JPH11224339A - Image area division method and image area division device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To area-divide images at a high speed and to divide the images into areas provided with an effective size not including unrequired fine areas by area-dividing thinned partial images for which object partial images are thinned by a prescribed thinning rate and integrating an isolated fine area to an adjacent area. SOLUTION: An area generation part 35 area-divides the thinned partial images for which the object partial images stored in a partial image storage part 80 are thinned by the prescribed thinning rate and calculates the number and center of the areas of the object partial images. An area establishment part 45 distributes all picture elements included in the object partial images to the most appropriate area among the areas calculated in the area generation part 35 and calculates an area set constituted of the object partial images. A fine area integration part 110 integrates the area smaller than a prescribed threshold value to the most appropriate area among the areas adjacent to it by using information established by the area establishment part 45 and stored in an area set storage part 70.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image area dividing method and an image area dividing apparatus for dividing an image represented by quantized multi-gradation image data into areas.

[0002]

2. Description of the Related Art Segmentation of an image such as an image read by an image scanner or an image generated by a computer is used as preprocessing in various fields of image processing.

[0003] For example, in image editing processing, area division is used to automatically cut out an area to be edited. In the image recognition processing, region segmentation is used to find a recognition target. Furthermore, region division can also be used when different image processing is performed for each region having a similar partial texture or color distribution state of an image.

In such a case, since the quality of the area greatly affects the performance of the entire image processing, it is required to use a high-performance area division method.

As a method of area division, there is a K-means algorithm (K-means clustering algorithm) which has been known for a long time (for example, Nakatani et al., "Detection of Object Boundary Line Based on Multiple Area Division Results", Electronic Information Communication). Journal D
2, Vol. J76-D2, No4, pp. 139-157. 914-9
16, '93 / 4).

In addition, RGB (red,
A method of performing a K-means algorithm on a five-dimensional feature space in which XY coordinates on an image plane are added to three components such as green and blue) has also been proposed (Izumi et al., "A segmentation method using color information and position information together. A Study ”, Proceedings of the 1991 IEICE Spring National Convention D-680).

Further, Japanese Patent Application Laid-Open No. Hei 8-30787 discloses that an image to be divided is divided into a plurality of block areas and a boundary area located between the respective block areas. A method has been proposed in which an image is divided into regions by dividing a boundary region that is in contact with the block region together with a small-capacity memory, in a short time, at high speed, without inconsistency between block regions.

[0008]

The area division by the K-means algorithm is a method in which the process of allocating each pixel in an image to an appropriate area is repeated until it converges to a predetermined state, and a better area is obtained little by little. is there. Therefore, a region corresponding to the feature space to be used can be obtained with high accuracy. However, since the divided areas are gradually converged to better areas, an enormous amount of time is required for processing.

The method proposed by JP-A-8-30787 can drastically reduce the processing time.
It still requires considerable time. Also, in this method, if the boundary area is set to be small, the boundary area may not work properly, and the area may be unnaturally divided. Conversely, if the boundary area is set to be large, the effect of the boundary area may be reduced. However, there is a drawback that the processing time is lengthened.

Therefore, a first object of the present invention is to enable an image to be divided into regions at higher speed.

A second object of the present invention is to make it possible to divide an image at high speed without causing unnatural area division.

A third object of the present invention is to divide an image into an area having an effective size that does not include an unnecessary small area as much as possible.

[0013]

In order to achieve the above object, an image area dividing method according to the present invention is a method for dividing an input image to be divided represented by quantized multi-tone input image data into a plurality of blocks. When the target image is divided into the target block image and the target partial image including the pixels included in the calculated region adjacent to the target block image, a partial image region dividing step is performed by sequentially dividing different block images in a plurality of block images. Is an image area dividing method for repeatedly dividing an input image into regions as a block image of interest. In the partial image region dividing step, the target partial image is divided into regions by decimating a thinned partial image at a predetermined thinning rate. An area calculating step of calculating the number and center of the areas of the partial image;
An area determination step of allocating all pixels included in the target partial image to the area calculated in the area calculation step, and a micro area integration step of integrating an isolated micro area into an adjacent area. And

Further, in the image area dividing method according to the present invention, in the micro area integrating step, an area smaller than a predetermined threshold is selected as a micro area, and the selected micro area is adjacent to the selected micro area. And integrating into an adjacent area having an area larger than a predetermined threshold value.

