JP2009251705A - Image evaluation device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image evaluation device clustering similar images without increasing calculation cost. <P>SOLUTION: This image evaluation device 100 comprises: a storage part 103 storing an image and attribute information of the image in association with each other; a feature value calculation part 104 reading the image from the storage part 103, calculating a feature value based on a color space of the image, and storing the feature value in the storage part 103 in association with the image; and a clustering execution part 105 classifying the image based on the feature value stored in the storage part 103, executing clustering processing for generating a plurality of clusters each comprising the similar images, and storing the cluster in the storage part 103 in association with the image. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image evaluation apparatus.

  Conventionally, from each image of a given image set, a statistic representing the feature of the image (hereinafter referred to as “feature”) is calculated, and the image set is based on the feature based on a certain standard. There are techniques for grouping, and among them, a number of techniques for automatically grouping image sets into similar image groups using a technique called clustering based on feature values have been proposed (for example, see Patent Document 1). Among them, a feature amount called a scalable color obtained by expressing an image in the HSV color space as a feature amount, creating a color histogram, and performing Haar transform on the color histogram is a color image. It is known that it has good characteristics for similar image search for, and is often used in this technical field.

The HSV color space is known as a color system that is close to the human color sense in color expression. Due to this, it is considered that the similar color search performance that adapts well to human senses can be achieved despite the simple statistics of scalable colors. On the other hand, for the same reason, many techniques for determining a print color (and color correction) based on color representation in the HSV color space have been proposed mainly in a technical area related to color printing. Among them, there has been proposed a lookup table (hereinafter referred to as “LUT”) for color space conversion and processing, mainly for the purpose of speeding up and improving the accuracy of color space conversion ( For example, see Patent Documents 1 and 2).
JP 2000-069306 A JP 2006-203896 A

  However, the conversion to the HSV color space used in the scalable color or the like described above is a simple calculation. However, if an attempt is made to adapt to an image with a scale of several million pixels taken by a recent digital still camera, the LUT is temporarily assumed. Even if conversion is performed, there is a problem that a certain calculation cost is required and the load on the platform to be mounted becomes high.

  The present invention has been made in view of such a problem, and an object thereof is to provide an image evaluation apparatus capable of clustering similar images without causing an increase in calculation cost.

  In order to solve the above problems, an image evaluation apparatus according to the present invention reads an image from the storage unit in which the image and attribute information of the image are stored in association with each other, and based on the image, A feature amount calculation unit that calculates a feature amount of a space, associates the feature amount with the image and stores the feature amount in a storage unit, and classifies the images based on the feature amount stored in the storage unit, and includes similar images. And a clustering execution unit that executes a clustering process for generating a plurality of clusters and stores the clusters in the storage unit in association with the cluster.

  In such an image evaluation apparatus, the feature amount calculation unit generates a color histogram based on a color space obtained by discretizing the first color system in which the image is expressed, and the color histogram is used as the second color specification. After color space conversion into the system, it is preferable to perform nonlinear quantization of frequency values and Haar conversion, and to use this calculation result as a feature amount.

  At this time, it is preferable that the first color system is a YCbCr color system. Here, in the color space obtained by discretizing the first color system, it is preferable that the discretization interval is subdivided according to the sensitivity to human color. Moreover, it is preferable that the second color system is any one of the HSV color system, the HMMD color system, and the RGB color system.

  Such an image evaluation apparatus is preset in the storage unit when the color histogram generated based on the color space obtained by discretizing the first color system is converted to the second color system. Preferably, it is configured to use a lookup table.

