JP2002190018A - Device and method for classifying image region - Google Patents

Abstract

PROBLEM TO BE SOLVED: To rapidly perform cluster classification conformed to an operator's intention in performing color classification. SOLUTION: This device is provided with a histogram preparing part 2 for preparing the histogram of an image at a lattice point in color space with each color component as an axis on the basis of the image composed of a plurality of color components; a histogram display part 5 for displaying the histogram prepared by the histogram preparing part 2, by an object of size corresponding to a histogram value; and a cluster classification part 3 for classifying all the lattice points of the histogram prepared by the histogram preparing part 2, according to a plurality of representative values.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image area classification apparatus and an image area classification method for classifying a full-color image into a certain number of representative colors, for example.

[0002]

2. Description of the Related Art When a full-color image such as a color photograph is read by an image input device such as a scanner, R / P
Digitalization is often performed using a total of 24 bits, each of 8 bits for (red), G (green), and B (blue). In this case, each pixel represents an arbitrary one of 2 24 = 16.770,000 colors.

However, in the case where a printed matter using a relatively small number of colors is digitized and used, if similar colors are grouped into one representative color and the entire image is represented by a small number of representative colors, the storage memory is required. Can greatly reduce the capacity and communication volume.

For example, when a full-color image is replaced with 16 types of representative colors, each pixel is represented by a 4-bit color number, whereby the amount of image data is reduced to 1/6 of the original. Since the number of colors is reduced, the area in the image is painted with a single color, so that the amount of image data can be further reduced. Further, since the obtained image is also a result of image processing in which a variation in density has been removed, the image can be used for region analysis and character recognition.

Here, the image data is converted into a representative color and
For display on a display device such as an RT, color data corresponding to the color number of the representative color is stored in a look-up table (LUT), and each time the data is read out to a CRT or the like, the LUT is referred to and displayed.

For this reason, when a full-color image is given, a process for selecting a representative color to be stored in the LUT is required. This representative color selection process is also called cluster classification or color classification in a color space, and a general cluster classification algorithm is applied.

FIG. 12 is a diagram for explaining a conventional image area classification apparatus for performing color classification. As shown in FIG. 12A, this image area classification device includes an image input unit 1, a histogram creation unit 2, a cluster classification unit 3, and an image output unit 4.

[0008] The image input unit 1 is a part that takes in a full-color image (black and white in the figure) as shown in FIG. Further, the histogram creating unit 2 obtains a frequency distribution (histogram) of pixels having the same RGB value from the input image as a non-negative integer scalar function defined in the three-dimensional color space.

The cluster classification unit 3 performs a process (cluster classification algorithm) for obtaining a group of points (colors) having a large histogram value and being close to each other in the RGB color space. Since these groups of colors represent clusters of similar colors, they are given the same color number and representative color. Finally, the entire grid point or the entire color space where the histogram value is not 0 is classified into N clusters.

The image output unit 4 reconstructs the classification result of the cluster classification unit 3 in an image space, and determines that all pixels of the input image have any color number (for example, 0 in the case of N clusters). To N-1), a lookup table of image data and representative colors is output.

The cluster classification algorithm applied by the cluster classification unit 3 is described in "Image Recognition Theory", "Maximum Distance Algorithm", and "K" described on pages 120 to 126 of Makoto Nagao, published by Corona. Averaging algorithm, self-convergence algorithm, and the like. The representative color is often selected based on the center of gravity of the histogram of each cluster.

[0012]

However, image region classification using such a cluster classification algorithm has the following problems. That is, the number of clusters does not match the distribution of the histogram, and the classification result may be different from the operator's intention. Even when there are a plurality of input images using the same type of color set, since the cluster classification process is performed from the beginning for each input image,
Requires a lot of processing time.

[0013]

SUMMARY OF THE INVENTION The present invention has been made to solve such problems. That is, the image region classification device of the present invention includes a histogram creating unit that creates a histogram of the image at a grid point in a color space with each color component as an axis, based on an image composed of a plurality of color components; Histogram display means for displaying the histogram created by the means with an object having a size corresponding to the histogram value, and cluster classification means for classifying all the grid points of the histogram created by the histogram creation means with a plurality of representative values. Have.

Further, according to the image area classification method of the present invention, based on an image composed of a plurality of color components, a histogram of the image at a grid point in a color space with each color component as an axis is created, and In a method of generating a plurality of cluster regions by classifying all grid points by a plurality of representative values, this is a method of displaying a histogram with an object having a size corresponding to the histogram value.

