度レベルの生起確率の平均値を示す。 After creating the normalized density value histogram in this way, the computer uses the following (Equation 1) to normalize the normalized density value histogram created based on the original image and the normalized image created based on the search target image. A correlation coefficient r with the density value histogram is calculated (step S104).

“X _i ” indicates the occurrence probability of the i-th luminance level of the original image, “x (upper bar)” indicates the average value of the occurrence probability of each luminance level in the original image, and “y _i ” indicates the search. The occurrence probability of the i-th luminance level of the target image is indicated, and “y (upper bar)” indicates the average value of the occurrence probability of each luminance level in the search target image.

And it is determined whether this correlation coefficient r is more than threshold value Th (step S105). Note that the threshold Th is preferably set to about 0.95. If the threshold value Th is set to a value that is too low, there is a possibility that even if the search target image and the original image are different, they are erroneously determined to be similar. On the other hand, if the threshold Th is set too high, the search target image is similar to the original image even if the search target image is obtained by adding some processing to the original image. Due to the influence of various noises, etc., these images may be determined not to be similar.
When it is determined that the correlation coefficient r is less than the threshold value Th (step S105: NO), this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r is greater than or equal to the threshold Th (step S105: YES), similar image determination processing is performed to determine that the original image and the search target image are similar (step S106). This process is temporarily terminated.

According to the embodiment described above, the following effects can be obtained.
(1) In the present embodiment, the similarity determination is executed through comparison of the occurrence probability of the luminance level, so that the number of processing steps is less than in the case where a function or the like is applied to display the feature of the image. Become. For this reason, when performing similarity determination of an original image and a search object image, it can suppress that a process becomes complicated.

  (2) Further, since the correlation coefficient is used in comparing the occurrence probability of the luminance level, it can be appropriately determined whether or not the original image and the search target image are similar. Become.

(Second embodiment)
Hereinafter, a second embodiment embodying the present invention will be described based on FIG. 4 and FIG. 5 with a focus on differences from the first embodiment. 4 and 5, the same elements as those described in the first embodiment are denoted by the same reference numerals, and redundant description of these elements is omitted.

  In recent years, for example, JPEG compression, a technique for reducing the amount of information contained in an image by cutting out high-frequency components that are difficult to perceive by human vision among signals included in pixels constituting the image has been widely used. Commonly used. According to such a technique, it is possible to provide an image in which the amount of information is greatly reduced even if it looks the same.

  When the search target image is an image obtained by applying such compression processing to the original image, the search target image and the original image are created based on these images even though they are similar in appearance. The normalized density value histograms are different.

For example, FIG. 4A shows a normalized density value histogram created based on the original image, and FIG. 4B shows a normalized density value histogram created based on an image obtained by subjecting the original image to JPEG compression. Indicates.
As shown in FIGS. 4A and 4B, the original image and the original image subjected to the JPEG compression process have different luminance level occurrence probabilities.

  In particular, for example, when quantization is performed so that only even values (or odd values) can be taken and recombined, the created normalized density value histogram is between adjacent luminance levels (adjacent luminance). The probability of occurrence varies greatly between levels).

  Therefore, in this embodiment, when the computer determines that the original image and the search target image are not similar, the computer performs smoothing processing on the original image and the search target image and compresses the original image. Apply processing. Then, it is determined whether or not the original image and the search target image are similar.

Hereinafter, the processing procedure of the similarity determination processing in the present embodiment will be described in detail with reference to FIG.
When this process is started, the computer first acquires an original image (step S201), and then acquires a search target image (step S202).
Then, normalized density value histograms are respectively created based on the original image and the search target image, and the correlation coefficient r is calculated (step S203).

Next, it is determined whether or not the correlation coefficient r is less than the threshold value Th (step S204). When it is determined that the correlation coefficient r is greater than or equal to the threshold Th (step S204; NO), a similar image determination process is executed (step S215), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r is less than the threshold value Th (step S204; YES), a smoothing process is performed on the normalized density value histogram created based on the search target image and the original image (step S204). S205).

  If the search target image has been subjected to compression processing, the smoothing processing is performed in this manner, thereby reducing the change in the probability of occurrence of the luminance level caused by the compression processing. It is possible to reduce the influence on the rejection determination process.

なお、Ｐ（ｉ）は、ｉ番目の輝度レベルの生起確率を示し、Ｐ´（ｉ）は、平滑化処理を施した後のｉ番目の輝度レベルの生起確率を示す。 Next, the smoothing process in the present embodiment will be described.
In the smoothing process according to the present embodiment, for example, when the smoothing process is performed on the i-th brightness level, the window W is set to a predetermined number of pixels centered on the i-th brightness level with respect to the normalized density value histogram. (In this embodiment, it is set to include an odd number). Then, based on the occurrence probability of each luminance level included in the window W, the following (Equation 2) is calculated.
The window W preferably includes 5 luminance levels. If the window W is set to a value that is too large, there is a possibility that even if the search target image and the original image are different, they are erroneously determined to be similar. On the other hand, if the window W is set to a too small value, even if the search target image is an image obtained by compressing the original image, the bias generated in the normalized density value histogram of the search target image due to the compression process is smoothed. There is a possibility that these images are not similar to each other.