Here, when the first color system of the image is the YCbCr color system and the second color system is the HSV color system, the feature amount calculation unit YCbCr obtained by discretizing the YCbCr color system. The color space is divided into 8 for each of the Y axis, Cb axis, and Cr axis to create a YCbCr histogram with a division number of 512, and the YCbCr histogram is color space converted to a histogram of the HSV color system. , look-up table, the value of YCbCr in YCbCr histograms in k-th Y _k, Cb _k, and Cr _k, k-th coordinate values on the HSV color space corresponding to the YCbCr histograms H _k, S _k, V _k and, red component, green component in the k-th YCbCr histograms, and the intensity of the blue component R _k, G _k, and B _k, the largest component among these color components and MAX _k, the smallest formed It was a MIN _k, respectively a function that returns the smallest value among the arguments and min (·), respectively a function that returns the maximum value of the argument as the max (·),% to when the modulo operation, the following equation
R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
S _k = (MAX _k −MIN _k ) / MAX _k
V _k = MAX _k
Preferably, it is calculated by

Alternatively, when the first color system of the image is the YCbCr color system and the second color system is the HMMD color system, the feature amount calculation unit obtains a YCbCr color obtained by discretizing the YCbCr color system. The space is divided into 8 with respect to each of the Y axis, Cb axis, and Cr axis to create a YCbCr histogram having a division number of 512, and the YCbCr histogram is configured to be color space converted into a histogram of the HMMD color system. lookup table, the value of YCbCr in YCbCr histograms in k-th and Y _k, Cb _k, Cr _k, the coordinates of the HMMD color space corresponding to the k-th YCbCr histograms H _k, and Diff _k, k-th red component in the YCbCr histograms, and green component, the intensity of the blue component R _k, G _k, and B _k, the largest component among these color components and MAX _k, most The Sai components and MIN _k, when each function that returns the smallest value among the arguments and min (·), respectively a function that returns the maximum value of the argument as the max (·), the% and remainder operation, Next formula
R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
Diff _k = (MAX _k −MIN _k )
Preferably, it is calculated by

  In addition, it is preferable that such an image evaluation apparatus has an image display part which reads and displays an image for every cluster from a memory | storage part.

  If the image evaluation apparatus according to the present invention is configured as described above, similar images can be clustered without increasing the calculation cost, and similar images can be grouped with high accuracy and high speed.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. First, the configuration of the image evaluation apparatus 100 according to the present embodiment will be described with reference to FIG. As shown in FIG. 1, the image evaluation apparatus 100 includes a control unit 101 that controls the entire apparatus, a database 103 (storage unit) that stores the image and attribute information of the image in association with each other, The CPU 102 that performs image evaluation processing according to an instruction, the feature amount calculation unit 104 that calculates the feature amount of the color space for the image of the database 103 on the CPU 102, classifies the images based on the calculated feature amounts, and similar images A clustering execution unit 105 for generating a plurality of clusters (executing clustering processing), a work area (work RAM) 106 for temporarily storing the results of clustering processing and intermediate calculations, and the control unit 101 Clustering start switch (S / W) 107 for giving a clustering start instruction, and clustering And an image display unit 108. for displaying an image based on the result.

  As shown in FIG. 2, the database 103 has an image information storage area 103a for storing the storage destination of the image data, the cluster number to which the image data belongs, and an image data storage area 103f for storing the image data itself. The image information storage area 103a has a data structure as shown in FIG. 2A, a file name column 103b in which the file name of the image data is stored, and a storage location of the image data including the image and attribute information. The image data storage address column 103c for storing the address of the image, the feature amount column 103d for storing the feature amount of the image as the result of the feature amount calculation described later, and the cluster number column 103e for storing the clustering result described later. And have. On the other hand, the image data storage area 103f has a data structure as shown in FIG. 2B, and stores an image column 103g in which the image itself is stored, and attribute information such as a header area and shooting time information. Attribute information column 103h, and information for one image data is managed as one record. Here, the storage location (address) of the image data stored in the image data storage area 103f is stored in the image data storage address column 103c of the image information storage area 103a, and is stored in the file name column 103b. The image data to be extracted is specified from the file name of the image data, the address of the image data is extracted from the image data storage address column 103c, and the corresponding image data itself is obtained from the address of the image data storage area 103f. Can do. The image column 103g and the attribute information column 103h may be configured as one file. As such a data structure, Exif (Exchangeable image file format) standardized by the Japan Electronics Industry Promotion Association (JEIDA) is known.