According to the present invention, a histogram of an image at a grid point in a color space with each color component as an axis is created based on the image to be classified, and the histogram is converted into an object having a size corresponding to the histogram value. , The histogram value in the color space can be visually recognized. Further, referring to the displayed object, the operator can easily set a cluster classification that suits his or her intention.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below with reference to the drawings. FIGS. 1A and 1B are diagrams illustrating an image area classification device according to the first embodiment. FIG. 1A illustrates a configuration diagram, and FIG. 1B illustrates a histogram display example.

That is, the image area classification device of the present embodiment includes an image input unit 1, a histogram creation unit 2, a cluster classification unit 3, and an image output unit 4, and also includes a histogram display unit 5 and an operation input unit 6. I have.

The image input unit 1 is a unit that takes in a full-color image as an input image. Further, the histogram creating unit 2 obtains a frequency distribution (histogram) of pixels having the same RGB value from the input image as a non-negative integer scalar function defined in the three-dimensional color space.

The cluster classification unit 3 performs processing (cluster classification algorithm) for obtaining a group of points (colors) having a large histogram value and being close to each other in the RGB color space. Since these groups of colors represent clusters of similar colors, they are given the same color number and representative color. Finally, the entire grid point or the entire color space where the histogram value is not 0 is classified into N clusters.

The image output unit 4 reconstructs the classification result of the cluster classification unit 3 in an image space, and determines that all the pixels of the input image have any color number (for example, 0 in the case of N clusters). To N-1), a lookup table of image data and representative colors is output.

The histogram display unit 5 performs a process of displaying the histogram created by the histogram creating unit 2 with an object having a size corresponding to the histogram value. For example, as shown in FIG. 1B, R (red), G (green), B
Displays a three-dimensional color space with each color component of (blue) as an axis,
The histogram of the image at each grid point in the color space is displayed as an object (a circle in the example in the figure) having a size corresponding to the histogram value.

The operation input unit 6 includes a cluster classification unit 3
Instruct a predetermined cluster classification condition to
This is a part for instructing the histogram display unit 5 on the viewpoint position of the histogram represented in the three-dimensional color space. By designating the viewpoint position, the operator can refer to the three-dimensional color space from a desired viewpoint position.

Next, based on the flowchart of FIG.
An image area classification method according to the embodiment will be described. In the following description, reference numerals not shown in FIG. 2 refer to FIG.

First, as shown in step S10, a full-color image is input by the image input unit 1. The size of the input image is W pixels in the horizontal (x-axis) direction and H pixels in the vertical (y-axis) direction, that is, a total of W × H pixels.

Next, as shown in step S11, the histogram of the input image in the RGB color space is obtained by the histogram creating section 2. That is, the RGB color space is discretized at, for example, 16 × 16 × 16 = 4096 grid points,
For example, the upper 4 bits of each of the 8 bits of the RGB value of each pixel of the input image are taken and mapped to these grid points, and the result of calculating the number of times (number of pixels) mapped for each grid point is a color histogram.

Next, from step S12 to step S2
Until 0, cluster selection processing is repeated N times. This processing will be described later. Then, in step S21, the output image and the color look-up table (LUT) classified into N are output. The value of N may be predetermined at 2, 8, 16, 256, or the like, or may be selected by the operator according to the contents of the input image.

Here, from step S12 to step S2
The processing up to 0 will be described. One cluster selection process is
In step S12, the entire RGB color space is initially set as a target region for cluster classification, and thereafter, each time a cluster region is determined, the region is updated with the remaining partial space from which the cluster region has been removed.

Steps S13 to S18 are processes in which the operator repeatedly displays and confirms the distribution state of the histogram and determines the cluster area on a trial basis before determining the cluster area in step S19.

Since the histogram is a non-negative integer scalar amount distributed in a three-dimensional RGB color space, this scalar amount is converted into a circle or square (cube) object (small object,
The histogram is represented by a three-dimensional image that is represented by the size of a (figure) and that shows the distribution state of all the objects.
In this case, only an object on a grid point having a histogram value equal to or larger than a threshold value arbitrarily designated by the operator can be selectively and dynamically displayed.