Note that P (i) indicates the occurrence probability of the i-th luminance level, and P ′ (i) indicates the occurrence probability of the i-th luminance level after the smoothing process is performed.

  As described above, when the search target image is an image obtained by performing compression processing on the original image, the normalization density value histogram generated based on the search target image and the original image are generated. This is different from the normalized density value histogram. That is, the occurrence probabilities of the respective luminance levels constituting the respective normalized density value histograms are different.

  In this regard, according to the present embodiment, the smoothing process is performed on the search target image and the original image as described above, so that the change in the probability of occurrence of the luminance level due to the compression process can be reduced. become. Therefore, it is possible to suppress erroneous determination that the search target image and the original image are not similar due to the fact that the compression processing has been performed even though the search target image and the original image are similar. it can.

Next, a correlation coefficient r _qk is calculated between the normalized density value histogram of the search target image subjected to the smoothing process and the normalized density value histogram of the original image (step S206). Since the correlation coefficient r _qk can be calculated by a method according to the method described in the first embodiment, a detailed description is omitted.
And it is determined whether this correlation coefficient _rqk is less than threshold value _Thqk (step S207). The threshold Th _qk is preferably set to about 0.95, similarly to the threshold Th described in the first embodiment.

When it is determined that the correlation coefficient r _qk is _{equal to} or greater than the threshold Th _qk (step S207; NO), a similar image determination process is executed (step S215), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _qk is less than the threshold Th _qk (step S207; YES), compression processing is performed on the original image (step S208).

  Then, a normalized density value histogram is created based on the original image subjected to the compression process (step S209), and a smoothing process is performed on the normalized density value histogram created based on the original image (step S209). Step S210).

  In this way, the original image is subjected to compression processing, and the normalized density value histogram created based on the original image is subjected to smoothing processing so that the search target image is subjected to compression processing on the original image. In the case of the image, the occurrence probability of each luminance level in the original image can be made closer to the occurrence probability of each luminance level in the search target image. In this embodiment, the original image is compressed at a predetermined compression rate (quality parameter).

Thereafter, the correlation coefficient r _qkm is calculated using the normalized density value histogram and the normalized density value histogram created based on the search target image (step S211). Since the correlation coefficient r _qkm can be calculated by a method according to the method described in the first embodiment, a detailed description is omitted.
And it is determined whether this correlation coefficient _rqkm is less than threshold value _Thqkm (step S212). Note that the threshold Th _qkm is preferably set to about 0.95, similarly to the threshold Th described in the first embodiment.

When it is determined that the correlation coefficient r _qkm is _{equal to} or greater than the threshold Th _qkm (step S212; NO), a similar image determination process is executed (step S215), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _qkm is less than the threshold Th _qkm (step S212; YES), the processing number n is incremented (step S213), and whether or not the processing number n is equal to or greater than the predetermined number C1. Is determined (step S214).

When it is determined that the processing number n is equal to or greater than the predetermined number of times C1 (step S214; YES), this processing is temporarily terminated.
On the other hand, when it is determined that the processing number n is less than the predetermined number C1 (step S214; NO), the processing from step S208 is repeated.

  When the search target image is an image subjected to compression processing, the compression rate (quality parameter) varies depending on the situation when the compression processing is performed. When the compression rate (quality parameter) is different, the created normalized density value histogram is different.

  For this reason, if the search target image and the original image are similar, even if the compression process is performed on the original image as described above, the search target image is different because the compression rate (quality parameter) is different. There is a possibility of determining that the image and the original image are not similar. That is, when the compression rate when the search target image is subjected to compression processing is larger than the above-described predetermined compression rate (when the quality parameter is set to a small value), the search target image and the original image are similar. It may be determined that it is not.

  In this regard, according to the present embodiment, the smoothing process is performed on the original image a predetermined number of times C1. For this reason, even though the search target image and the original image are similar, the compression processing is performed and the compression rate (quality parameter) is different, so that they are not similar It can suppress judging.

  The predetermined number of times C1 is preferably set to 2 to 4 times, and more preferably set to 4 times. If the predetermined number of times C1 is too large, the processing time may be longer than necessary. On the other hand, if the predetermined number of times C1 is too small, even if the search target image is an image compressed with a large compression ratio (by setting the quality parameter small) with respect to the original image, this compression ratio (quality parameter) ), The normalized density value histograms created based on the two images are different, and as a result, it may be determined that these images are not similar.