(Image evaluation processing)
The image evaluation process performed by the image evaluation apparatus 100 will be described with reference to the flowchart of FIG. When the user presses the clustering start switch 107, a clustering start instruction is transmitted to the control unit 101 (step S100). When receiving the clustering start instruction, the control unit 101 issues a clustering start command to the CPU 102 (step S200). When the CPU 102 receives this clustering start command, the CPU 102 reads out the image data in the image data storage area 103f one by one in association with the file name (for example, 0001.jpg) stored in the file name column 103b of the database 103. Then, the image data is transferred to the feature amount calculation unit 104 and an instruction to execute feature amount calculation is given (step S300). The feature amount calculation unit 104 calculates the feature amount of the image based on the passed image data, and returns the feature amount as the calculation result to the CPU 102 (step S400). Details of the feature amount calculation processing will be described later. The CPU 102 stores the feature quantity as a calculation result in the feature quantity column 103d of the database 103 in association with the image data and stores it (step S500). These image data reading processing (step S300), feature amount calculation processing (step S400), and feature amount storage processing in the database 103 (step S500) are executed for all the images stored in the database 103. Repeat until.

  As described above, for all the images stored in the database 103, the image evaluation apparatus 100 continues the image clustering process after the image data has been read and the feature amount calculation and the storage in the database 103 have been completed. In the image clustering process, the CPU 102 reads all feature quantities associated with each image stored in the database 103, passes the feature quantities for all the images (hereinafter referred to as feature quantity set) to the clustering execution unit 105, and executes clustering execution. An instruction is issued to the clustering execution unit 105 (step S600). The clustering execution unit 105 executes clustering for classifying images and generating a plurality of clusters based on the given feature quantity set, and as shown in FIG. 9, which image is associated with each feature quantity. A calculation result indicating whether it belongs to a cluster (hereinafter referred to as a clustering result) is returned to the CPU 102 (step S700). Note that various data such as the image data, the feature amount set, and the clustering result as described above are temporarily stored in the work area 106, and each part uses the data to exchange data between the parts. Done.

  Next, the CPU 102 performs an indexing process based on the crystallizing result received from the clustering execution unit 105. That is, cluster numbers are assigned to all the generated clusters in ascending order, and a cluster number belonging to each image is associated with each image in the database 103 and stored in the cluster number column 103e of the database 103 (step S800). For example, in this embodiment, as shown in FIG. 9, the images are clustered into four clusters (similar image groups), and each cluster is given a cluster number of 1 to 4, and as shown in FIG. The cluster number to which the image belongs is stored in the image cluster number column 103e.

  After the image clustering process (step S700) and the indexing process (step S800) are completed, the CPU 102 notifies the control unit 101 to that effect (step S900). The control unit 101 accesses the database 103 when it receives a notification of the end of clustering processing, etc. from the CPU 102, reads all the image data and the cluster number to which it belongs, and instructs the image display unit 108 to display the clustering result. Each image data is issued (step S1000).

  Upon receiving the clustering result display command and the image data from the control unit 101, the image display unit 108 sequentially displays each image set that is the clustering result according to a predetermined display format (screen layout) (step S1100). ). FIG. 10 shows an example of displaying an image on the image display unit 108. Thus, an image is displayed for each cluster (similar group). Here, the image display method follows the format (size and position) defined in the selected layout. The above is the operation of the image evaluation process by the image evaluation apparatus 100.

(Feature calculation processing)
Next, the procedure of the feature amount calculation process (step S400) performed by the feature amount calculation unit 104 for one image will be described based on the flowchart of FIG. Hereinafter, in the present embodiment, it is assumed that the image data input to the feature amount calculation unit 104 is expressed in the YCbCr (hereinafter simply referred to as “YCC”) color system as the first color system. (In the following description, this image data is simply referred to as “YCC data”). As shown in FIG. 6A, the YCC data input to the feature amount calculation unit 104 is plotted on a segmented YCC color space, and the number plotted for each block is counted to obtain a three-dimensional data. A YCC histogram is generated (step S410).