FIG. 3 is a diagram showing a display example in which the threshold value of the three-dimensional histogram is changed in four ways. FIG. 3 (a)
This is a case where THL = 1000 is set as the first histogram threshold value in step S13 (see FIG. 2). In this case, the histogram value is 1 in step S14 (see FIG. 2).
A circle having a size proportional to the logarithm of the histogram value is arranged at all grid point positions of 000 or more, and step S
An image viewed from the viewpoint position designated at 15 (see FIG. 2) is generated as a three-dimensional image at step S16 (see FIG. 2) and displayed on the screen of the histogram display unit 5.

FIG. 3B shows that the second threshold value is TH.
FIG. 3C shows a case where THL = 50 is set as the third threshold value, and FIG. 3D shows a three-dimensional case where THL = 10 is set as the fourth threshold value. An image was similarly generated. The operator inputs the histogram threshold value through the operation input unit 6.

When the threshold value is increased as shown in FIG.
When the position and the maximum value of the maximum point of the histogram selected as the representative point of the cluster are known, and the threshold value is reduced as shown in FIG. 3D, the spread of the base of the cluster and its value can be confirmed. This spread is considered to indicate the noise component of the pixel density and the number of pixels.

A plurality of three-dimensional images generated for a plurality of threshold values in this way can be displayed in time series with the threshold values arbitrarily changed. When the cluster area is experimentally set in step S17, a set of objects included in the cluster area can be highlighted. As a result of the above operation, the process exits the loop in step S18 and proceeds to step S18.
At 19, the cluster region and the representative color are formally determined.

The representative color is obtained by arithmetic processing as the center of gravity of the histogram of the cluster area. This calculation is executed by the cluster classification processing unit 3.

The operation of the histogram display unit 5 will be described.
In the present embodiment, a three-dimensional image that is constantly updated following the operation is generated so that the operator can fully understand the distribution shape of the histogram.

In the generation of the three-dimensional image in step S16,
The following points were considered. (1) An operator can grasp the position and size of the cluster distribution by generating an image in which the threshold is arbitrarily changed. (2) By displaying an image whose viewpoint is arbitrarily changed, an image which is not disturbed by the occlusion relation between clusters can be obtained. (3) To facilitate understanding of the distribution of the color space, the degree of freedom of the visual field conversion is determined, for example, by setting the B axis above the screen.

FIG. 4 is a diagram showing the relationship (view conversion) between the color space coordinate system O-RGB and the viewpoint coordinate system O-xyz. In response to an instruction from the operation input unit 6, the histogram display unit 5 performs a view conversion using the following coordinate conversion formula.

View point position: OE = (Re, Ge, Be) Projection plane: OE is set on the z-axis and projected on an xy plane perpendicular to the z-axis. Project the B axis onto the y axis.

X-axis direction: outer product direction of B-axis and OE, x-axis unit vector i = (− Ge / Te, Re / Te, 0) y-axis direction: outer product direction of OE and x-axis, y-axis unit vector j
= (-BeRe / SeTe, -BeGe / SeTe, T
e / Se) z-axis direction: OE direction, z-axis unit vector k = (Re / S
e, Ge / Se, Be / Se)

Assuming that the x, y, and z coordinates of the point P (R, G, B) are (xp, yp, zp), xp = OP · i = (− GeR + ReG) / Te yp = OP · j = (− BeReR-BeGeG + TeT
eB) / SeTe zp = OP · k = (ReR + GeG + BeB) / Se

Assuming that the projection point of the point p is Q (xd, yd, 0), EQ = (xd, yd, -Se) and EP = (xp, y
p, zp-Se) are in the same direction, so xd / xp =
yd / yp = -Se / (zp-Se) = kp, xd
= Xpkp, yd = ypkp.

However, Se = sqrt (ReRe + Ge
Ge + BeBe), Te = sqrt (ReRe + GeG)
e) is obtained when the viewpoint is determined. xp, yp, zp and k
p = -SeSe / (ReR + GeG + BeB-SeS
e) is obtained at the time of coordinate transformation.

Next, a method for experimentally setting a cluster area in step S17 will be described. For example, in the example shown in FIG. 3B, since the existence of about seven clusters can be confirmed, as shown in FIG.
The case where a rectangular parallelepiped surrounding 6 is specified is taken as an example.