Further, when the image compression method is the JPEG method, it is preferable to set the predetermined number of times C1 when the compression process is performed on the original image to 2 as described above. At this time, it is more preferable to first set the quality parameter to 0.6 and perform compression processing, and then set the quality parameter to 0.2 and perform compression processing.
This is because the distribution of the normalized density value histogram changes significantly when the quality parameter is compressed with a quality parameter lower than about 0.2.

According to the embodiment described above, the following effects can be obtained in addition to the effects (1) and (2) described above.
(3) According to the present embodiment, a smoothing process is performed on the normalized density value histogram created based on the search target image, thereby suppressing a change in the occurrence probability of the luminance level due to the compression process. be able to. For this reason, even when the search target image is an image obtained by performing compression processing on the original image, it is possible to suppress the determination that the search target image and the original image are not similar. It becomes like this.

  (4) Further, since the smoothing process is performed on the original image for a predetermined number of times C1, the search target image is obtained by compressing the original image with a large compression ratio (with a small quality parameter). Even in some cases, it is possible to suppress erroneous determination that the search target image and the original image are not similar.

(Third embodiment)
Hereinafter, a third embodiment of the present invention will be described with reference to FIGS. 6 and 7 with a focus on differences from the first embodiment. 6 and 7, the same elements as those described in the first embodiment are denoted by the same reference numerals, and redundant description of these elements is omitted.

When the search target image is an image obtained by trimming a part of the original image, the normalized density value histogram generated based on the search target image even though the search target image is similar to a part of the original image And a normalized density value histogram created based on the original image. For this reason, there is a possibility that it may be determined that the original image and the search target image are not similar.
Therefore, in this embodiment, it is determined whether or not the search target image is included in the original image.

  Hereinafter, the processing procedure of the similarity determination processing in the present embodiment will be described in detail with reference to FIGS. 6 and 7. 7A shows an example of the original image, FIG. 7B shows an example of the search target image, and FIG. 7C shows the original image shown in FIG. 7A. A part of

When this processing is started, as shown in FIG. 6, first, the computer acquires an original image (step S301), and then acquires a search target image (step S302).
Then, the search target image size (Rs, Cs) (corresponding to the “dimension of the search target image” in the present invention) and the original image size (Ro, Co) (corresponding to the “dimension of the original image” in the present invention) are detected. (Step S303). “Ro” indicates the vertical width of the original image, “Co” indicates the horizontal width of the original image, “Rs” indicates the vertical width of the search target image, and “Cs” indicates the horizontal width of the search target image (see FIG. 7). ).

Then, it is determined whether or not the vertical width Rs of the search target image is less than the vertical width Ro of the original image and the horizontal width Cs of the search target image is less than the horizontal width Co of the original image (step S304).
When it is determined that the vertical width Rs of the search target image is equal to or larger than the vertical width Ro of the original image, or the horizontal width Cs of the search target image is equal to or larger than the horizontal width Co of the original image (step S304; NO), this process is temporarily ended. To do.

On the other hand, if it is determined that the vertical width Rs of the search target image is less than the vertical width Ro of the original image and the horizontal width Cs of the search target image is less than the horizontal width Co of the original image (step S304; YES), then the original In the image, the first line (corresponding to the “first search area” of the present invention) of the search target image and the area where the correlation coefficient r is maximized (hereinafter “start position”) are searched (step S305).
Here, the first line indicates an arbitrary part extracted from the search target image.

  Next, with reference to FIG. 7, a method for searching for the head position where the correlation coefficient r is maximized with the first line of the search target image in the original image (step S305) will be described in detail.

As shown in FIG. 7A, the computer first extracts the first line 300 from the search target image 301. In the present embodiment, the upper area of the search target image 301 is extracted as the first line 300.
Then, the computer normalizes the area having the horizontal width w1 and the vertical width h1 located in the upper left of the original image (the area indicated by the solid line in FIG. 7A and corresponding to the “second search area” of the present invention). Create a density density histogram. Then, a correlation coefficient r is calculated between the normalized density value histogram and the normalized density value histogram of the first line 300.

  Next, the computer creates a normalized density value histogram of an area indicated by a broken line in FIG. 7A (corresponding to the “second search area” of the present invention), and the normalized density value histogram and the first line. The correlation coefficient r is calculated between 300 normalized density value histograms. Thereafter, a normalized density value histogram of a region (corresponding to the “second search region” of the present invention) indicated by a one-dot chain line in FIG. 7A is created, and the normalized density value histogram and the normality of the first line 300 are created. The correlation coefficient r is calculated with the normalized density value histogram.

  The computer repeats such processing, creates a normalized density value histogram of a region (corresponding to the “second search region” of the present invention) having a horizontal width w1 and a vertical width h1 for the entire original image. A correlation coefficient r is calculated between the normalized density value histogram created based on the line 300.