Further, in the image area dividing method according to the present invention, the area calculating step divides the first partial image obtained by thinning the target partial image at the first thinning rate, and roughly calculates the area of the target partial image. Estimating the number and the center of the region, and using the number and the center of the regions estimated in the region estimating step, the target partial image is thinned at a second thinning rate smaller than the first thinning rate. And a region generation step of calculating the number and center of the regions of the target partial image by dividing the thinned partial image into regions.

Further, in the image area dividing method according to the present invention, the area calculating step, the area estimating step, the area generating step, or the partial image area dividing step is performed by three components of the color vector of the pixel and the coordinates on the image plane. In the dimensional feature space, the thinned image, the target partial image, or the thinned partial image is divided into regions by a K-means algorithm.

Further, in the image area dividing method according to the present invention, the area determining step allocates the pixels to an area having the smallest weighted distance in the five-dimensional feature space based on the three components of the color vector of the pixel and the coordinates on the image plane. It is characterized by the following.

The image dividing apparatus according to the present invention further comprises an image input unit for obtaining quantized multi-tone input image data representing an input image to be divided, and an image input unit for dividing the input image into a plurality of block images. A target partial image extracting unit for sequentially extracting a target partial image composed of pixels included in a target block image and a calculated area adjacent to the target block image by sequentially using different block images in a plurality of block images as a target block image And a partial image region dividing unit that sequentially divides the target partial images sequentially extracted by the target partial image extracting unit, and thins out the target partial images extracted by the target partial image extracting unit at a predetermined thinning rate. A region calculating unit that divides the thinned partial image into regions and calculates the number and center of the regions of the target partial image extracted by the target partial image extracting unit; Region determining unit that distributes all the pixels included in the target partial image extracted by the target partial image extracting unit to the isolated region, and a minute region integrating unit that integrates the isolated minute region into an adjacent region And a dividing unit.

Further, in the image area dividing apparatus according to the present invention, the minute area portion selects an area smaller than a predetermined threshold as a minute area, and selects the selected minute area as
It is a means for integrating into an adjacent area adjacent to the selected minute area and having an area larger than a predetermined threshold.

Further, in the image area dividing apparatus according to the present invention, the area calculating section divides the area of the first thinned partial image obtained by thinning the target partial image extracted by the target partial image extracting section at a first thinning rate. A region estimating unit that estimates the approximate number and center of the regions of the target partial image extracted by the target partial image extracting unit; and a target partial image using the number and center of the regions estimated by the region estimating unit. A second thinned-out partial image obtained by thinning out the target partial image extracted by the extracting unit at a second thinning-out rate smaller than the first thinning-out rate is divided into regions,
A region generation unit that calculates the number and center of the regions of the target partial image extracted by the target partial image extraction unit.

Further, in the image region dividing apparatus of the present invention, the region calculating unit, the region estimating unit, the region generating unit, or the partial image region dividing unit includes three components of the color vector of the pixel and coordinates based on the coordinates on the image plane. In the dimensional feature space, the thinned image, the target partial image, or the thinned partial image is divided into regions by a K-means algorithm.

Further, in the image area dividing apparatus according to the present invention, the area determining section allocates the pixel to an area having the smallest weighted distance in the five-dimensional feature space based on the three components of the color vector of the pixel and the coordinates on the image plane. It is characterized by the following.

Further, the present invention relates to a program recording medium on which a program for executing each of the above-described image area dividing methods is recorded.