First, in the stereoscopic image display shown in FIG.
A rectangle is designated by dragging from an upper left point 22 to a lower right point 23 on one surface of an RGB cube, for example, a surface of R = 255, using a mouse device or the like. As a result, not only this rectangle but also a rectangular parallelepiped obtained by translating this rectangle in the R-axis direction to the plane of R = 0 is displayed on the display screen, and all objects included in the rectangular parallelepiped are displayed in an emphasized manner. Will be.

As the emphasizing method, for example, the brightness or hue of the object may be changed, or the object may be blinked. Thereby, the object set 2
6 and a part of the object set 27 are emphasized.

Thus, the target object set is represented by BG
After being surrounded by the coordinates, the section of the R coordinate shown in FIG. 5B is next specified to obtain a rectangular parallelepiped indicating the target cluster area.
To do this, the user drags and specifies between two points on the side parallel to the R axis of the first rectangular parallelepiped with the mouse. For example, points 24 and 25
Is designated, this time, an image is displayed in which only the target cuboid and the object set 26 included therein are highlighted.

Next, in order to allow the operator to determine whether or not to determine the target rectangular parallelepiped as a cluster area, the histogram threshold and the viewpoint position can be changed. If the histogram threshold is changed, the number of objects included in the rectangular parallelepiped, that is, the number of objects to be highlighted is increased or decreased, and if the viewpoint position is changed, the positional relationship between the rectangular parallelepiped and the object can be grasped from various angles. The position and size of this rectangular parallelepiped are changed so as to be optimal as a cluster area. The above operation is performed in step S shown in FIG.
This is realized by a loop that repeatedly executes steps 13 to S17.

Thus, a confirmation input is made to use this rectangular parallelepiped as a cluster area. Then, step S18
Then, the process proceeds to step S19, where the cluster region and the representative color are determined.

It should be noted that the cluster area is a set of all grid points included in the rectangular parallelepiped, and is not limited to the displayed object. The representative color is the center of gravity obtained by weighting these grid points with the histogram frequency,
Alternatively, it is obtained as a color specified by the operator. A new object having a position and a size representing the cluster may be displayed at the position of the obtained representative color.

Since one cluster area has been determined as described above, the process returns to step S12, the data of this cluster area is deleted from the histogram data to 0, the histogram is updated, a new histogram is displayed three-dimensionally, and the second histogram is displayed. , The third cluster area is repeatedly designated.

When a color space portion that does not belong to any of the cluster regions is generated by specifying the cluster region N times, the color space portion is appropriately divided and added to the default cluster region, or the grid point of the space portion is set to the closest distance. You may make it belong to the cluster of a near representative color.

As described above, in the present embodiment, the color histogram is displayed as a set of objects in the three-dimensional RGB color space, and the display result is presented to the operator so that the cluster distribution can be accurately grasped. Since the cluster area is specified, it is possible to obtain the optimum color classification result desired by the operator.

Further, by generating an image in which the histogram threshold value and the stereoscopic display viewpoint are arbitrarily changed, the position and size of the cluster distribution can be accurately grasped, and small clusters and dispersed clusters can be confirmed without being overlooked. Become like

Next, a second embodiment will be described. The second embodiment is characterized in that automatic calculation of clustering is executed after the operator designates the initial cluster centers which are N temporary representative colors. As the automatic calculation of clustering, for example, a K-means algorithm is applied.
The calculation is repeated until the cluster center position converges.

FIG. 6 is a flowchart illustrating an image area classification method according to the second embodiment. First, as shown in step S30, a full-color image is input by the image input unit 1, and as shown in step S31, the histogram of the input image in the RGB color space is obtained by the histogram creating unit 2.

Next, from step S32 to step S36
Is a temporary cluster center designation process that is repeated N times. In steps S32 and S33, setting or changing of a histogram threshold, change of display attributes of objects,
In step S34, setting or changing of a viewpoint position and generation and display of a three-dimensional image are performed.

In step S35, the center of the cluster is specified. As in the first embodiment, this includes a trial setting of the center of the cluster and a process of reconfirmation by changing the histogram threshold and the viewpoint position. And

When the N cluster centers are designated in this manner, the process exits the loop in step S36, and then proceeds to step S3
7 and the cluster center updating process in step S38 are repeated. Step S when the cluster center converges
The process exits the loop at 39 and proceeds to step S40. In step S40, an image and a color LUT, which have been converted into color numbers from the grid point classification results in the color space, are output.