  That is, as shown in FIG. 7C, the computer sequentially creates a normalized density value histogram of the region having the horizontal width w1 and the vertical width h1 while shifting the position little by little in the original image. Correlation coefficient r is calculated with respect to the normalized density value histogram of line 300, respectively. Then, a normalized density value histogram of an area (corresponding to the “second search area” of the present invention) having a horizontal width w1 and a vertical width h1 is generated for the entire original image, and is generated based on the first line 300. The correlation coefficient r is calculated with the normalized density value histogram.

  Then, with the region having the largest correlation coefficient r (the region M shown in FIG. 7A in this embodiment) as the head position, the region having the vertical width Rs and the horizontal width Cs that are the same size as the search target image ( A normalized density value histogram of the area M1) shown in FIG. That is, in the present embodiment, the upper area of the search target image 301 is the first line 300. For this reason, the region M having the largest correlation coefficient r with the first line 300 is assumed to be the upper region, and it is estimated that an image similar to the search target image is located in the region M1 having the vertical width Rs and the horizontal width Cs. .

Then, a correlation coefficient r _trm is calculated between the normalized density value histogram created based on the region M1 and the normalized density value histogram created based on the search target image 301 (step S306). Since the correlation coefficient r _trm can be calculated by a method according to the method described in the first embodiment, a detailed description is omitted.

Next, it is determined whether or not the correlation coefficient r _trm is _greater than or _equal to a threshold Th _trm (step S307). The threshold Th _trm is preferably set to 0.95 or more, and more preferably set to 0.99.
If it is determined that the correlation coefficient r _trm is _greater than or _equal to the threshold Th _trm (step S307; YES), a similar image determination process is executed (step S308), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _trm is less than the threshold Th _trm (step S307; NO), this process is temporarily ended.

According to the embodiment described above, the following effects can be obtained in addition to the effects (1) and (2) described above.
(5) According to the present embodiment, a normalized density value histogram of the region M1 in which the search target image is considered to be displayed is created from the original image, and the normalized density value histogram and the search target image are generated. The correlation coefficient r _trm is calculated between the normalized density value histogram created on the basis thereof. For this reason, even if the search target image is an image obtained by trimming a part of the original image, it is possible to appropriately determine whether or not the original image and the search target image are similar.

  (6) Further, a part of the search target image 301 is extracted as the first line 300, and the normalized density value histogram created based on the first line 300 and the normalized density created based on the part of the original image. The correlation coefficient r is calculated from the value histogram. As described above, when creating the normalized density value histogram, a part of the search target image is used instead of the entire area, so that it is possible to more suitably suppress the processing from becoming complicated. .

(Fourth embodiment)
Hereinafter, a fourth embodiment embodying the present invention will be described based on FIG. 8 and FIG. 9 with a focus on differences from the first embodiment. In FIGS. 8 and 9, the same elements as those described in the first embodiment are denoted by the same reference numerals, and redundant descriptions of these elements are omitted.

When the search target image is an image obtained by rotating the original image with another image added thereto, a normalized density value histogram created based on the search target image and a normalized density created based on the original image Since it is different from the value histogram, there is a risk that it is determined that the original image and the search target image are not similar.
Therefore, in this embodiment, first, rotation determination processing is performed to determine whether or not the search target image may include an image obtained by rotating the original image by a predetermined angle. When it is determined that there is a possibility that the search target image includes an image obtained by rotating the original image by a predetermined angle, the similarity determination process is executed by the method described below.

Hereinafter, the procedure of the rotation determination process and the similarity determination process in the present embodiment will be described in detail with reference to FIGS. 8 and 9.
When this processing is started, first, the computer acquires an original image (step S401) and also acquires a search target image (step S402).
Then, the search target image size (Rs, Cs) and the original image size (Ro, Co) are detected (step S403). Then, a table Θ (Ro, Co) is created based on the original image (step S404).

Here, the table Θ (Ro, Co) will be described with reference to FIG.
The original image in the FIG. 9 (a), the figure 9 (b) at four corners on the other image (corresponding to the "rotation original image" in the present invention) the original image image G1 obtained by the rotation angle theta _n respectively An image G3 _n (corresponding to the “rotating image for search” of the present invention) made G2 added to form a rectangle, and FIG. 9C shows a search target image including an image obtained by rotating the original image.
As shown in FIG. 9A, “Ro” indicates the vertical width of the original image, and “Co” indicates the horizontal width of the original image. Further, as shown in FIG. 9B, “Rg _n ” indicates the vertical width of the image G3 _n , and “Cg _n ” indicates the horizontal width of the image G3 _n . Furthermore, as shown in FIG. 9C, “Rs” indicates the vertical width of the search target image, and “Cs” indicates the horizontal width of the search target image.

In the present embodiment, as shown in FIG. 9A, the computer first obtains an original image having dimensions of a vertical width Ro and a horizontal width Co.
Then, as shown in FIG. 9 (b), this original image is rotated by an angle Θ _n (n = 1, 2, 3,... N), and this is set as an image G1 _n . Next, the image G1 _n corners to other images G2 of adding (regions black shown in FIG. 9 (b)), respectively and the image G3 _n. This angle Θ _n (n = 1, 2, 3... N) is preferably between π / 2 and + π / 2 at π / 180 intervals, but is not limited thereto.