  FIG. 6A shows an example of the YCC color space when the Y component is divided into 8 planes. FIG. 6B shows the CbCr plane of the third plane from the bottom of this figure. FIG. 6C shows the YCb plane of a predetermined plane in the YCC color space when the Cr component is divided into 8 planes. In FIG. 6B, the CbCr plane is divided into vertical and horizontal 8 × 8 block areas (converted from RGB to YCbCr, and the corresponding areas display coloring and dividing lines. The areas not to be displayed are white and the dividing lines are not displayed.) The color distribution density is different between the vicinity of the center and the vicinity of the periphery. That is, the color distribution is dense near the center and sparse near the periphery. From this color distribution characteristic, the present inventors have found that the number of blocks when dividing each plane can be reduced by making the area distribution fine near the center and rough around the periphery. That is, for example, when the CbCr plane is equally divided in FIG. 6B, the number of blocks is 16 × 16 = 256. However, as in the dividing line shown in FIG. Can be reduced to 8 × 8 = 64 blocks (compression of information amount). As a result, the number of calculations to be described later is reduced, and the operation speed can be increased. Furthermore, since the color distribution near the periphery is sparse, even if the blocks near the periphery are combined, there is no large difference in color information in units of blocks, so calculation errors are reduced and high-precision calculations are possible. Become.

As shown in FIG. 6A, since the YCC color space is divided into eight parts for each YCC axis as described above, the number of divisions when the YCC color space is divided (in the following description, this division number is referred to as “division number”). (Referred to as “bin”) is the number of bins = 8 ³ = 512 bins, that is, the generated YCC histogram is a 512-bin histogram. After creating the YCC histogram, the feature quantity calculation unit 104 refers to the LUT in which the correspondence between the YCC histogram and the HSV histogram is stored in advance in the database 101 from the 512-divided YCC color space, as shown in FIG. Color space conversion is performed on the divided HSV color space (step S420). 7A shows the HSV color space, FIG. 7B shows the H plane divided into 16 parts, and FIG. 7C shows a part of the SV plane. In FIG. 7, the H plane and the SV plane are divided at equal intervals, but are not necessarily divided at equal intervals. The conversion LUT from YCC to HSV can be created by applying the following equation (1) to the representative value of each bin of the YCC histogram, that is, the center point coordinates of each bin.

R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
S _k = (MAX _k −MIN _k ) / MAX _k
V _k = MAX _k (1)

The formula (1), the k subscript, represents the index of the bin in the YCC histogram, Y _k, Cb _k, Cr _k denotes the value of the YCC in bin = k of YCC histogram, H _k, S _k , V _k represent coordinate values in the HSV color space corresponding to bin = _k of the YCC histogram, and this value is a representative value of each bin in the HSV histogram. R _k , G _k , and B _k represent the intensities of the red, green, and blue components at bin = k, MAX _k represents the largest of these color components, and MIN _k represents these color components. Represents the smallest component. Also, min (•) and max (•) mean functions that return the minimum and maximum values of the arguments, respectively, and% means a remainder operation.

By such calculation, a conversion LUT into an HSV histogram corresponding to a predetermined YCC histogram is created. When converting a YCC histogram to an HSV histogram using this LUT, if frequency values belonging to different bins in the YCC histogram are converted to the same coordinates in the HSV color space, the frequency of the corresponding bin in the HSV histogram The value is the total value of the frequency values in the conversion source YCC histogram. That is, a coordinate _{(Y k, Cb k, Cr} k) frequency value n _{k of, (Y m, Cb m,} Cr m) is the frequency value of n _m of and (Y _k, Cb _k, Cr _k ) and (Y _m , Cb _m , Cr _m ) are converted to the same coordinates (H _k , S _k , V _k ) in the HSV color space, the frequency of (H _k , S _k , V _k ) The value is (n _k + n _m ). _{Incidentally, (Y k, Cb k,} Cr k) and _{(H k, S k, V} k) If there was a one-to-one _{correspondence, (H k, S k,} V k) frequency value of the intact n _k .