Next, a specific operation in the second embodiment will be described. First, the histogram distribution HGM [16] [16] [16] obtained in step S31 is obtained in step S3.
It is assumed that an image is displayed in the process of No. 4 and that it is confirmed that the image has seven main clusters.

In step S35, the operator designates N = 7 cluster centers, and sets arrays PB [7], PG
[7] and PR [7]. Here, PB [m],
PG [m] PR [m] indicates a representative color having a color number m. (M = 0 to N-1)

After the provisional cluster center is designated by the operator as described above, the following processing is executed by the cluster classification unit 3.

(A) First, each point (all grid points) of the histogram is made to belong to the cluster at the center of the cluster closest to the distance. The classification result is stored in IDX [16] [16] [16]. IDX is 0, 1, 2, 3, 4, 5, 6 (cluster number).

That is, for each m, the distance to the cluster center is (b-PB [m]) (b-PB
[M]) + (g-PG [m]) (g-PG [m]) +
(R-PR [m]) m that minimizes (r-PR [m])
Is the value of IDX [b] [g] [r].

(B) Next, the center of each cluster is changed to the center of gravity of the histogram points belonging to the cluster. That is, for each m, IDX [b] [g] [r] =
m for all b, g, and r, (CB [m], C
G [m], CR [m]) = {HGM [b] [g]
[R] × (b, g, r)} / {HGM [b] [g]
[R]｝ is calculated.

(C) Then, a convergence check is performed. That is, for all m, (CB [m], CG [m], C
R [m]) = (PB [m], PG [m], PR [m])
Then converge.

When the convergence has occurred, the color classification result IDX [1
6] Inversely map [16] [16] to the image space.

That is, the input image is represented by B [H] [W], G
[H] [W], R [H] [W], output image = I
DX [B [y] [x], G [y] [x], R [y]
[X], y = 0,..., H-1, x = 0,.

As the color LUT, PB [m],
PG [m], PR [m], m = 0,...

On the other hand, when convergence does not occur, (CB [m], CG [m], CR [m]) is changed to (PB
[M], PG [m], PR [m]), and the above (a), (b), and (c) are executed again.

FIG. 7 is a diagram showing an example of a numerical example of the processing in the processing of the cluster classification section. FIG. 7A shows a cross section HGM [2] [g] [r] of the histogram. Two of these points are designated as the center of the initial cluster.

FIG. 7B shows the result of cluster classification of the space, and FIG. 7C shows the recalculated cluster center. The positions of these two points (m = 0 and m = 3) have not changed.

In the second embodiment, the color histogram is displayed as a set of objects in the three-dimensional RGB color space, and the display result is presented to the operator so that the cluster distribution can be accurately grasped. Since the representative color or the center of the cluster is temporarily specified, if the cluster classification is automatically calculated using the temporary color, the convergence can be achieved in a short time, and the optimum color classification result desired by the operator can be obtained quickly. Further, the assumption that the temporary cluster center moves and converges, and the progress of the cluster classification can be visually grasped by the color-coded image of the object.

Next, a third embodiment will be described. In the third embodiment, a plurality of image data created using the same set of colors is targeted, and the boundary data of the cluster region specified in the color classification for the first image is classified into any subsequent input image classification. The feature is that it can be applied.

FIG. 8 is a flowchart illustrating an image area classification method according to the third embodiment. First, in step S50, the first full-color image is input, and in step S5
In step 1, the histogram is obtained.

Steps S52 to S55 correspond to steps S12 to S21 of the first embodiment.
This is a simple illustration of the processing up to this point. Step S56
Then, the LUT of the shape and the representative color of the cluster area is stored so that it can be repeatedly used for the same type of input image thereafter.

Steps S57 to S60 show the color classification processing for the second and subsequent image data. In step S59, the saved shape data is loaded and used instead of the operator designating the cluster area. .

Next, storage and reuse of shape data will be described. In order to store and reuse the shape data, the shape is defined by a space R
It is assumed that the space can be expressed by a mathematical expression such as a rectangular parallelepiped placed inside the sphere of the GB coordinate axis, the inside of a sphere, or the like.

For example, when an inclined slender solid is designated, a parallelepiped that can be expressed as a common part of six half spaces partitioned by six planes is suitable. A cuboid placed parallel to the RGB coordinate axes is a special case.
The sphere is the only usable surface because it can be easily specified and can be processed analytically. In addition to common parts, union operations are allowed.