As shown in FIG. 9B, the computer performs the operation for obtaining the image G3 _n (n = 1, 2, 3,... N) based on the original image in this way for each angle Θ _n . Then, the vertical width Rg _n and width Cg _n of each image G3 _n corresponding to each angle theta _n stored respectively, this is a table Θ (Ro, Co).
Thus, the computer stores the original image vertical width Rg _n and width Cg _n of each image G3 _n obtained by rotating the angle theta _n, as a table Θ (Ro, Co).

Each vertical width Rg _n and width Cg _n image G3 _n, based on the vertical width Ro and breadth Co of the original image, the following equation (3) can be calculated using (4).
Rg _n = Rocos Θ _n + Cosin Θ _n (3)
Cg _n = Rosin Θ _n + Cocos Θ _n (4)

  Next, the computer reads the table Θ (Ro, Co) (step S405).

Then, the table theta (Ro, Co) using each of the vertical width Rg _n and width Cg _n of all image G3 _n constituting the enlarged longitudinal retrieval target image (reduction) ratio Kr (Kr = Rs / Rg) and the horizontal expansion (reduction) rate Kc (Kc = Cs / Cg). The table Θ (Ro, Co) has an absolute value of the difference between the vertical compression (reduction) rate Kr and the horizontal compression (reduction) rate Kc less than a predetermined value C2 (Rg, Cg). (Step S406, corresponding to the “rotation determination process” of the present invention).

  When the absolute value of the difference between the compression (reduction) rate Kr in the vertical direction and the compression (reduction) rate in the horizontal direction is less than the predetermined value C2 (Rg, Cg) is in the table Θ (Ro, Co) (step) S406; YES), it is determined that the search target image may be an image obtained by rotating an image having substantially the same aspect ratio as the original image by an angle Θ and adding other images to the four corners.

Therefore, the computer rotates the original image by the angle Θ (step S407), and creates normalized density value histograms of the original image and the search target image (step S408).
Here, when creating a normalized density value histogram based on the original image, first, black images (images having a luminance of “0”) are added to the four corners of the original image rotated by angle Θ in step S407. . Then, the normalized density value histogram is created by ignoring the added image, that is, using only the luminance signal of the pixel in the area where the original image is displayed.

Next, a method for creating a normalized density value histogram in the search target image in step S408 will be described.
If the search target image is obtained by rotating the original image by an angle Θ and adding other images to the four corners, the search target image is displayed as an image G1 as shown in FIG. 9C. It is considered to include a region and a region where an image unrelated to the image G1 is displayed (a region indicated by hatching). Therefore, in this embodiment, when the normalized density value histogram is created based on the search target image in step S408, only the luminance signal of the pixel in the region where the above-described original image is actually displayed is used. That is, the normalized density value histogram is created based only on the same address as the address of the area where the original image is considered to be displayed among the search target images. In other words, in the search target image, the pixels in the region where the image other than the image G1 is considered to be displayed (the region indicated by hatching in FIG. 9C) is the target of the normalized density value histogram. I'm outside.

Then, a correlation coefficient r _rot is calculated between the normalized density value histogram created based on the search target image and the normalized density value histogram created based on the original image (step S409). It is determined whether or not the relationship number r _rot is greater than or equal to a threshold Th _rot (step S410). Since the correlation coefficient r _rot can be calculated by a method according to the method described in the first embodiment, a detailed description is omitted. Further, the threshold value Th _rot is preferably set to about 0.95 like the threshold value Th described in the first embodiment.

When it is determined that the correlation coefficient r _rot is equal to or greater than the threshold Th _rot (step S410; YES), a similar image determination process is executed (step S411), and this process is temporarily terminated.
On the other hand, the absolute value of the difference between the compression (reduction) rate Kr in the vertical direction and the compression (reduction) rate in the horizontal direction is less than the predetermined value C2 (Rg, Cg) is the table Θ (Ro, Co). If it is determined that the correlation coefficient r _rot is less than the threshold value Th _rot (step S410; NO), this process is temporarily terminated.

  As described above, in the present embodiment, pixels in a region where an image other than the image G1 is considered to be displayed are not included in the normalized density value histogram. For this reason, even when the search target image is obtained by rotating the original image by a predetermined angle and adding other images to the four corners, only the area corresponding to the original image is appropriately extracted from the search target image, A normalized density value histogram can be created based on the region. For this reason, it becomes possible to appropriately determine whether or not the rotated original image is included in the search target image.

According to the embodiment described above, the following effects can be obtained in addition to the effects (1) and (2) described above.
(7) According to the present embodiment, even when the search target image is obtained by adding other images to the four corners of the rotated original image, the search target image and the original image are similar. It becomes possible to appropriately determine whether or not there is.