  The above is the histogram conversion processing performed by the feature amount calculation unit 104. After creating the HSV histogram as described above, the feature amount calculation unit 104 performs non-linear quantization and Haar transform of the frequency value in the same manner as the normal scalable color calculation, and the calculation result is finally obtained as the feature amount. The CPU 102 is notified as a return value (step S430).

  In this embodiment, an example in which color conversion is performed from the YCC color space to the HSV color space has been described. However, the conversion destination color space is not limited to the HSV color space, and HMMD (Hue, Max, Min, Diff) colors are used. It may be a space (these are defined as MPEG-7). At that time, the above formula (1) is changed as the following formula (2).

R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
Diff _k = (MAX _k −MIN _k ) (2)

The HMMD color system is a color space having (H _k , MAX _k , MIN _k , Diff _k ) in the above equation (2) as coordinate values of the color space. In this embodiment, the description has been made on the assumption that the color system of the conversion source is the YCC color system, but the present invention is not limited to this, and for example, the conversion source is the RGB color system. Even other color systems can be applied.

(Clustering process)
Next, the procedure of clustering processing (step S700) performed by the clustering execution unit 105 based on the feature value set will be described with reference to the flowchart of FIG. The clustering execution unit 105 acquires feature quantity sets from the CPU 102 (step S710), and executes hierarchical clustering, which is a kind of agglomerative clustering, for these (step S720). For this hierarchical clustering, a conventionally known method can be used. For example, the distance between images is obtained in an n-dimensional space based on the feature amount (n-dimensional) of the color space of each image, and hierarchical clustering is performed based on the distance. As a result of the hierarchical clustering, a system diagram as shown in FIG. 8 is created for all images corresponding to the feature amount. By cutting this system diagram at a predetermined threshold value _Td , a clustering result as shown in FIG. 9 is obtained (step S740). The clustering execution unit 105 notifies the CPU 102 of the completion of the image clustering process and the obtained clustering result (step S750).

  As described above, conventionally, when converting an image into an HSV color space or the like, all millions of pixels were calculated, so the calculation cost was enormous and the load on the implementation platform was high. According to the image evaluation apparatus of the embodiment, the color space conversion of the image is performed without increasing the calculation cost, the color space feature amount is calculated, and clustering of similar images is performed with high accuracy and high speed based on the feature amount. Can be done. As a result, similar images can be displayed for each group.

It is a block diagram showing the structure of the image evaluation apparatus of this embodiment. It is explanatory drawing which shows the data structure of a database. It is a flowchart showing the image evaluation process sequence performed with the image evaluation apparatus of this embodiment. It is a flowchart showing the feature-value calculation process procedure in a feature-value calculation part. It is a flowchart showing the clustering process sequence in a clustering execution part. (A) is a schematic diagram showing a YCC color space, (b) is a schematic diagram showing a CbCr plane, and (c) is a schematic diagram showing a YCr plane. (A) is a schematic diagram showing HSV color space, (b) is a schematic diagram showing H plane, (c) is a schematic diagram showing SV plane. It is a schematic diagram showing the result of having cut _| disconnected the systematic diagram obtained as a result of hierarchical clustering by _Td . It is a schematic diagram showing the example of the clustering result finally obtained. It is a schematic diagram which shows the example of the screen displayed on an image display part according to the screen layout for an image display.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Image evaluation apparatus 101 Control part 103 Database (memory | storage part)
104 feature amount calculation unit 105 clustering execution unit 108 image display unit

Claims (9)