As described above, if expressed in an analytical and logical manner, the designated cluster area can be reused as data. In order to easily perform classification without leaving the cluster space, it is better to hierarchically divide into two on a plane parallel to the coordinate axis. The data in FIG. 9 is a diagram showing boundary surface data in the case where the RGB color cube is divided into eight simplest based on threshold values for three coordinate axes.

The representative color of the cluster area based on the shape data is not limited to the center of gravity of the histogram but may be the geometric center of the space, the maximum point of saturation, or a color specified by the operator.

In the third embodiment, an operator intervenes and instructs cluster classification for only the first image among a plurality of input images. However, other input images are initialized based on the stored shape data. Cluster classification can be set, and color classification can be automatically performed. In addition, since the boundary surface is given by a mathematical expression as shape data for cluster classification, it is possible to quickly perform cluster classification without extensive use of distance calculation.

Next, a fourth embodiment will be described. The fourth embodiment is characterized by a display method in the histogram display unit.

FIG. 10 is a diagram for explaining a display method in the fourth embodiment. In the histogram display, the display method can be changed as follows, including the display method described above.

(A) The shape of the object is displayed as a circle, a square, or a point. (B) The object is displayed in the color of the grid point position in the RGB color space. (C) The object whose histogram value is maximum in the RGB color space is displayed in a color that distinguishes it from the others. (D) The placed object is displayed as two images for binocular stereopsis. (E) The viewpoint position is displayed as a moving image that is continuously moved. These can be realized by known image generation techniques.

As shown in FIG. 10A, the following recognition effect can be obtained by displaying the object in the color unique to the lattice points in the RGB color space, that is, the same color as the pixels of the original image. .

(1) Since each image area of the original image corresponds to the cluster of the color space with the same color, the meaning and structure of the histogram can be understood without prior knowledge. (2) It is possible to easily confirm clusters and determine where to place cluster boundaries. (3) It is easy to recognize the position of the color used in the image in the color space, the distance between the colors, and the like. (4) By making the shape square, there is a visual effect as if the original image was chopped and scattered in the color space.

Further, as shown in FIG. 10B, the following recognition effects can be obtained by displaying the objects in dots.

(1) The image can be quickly updated following the operation of changing the histogram threshold, viewpoint, and cluster area. (2) The number of highlighted grid points belonging to the designated cluster area can be easily understood.

Further, as shown in FIG. 1 (b), by displaying the placed object in two images for binocular stereopsis, the distance to all objects can be recognized simultaneously.

Also, in any of the display methods described above, by sequentially moving the viewpoint position and displaying a moving image, it is possible to easily recognize the context and the distribution shape from the relative movement of all objects. Become like

As described above, by changing the shape, size, and color representation method of the object, it is possible to grasp the distribution of clusters from many sides and set an appropriate cluster area. In addition, a plurality of clusters can be visually and clearly distinguished by binocular stereovision or continuous change of the viewpoint, and it becomes possible to observe the shape of the cluster in detail.

Next, a fifth embodiment will be described. FIG.
FIG. 14 is a diagram illustrating a fifth embodiment. In the above-described embodiment, the space in which the histogram is displayed as an object is a color space in which RGB is used as a coordinate axis.
The embodiment is characterized in that an image in which an object is subjected to coordinate transformation from an RGB color space to an HSI color space having hue (H), saturation (S), and luminance (I) as coordinate axes is displayed. In addition, in FIG. 11, the outer surface of the color cube is deformed into a hexagonal pyramid, but only the 12 sides are displayed.

The histogram display unit 5 performs coordinate conversion from the RGB color space to the HSI color space using the following conversion formula.

RGB coordinate system (R, G, B) values are from 0 to 2
55 Maximum and minimum values of luminance I ... Imax = Max (R, G,
B), Imin = Min (R, G, B) Median value of luminance I... I = (Imax + Imin) / 2 Saturation S ... When I ≦ 128, S = 256 × (Imax−I
min) / (Imax + Imin), when I> 128,
S = 256 × (Imax−Imin) / (512−Im)
ax + Imin) Hue H ... When R = Imax, H = (GB) / (Ima
x−Imin) × (π / 3), when G = Imax, H =
(BR) / (Imax-Imin) × (π / 3) +2
When π / 3 and B = Imax, H = (R−G) / (Imax
x-Imin) × (π / 3) + 4π / 3 HSI coordinate system (S × cosH, S × sinH, I)

In such an HSI color space, the hue can be understood as a rotation angle around the I axis when viewed from above the I axis connecting black and white. The magnitude of the saturation can be determined based on whether the color is in close contact with the I-axis or apart from the I-axis.