(Fifth embodiment)
Hereinafter, a fifth embodiment embodying the present invention will be described based on FIGS. 10 to 13 with a focus on differences from the first embodiment. In addition, in these FIGS. 10-13, about the element same as each element demonstrated in 1st embodiment, the respectively same code | symbol is shown, The overlapping description about these elements is omitted.

  If the search target image is one of the original image trimmed, compressed original image, or rotated original image, a normalized density value histogram created based on the search target image Therefore, the normalized density value histogram created based on the original image is different, and it is determined that the original image and the search target image are not similar.

  Therefore, in this embodiment, even when the search target image is obtained by performing at least one of trimming, compression, and rotation on the original image, the search target image and the original image are similar. The similar image determination process that can appropriately determine whether or not the image is determined is executed.

Hereinafter, the processing procedure of the similarity determination processing in the present embodiment will be described in detail with reference to FIGS. 10 and 11.
When this processing is started, first, the computer acquires an original image (step S501) and also acquires a search target image (step S502).
If the original image or the detection target image is a color image, the color signal is converted into a luminance signal (step S503), a normalized density value histogram is created based on the search target image and the original image, and the correlation coefficient r Is calculated (step S504).

And it is determined whether this correlation coefficient r is less than threshold value Th (step S505).
When it is determined that the correlation coefficient r is greater than or equal to the threshold Th (step S505; NO), a similar image determination process is executed (step S529), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r is less than the threshold value Th (step S505; YES), it is determined whether or not the search target image has been subjected to compression processing on the original image.

That is, first, smoothing processing is performed on the search target image and the original image (step S506). Then, a correlation coefficient r _qk is calculated between the search target image subjected to the smoothing process and the original image (step S507).
Next, it is determined whether or not this correlation coefficient r _qk is less than a threshold value Th _qk (step S508).

When it is determined that the correlation coefficient r _qk is _{equal to} or greater than the threshold Th _qk (step S508; NO), a similar image determination process is executed (step S529), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _qk is less than the threshold Th _qk (step S508; YES), the original image is subjected to compression processing (step S509), and based on the original image subjected to this compression processing. A normalized density value histogram is created (step S510).

Then, a smoothing process is performed on the normalized density value histogram created based on the original image (step S511), and the normalized density value histogram of the original image subjected to the smoothing process and the search target image. A correlation coefficient r _qkm is calculated between the normalized density value histogram (step S512).
Then, it is determined whether or not the correlation coefficient r _qkm is less than the threshold Th _qkm (step 513).

When it is determined that the correlation coefficient r _qkm is _{equal to} or greater than the threshold Th _qkm (step 513; NO), a similar image determination process is executed (step S529), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _qkm is less than the threshold Th _qkm (step 513; YES), the processing number n is incremented (step S514), and whether or not the processing number n is equal to or greater than the predetermined number C1. Is determined (step S515).

When it is determined that the processing number n is less than the predetermined number C1 (step S515; NO), the processing after step S509 is repeated again.
On the other hand, when it is determined that the processing number n is equal to or greater than the predetermined number C1 (step S515; YES), it is determined whether or not the search target image is a rotation of the original image.

  That is, first, the search target image size (Rs, Cs) and the original image size (Ro, Co) are detected (step S516). Then, a table Θ (Ro, Co) is created based on the original image (step S517).

Next, Θ (Ro, Co) is read (step S518), and for each angle Θ, the vertical enlargement (reduction) rate Kr (Kr = Rs / Rg) and the horizontal enlargement (reduction) rate Kc ( Kc = Cs / Cg) is calculated (step S519). In other words, the vertical enlargement (reduction) rate Kr and the horizontal enlargement (reduction) rate Kc are calculated for all Θ (Ro, Co) constituting the table Θ (Ro, Co).
Next, the absolute value of the difference between the vertical compression (reduction) rate Kr and the horizontal compression (reduction) rate Kc is less than a predetermined value C2 (Rg, Cg) in the table Θ (Ro, Co). It is determined whether or not there is (step S520).

  It is determined that the absolute value of the difference between the compression (reduction) rate Kr in the vertical direction and the compression (reduction) rate in the horizontal direction is less than the predetermined value C2 (Rg, Cg) is in the table Θ (Ro, Co). If so (step S520; YES), the original image is rotated by an angle Θ (step S521), and a normalized density value histogram is created based on each of the original image and the search target image (step S522). In addition, since the creation method of the normalized density value histogram based on the search target image in this case has been described in the fourth embodiment, detailed description thereof is omitted.

Then, a correlation coefficient r _rot is calculated between the normalized density value histogram created based on the search target image and the normalized density value histogram created based on the original image (step S523), and this correlation coefficient r It is determined whether or not _{rot is} greater than or _equal to a threshold value Th _rot (step S524).