A storage unit in which an image and attribute information of the image are associated and stored;
A feature amount calculation unit that reads the image from the storage unit, calculates a feature amount of a color space based on the image, and associates the feature amount with the image and stores the feature amount in the storage unit;
The image is classified based on the feature amount stored in the storage unit, a clustering process is performed to generate a plurality of clusters composed of similar images, and the cluster is associated with the image and stored in the storage unit A clustering execution unit,
An image evaluation apparatus. The feature amount calculation unit includes:
A color histogram is generated based on a color space obtained by discretizing the first color system in which the image is expressed, and the color histogram is converted into a second color system. The image evaluation apparatus according to claim 1, wherein conversion and Haar conversion are performed and the calculation result is used as a feature amount.   The image evaluation apparatus according to claim 2, wherein the first color system is a YCbCr color system.   The image evaluation apparatus according to claim 2 or 3, wherein a discretization interval is subdivided in a color space obtained by discretizing the first color system in accordance with sensitivity to human colors.   The image evaluation apparatus according to any one of claims 2 to 4, wherein the second color system is one of an HSV color system, an HMMD color system, and an RGB color system.   When the color histogram generated based on the color space obtained by discretizing the first color system is color space converted to the second color system, a lookup table preset in the storage unit is used. The image evaluation apparatus according to any one of claims 2 to 5, wherein the image evaluation apparatus is configured to be performed. When the first color system of the image is a YCbCr color system and the second color system is an HSV color system,
The feature amount calculation unit divides the YCbCr color space obtained by discretizing the YCbCr color system into eight parts for each of the Y-axis, Cb-axis, and Cr-axis to create a YCbCr histogram having a division number of 512. Is converted to a color space histogram of the HSV color system,
The look-up table, the value of YCbCr in YCbCr histograms in k-th Y _k, Cb _k, and Cr _k, k-th coordinate values H _k on HSV color space corresponding to the YCbCr histograms, S _k, V _k In the k-th YCbCr histogram, the red, green, and blue component intensities are R _k , G _k , B _k , the largest component among these color components is MAX _k, and the smallest component is MIN _k. When the function that returns the minimum value among the arguments is min (·), the function that returns the maximum value among the arguments is max (·), and% is the remainder operation,
R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
S _k = (MAX _k −MIN _k ) / MAX _k
V _k = MAX _k
The image evaluation apparatus according to claim 6, which is calculated by: When the first color system of the image is a YCbCr color system and the second color system is an HMMD color system,
The feature amount calculation unit divides the YCbCr color space obtained by discretizing the YCbCr color system into eight parts for each of the Y-axis, Cb-axis, and Cr-axis to create a YCbCr histogram having a division number of 512. Is converted to a color space histogram of the HMMD color system,
The look-up table, the value of YCbCr in YCbCr histograms in k-th and Y _k, Cb _k, Cr _k, the coordinates of the HMMD color space corresponding to the k-th YCbCr histograms H _k, and Diff _k, k The intensity of the red component, green component, and blue component in the second YCbCr histogram is R _k , G _k , B _k , the largest component among these color components is MAX _k , the smallest component is MIN _k , When the function that returns the minimum value is min (•), the function that returns the maximum value of each argument is max (•), and% is the remainder operation,
R _k = Y _k + 1.40200 × (Cr _k −128)
G _k = Y _k −0.34414 × (Cb _k −128) −0.71414 × (Cr _k −128)
B _k = Y _k + 1.77200 × (Cb _k −128)
MAX _k = max (R _k , G _k , B _k )
MIN _k = min (R _k , G _k , B _k )
H _k = (60 × (G _k −B _k ) / (MAX _k −MIN _k ))% 360, if MAX _k = R _k
= (60 × (B _k −R _k ) / (MAX _k −MIN _k ) +120)% 360, if MAX _k = G _k
= (60 × (R _k −G _k ) / (MAX _k −MIN _k ) +240)% 360, if MAX _k = B _k
Diff _k = (MAX _k −MIN _k )
The image evaluation apparatus according to claim 6, which is calculated by:   The image evaluation apparatus according to claim 1, further comprising an image display unit that reads and displays the image for each cluster from the storage unit.

Images