In the fifth embodiment, the object is coordinate-transformed and displayed in the HSI color space having luminance, hue, and saturation as coordinate axes, whereby the structure based on hue becomes clear, and a method of cluster classification is adopted. Is easy to stand.

In the fifth embodiment, the conversion from the RGB color space to the HSI color space has been described as an example. However, the present invention is not limited to this. For example, the conversion from the RGB color space to Y (yellow) M (magenta) Y ^* color space composed of C (cyan), L ^* a ^* b
^{* The} same applies to the case of conversion to another color space such as a color space or an XYZ color space.

[0098]

As described above, the present invention has the following effects. That is, based on the image to be classified, a histogram of an image at a grid point in a color space with each color component as an axis is created, and this histogram is displayed as an object having a size corresponding to the histogram value. The histogram value in the image can be visually recognized. As a result, the operator can set the cluster classification according to his or her intention by referring to the displayed object, and perform accurate and quick color classification.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating an image area classification device according to a first embodiment.

FIG. 2 is a flowchart illustrating an image area classification method according to the first embodiment.

FIG. 3 is a diagram showing a display example in which threshold values of a three-dimensional histogram are changed in four ways.

FIG. 4 is a diagram showing a view conversion.

FIG. 5 is a diagram illustrating designation of an object set using stereoscopic image display.

FIG. 6 is a flowchart illustrating an image area classification method according to a second embodiment.

FIG. 7 is a diagram illustrating an example of a numerical example of a process in a process of a cluster classification unit.

FIG. 8 is a flowchart illustrating an image area classification method according to a third embodiment.

FIG. 9 is a diagram showing boundary surface data when an RGB color cube is divided into eight.

FIG. 10 is a diagram illustrating a display method according to a fourth embodiment.

FIG. 11 is a diagram illustrating a fifth embodiment.

FIG. 12 is a diagram illustrating a conventional image area classification device that performs color classification.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Image input part 2 Histogram preparation part 3 Cluster classification part 4 Image output part 5 Histogram display part 6 Operation input part

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Koichi Higuchi 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. F-term (reference) 5B057 AA11 BA02 CA01 CA12 CA16 CB01 CB12 CB16 CC01 CE17 DA16 DC19 DC25 5L096 AA02 CA14 DA04 FA37 GA40 GA41 JA22

Claims (10)

[Claims] 1. A histogram creating means for creating a histogram of the image at a grid point in a color space with each color component as an axis, based on an image composed of a plurality of color components, and a histogram created by the histogram creating means. A histogram display unit that displays the histogram with an object having a size corresponding to the histogram value; and a cluster classification unit that classifies all the grid points of the histogram created by the histogram creation unit with a plurality of representative values. Image area classification device. 2. The image area classification device according to claim 1, further comprising an operation input unit for instructing the histogram display unit of a threshold of a histogram value for displaying the object. 3. The apparatus according to claim 2, wherein the operation input unit instructs the histogram display unit to determine a viewpoint position serving as a base point for displaying an arrangement state of the object. 4. The apparatus according to claim 1, wherein the histogram display unit uses a circle, a square, or a point as the object. 5. The image area classification device according to claim 1, wherein said histogram display means displays said object in a color at a grid point in said color space. 6. The image area classification device according to claim 1, wherein the histogram display unit displays the object having the maximum histogram value separately from other objects. 7. The apparatus according to claim 1, wherein the histogram display unit displays the object as two images for binocular stereoscopic viewing. 8. The image area classification apparatus according to claim 1, wherein said histogram display means displays said object as a moving image in which a viewpoint position is continuously moved. 9. A histogram of the image at grid points in a color space centered on each color component is created based on an image composed of a plurality of color components, and all grid points of the histogram are represented by a plurality of representative values. An image area classification method for generating a plurality of cluster areas by classifying the histograms as follows: displaying the histogram with an object having a size corresponding to a histogram value. 10. The method according to claim 1, wherein shape data of each cluster area classified by the plurality of representative values is stored, and the shape data is read and reused when classifying the cluster area in another image. The image area classification method according to claim 9, wherein