When it is determined that the correlation coefficient r _rot is greater than or equal to the threshold Th _rot (step S524; YES), a similar image determination process is executed (step S529), and this process is temporarily terminated.
On the other hand, the absolute value of the difference between the compression (reduction) rate Kr in the vertical direction and the compression (reduction) rate in the horizontal direction is less than the predetermined value C2 (Rg, Cg) is the table Θ (Ro, Co). If it is determined that the correlation coefficient r _rot is less than the threshold value Th _rot (step S524; NO), then the search target image is a part of the original image. It is determined whether or not it is trimmed.

That is, first, it is determined whether or not the vertical width Rs of the search target image is less than the vertical width Ro of the original image and the horizontal width Cs of the search target image is less than the horizontal width Co of the original image (step S525).
When it is determined that the vertical width Rs of the search target image is equal to or larger than the vertical width Ro of the original image, or the horizontal width Cs of the search target image is equal to or larger than the horizontal width Co of the original image (step S525; NO), the search target image is Since the image is not partly trimmed, this process is temporarily terminated.

On the other hand, when it is determined that the vertical width Rs of the search target image is less than the vertical width Ro of the original image and the horizontal width Cs of the search target image is less than the horizontal width Co of the original image (step S525; YES), The first position of the search target image and the head position where the correlation coefficient r is maximized are searched (step 526).
Then, an area of the same size as the search target image is extracted from the original image based on the head position, and a normalized density value histogram is created based on this area. Then, a correlation coefficient r _trm is calculated between the normalized density value histogram and the normalized density value histogram created based on the original image (step S527).

Next, it is determined whether or not the correlation coefficient r _trm is _greater than or _equal to a threshold Th _trm (step S528). When it is determined that the correlation coefficient r _trm is _{equal to} or greater than the threshold Th _trm (step S528; YES), a similar image determination process is executed (step S529), and this process is temporarily terminated.
On the other hand, when it is determined that the correlation coefficient r _trm is less than the threshold Th _trm (step S528; NO), this process is temporarily terminated.

  FIG. 12A shows an example of an original image, a search target image obtained by reducing the original image by 0.75 times, and a search target image obtained by enlarging the original image by 1.25 times. FIG. 12B shows the calculation result of the correlation coefficient between the search target image thus geometrically converted and the original image.

  As shown in FIG. 12B, when the search target image is an image obtained by reducing the original image by 0.75 times or an image obtained by enlarging the image by 1.25 times, The correlation coefficient calculated between the search target images was 0.99 or more.

FIG. 13A shows a search target image including the original image rotated right by 30 degrees. FIG. 13B shows a search target image between the search target image and each image G3 included in the table Θ (Ro, Co). The calculation result of a correlation coefficient is shown.
Note that the size of the original image is 202 pixels long and 230 pixels wide (a part of the test image “Lena” (size 256 × 256) is cut out and used as the original image), and the size of the search target image is vertical. It is 224 pixels and the width is 256 pixels. The search target image is an image obtained by adding another image to the four corners of the image obtained by rotating the original image 30 degrees to the right and further compressing the original image with the quality parameter 60.

  As shown in FIG. 13B, the correlation coefficient between the image G3 including the original image rotated right by 30 degrees and the search target image was as high as 0.984. On the other hand, the correlation coefficients between the image G3 including the original image rotated to an angle other than this and the search target image were all less than 0.9.

According to the embodiment described above, the following effects can be obtained.
(8) According to the present embodiment, the search target image is obtained by trimming or compressing the original image, or another image is added to the four corners of the rotated original image. In any case, it is possible to appropriately determine whether or not the original image and the search target image are similar.

(Other embodiments)
In addition, the similar image determination apparatus 1 concerning this invention is not limited to the structure illustrated in the said embodiment, For example, it can also implement as the following forms which changed the embodiment suitably. The following modifications are not applied only to the above-described embodiment, and different modifications can be combined with each other.

  In the second embodiment, the window W is set for the normalized density value histogram created based on the search target image and the normalized density value histogram created based on the original image. Although the smoothing process is performed, the smoothing process in the present invention is not limited to this, and various known methods can be used.

In the third embodiment, the upper area of the search target image is extracted as the first line. However, the first line of the present invention is not limited to this, and is an arbitrary position area of the search target image. I just need it. Further, the correlation coefficient r _trm may be calculated between the entire search target image and the original image without extracting the first line.

  In the fifth embodiment, after determining whether or not the search target image is an image subjected to compression processing, it is determined whether or not the search target image is an image obtained by rotating the original image, and then Although it is determined whether or not the original image is a trimmed image, the determination order in the present invention is not limited to this. For example, after determining whether or not the search target image is an image obtained by trimming a part of the original image, it is determined whether or not the search target image is an image obtained by rotating the original image, and then compression processing is performed. You may make it determine whether it is an applied image. Alternatively, after determining whether or not the search target image is an image obtained by rotating the original image, it is determined whether or not the search target image is an image obtained by trimming a part of the original image, and then compression processing is performed. It may be determined whether or not the image has been processed.

  In each of the above embodiments, the normalized density value histogram is created based on the entire images of the search target image and the original image, but the present invention is not limited to this. For example, the search target image and the original image are each divided into a plurality of blocks, a histogram such as an average value or standard deviation is created for each block, and the search target image and the original image are similar based on this histogram It may be determined whether or not.

  In the first embodiment, the second embodiment, the third embodiment, and the fourth embodiment, the original image and the search target image are similar using the normalized density value histogram created based on the luminance signal. It is determined whether or not there is, but the present invention is not limited to this. For example, a color signal may be converted into a luminance signal, and a normalized density value histogram may be created in which the occurrence probability of the obtained luminance signal is normalized by the total number of pixels.

  In each of the above embodiments, the normalized density value histogram is created based on the luminance level, but the present invention is not limited to this. For example, a difference value between adjacent pixels is calculated, a histogram between the search target image and the original image is created using the difference value and the occurrence probability of the difference value, and this histogram is used as a feature amount. Also good.

  In each of the above embodiments, it is determined whether or not the search target image and the original image are similar using a normalized density value histogram (statistic) created based on the luminance level. The invention is not limited to this. For example, the search target image and the original image are subjected to orthogonal transform such as discrete cosine transform and wavelet transform, and a normalized histogram is created as described above based on the transform coefficient pattern or transform coefficient. Whether the search target image and the original image are similar may be determined based on the above.

  In each of the above embodiments, the normalized density value histogram is created using the luminance level, but the present invention is not limited to this, and the normalized density value histogram is created using the color signal itself. You may do it.

  In each of the above embodiments, the normalized density value histogram is created using the luminance signal (Y signal), but the present invention is not limited to this, and the original image and the search target image are color images. In some cases, a normalized density value histogram may be created using another color signal (color signal).

  In each of the above-described embodiments and modifications, the case where a program is used as a component of the similar image determination device 1 in the present invention has been described, but the present invention is not limited to computer software, and an electronic circuit (integrated circuit) Etc.), or may be realized by cooperation between software and hardware.

DESCRIPTION OF SYMBOLS 1 ... Similar image determination apparatus 12 ... Interface part 13 ... Original image process part 21 ... Interface part 22 ... Search object image process part 23 ... Image comparison part 24 ... Similar image determination part

Claims (6)

A similar image determination device that determines whether a search target image and an original image are similar,
Search target image processing means for extracting feature quantities of the search target image, original image processing means for extracting feature quantities of the original image, and determining whether the search target image is similar to the original image Determination means for executing similarity determination processing,
The similarity determination processing compares the frequency distribution of the feature quantity of the search target image obtained through the search target image processing means with the frequency distribution of the feature quantity of the original image obtained through the original image processing means. And determining whether the search target image and the original image are similar to each other. The similar image determination device according to claim 1,
The similarity determination process calculates a correlation coefficient between the occurrence probability of the feature quantity in the search target image and the occurrence probability of the feature quantity in the original image, and the correlation coefficient is equal to or greater than a predetermined value. It is determined that the search target image and the original image are similar to each other. The similar image determination device according to claim 1 or 2,
The search target image processing means performs a smoothing process on the frequency distribution of the feature amount of the search target image,
The similar image determination apparatus, wherein the original image processing means performs a smoothing process on a frequency distribution of the feature values of the original image. The similar image determination device according to any one of claims 1 to 3,
The search target image processing means detects a dimension of the search target image,
The determination means sets a first search area for the search target image when the dimension of the search target image is smaller than the dimension of the original image, and sets the first search area for the original image. A second search area having the same dimensions is set, and the similarity determination process is executed between the first search area and the second search area.
When it is determined that the image of the first search area and the image of the second search area are similar,
Based on the first search area and the second search area, an area in which the search target image is estimated to be displayed is extracted from the original image, and the similarity between the area and the search target image is extracted. A similar image determination device characterized by executing determination processing. The similar image determination device according to any one of claims 1 to 4,
The determination means executes a rotation determination process for determining whether or not the search target image may include the original image rotated by a predetermined angle,
When it is determined that there is a possibility that the search target image includes the original image rotated by a predetermined angle through the rotation determination process, a search rotated image including the rotated original image obtained by rotating the original image by the predetermined angle A similar image determination device that creates and executes the similarity determination process between an area corresponding to the rotated original image in the search target image and the original image. A program for causing a computer to execute a similar image determination process for determining whether a search target image and an original image are similar,
In the computer,
A search target image processing step of extracting a feature amount of the search target image; and an original image processing step of extracting a feature amount of the original image;
Compare the frequency distribution of the feature amount of the search target image obtained in the search target image processing step with the frequency distribution of the feature amount of the original image obtained in the original image processing step, and A program for executing a similarity determination step for determining whether or not the original image is similar.

Images