[0024]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a diagram for explaining the configuration of an image area dividing apparatus for dividing an input image into areas by utilizing image thinning in the image area dividing apparatus according to the present invention.
This image area dividing device includes an image input unit 10, an area estimation unit 20, an area generation unit 30, and an area determination unit 40.

The image input unit 10 receives an input image I to be divided.
It obtains quantized multi-tone input image data Di indicating i, and stores it in a storage device such as an image scanner that reads an image on paper or a film, a digital camera that shoots a subject image, or a hard disk. A device that reads out image data that has been transmitted, a device that receives and captures image data transmitted via a network, or a device that generates image data by drawing with a computer.

In the image input unit 10, the color system of the input image data is converted to the color system used in the subsequent processing, if necessary. In the following example, the input image data is
It is obtained as RGB image data, converted into L ^* a ^* b ^* image data, and represented by a two-dimensional map of a color vector having L ^* , a ^* , and b ^* components. In addition, the image input unit 10 includes a storage device such as a memory that stores part or all of the input image data as necessary.

The region estimating section 20 divides the input image represented by the above-mentioned input image data into thinned-out images at a predetermined thinning-out rate Ra, and divides the region into the approximate number and center of the regions of the input image. Using the number and center of the regions estimated by the region estimating unit 20, the region generating unit 30 performs region division on the thinned image obtained by thinning the input image at the thinning ratio Rb smaller than the above thinning ratio Ra. Then, the number and center of the regions of the input image are calculated. The following example shows a case where simple thinning is used.

As shown in FIG. 9A, an input image Ii has pixels Pi continuous in the X direction (horizontal direction) and the Y direction (vertical direction).
9 is thinned out in the X direction and the Y direction at a thinning rate Ra.
A thinned-out image Ia composed of some pixels such as a pixel Pa painted out in (B), and a pixel Pb filled out in FIG. 9C, in which pixels Pi are thinned out at a thinning-out rate Rb in the X and Y directions. Thinned image I consisting of some pixels like
b. The thinning rate R (Ra or Rb) means that one pixel is extracted for every R2 pixels.

Then, the area estimating section 20 divides the thinned image Ia into areas as described later and estimates the approximate number and center of the areas of the input image Ii.
Then, using the estimation result, the thinned image Ib is divided into regions as described later, and the number and center of the regions of the input image Ii are calculated.

The area determining section 40 allocates all pixels included in the input image Ii to the most appropriate area among the areas calculated by the area generating section 30, as described later.

FIG. 2 shows an image area dividing method corresponding to the image area dividing apparatus shown in FIG. 1. First, in step S20 in the area estimating section 20, the thinned image with the thinning rate Ra is divided into areas, and the input image is divided. The approximate number and center of the regions are estimated, and then, in step S30 in the region generation unit 30, the thinned image is divided into regions by the thinning rate Rb using the estimation result, and the number and center of the regions of the input image are determined. Is calculated, and in step S40 in the area determination unit 40, all pixels included in the input image are allocated to the calculated area.

Area estimation at step S20, step S20
The region generation at 30 and the region determination at step S40 are performed by a clustering method obtained by improving the K-means algorithm as described below. This is shown in FIGS.
Since this is the same as the case where a target partial image described later and shown in FIG. 8 is replaced with the input image itself, FIGS. 6 to 8 are shown by diversion.

In step S20, as shown in FIG. 6 (however, the target partial image in step S202 is replaced with an input image), first in step S201, the number of clusters K is appropriately selected, and an initial area is set for the input image. The center is set, and in step S202, one pixel of the input image is thinned out at a thinning rate R = Ra to select one.

The selection of the pixel, for example, a feature vector P _n = the pixels of the input image _{(L n, a n, b} n, X n, Y n)
(Where 1 ≦ n ≦ the total number of pixels of the input image), by sequentially selecting pixels that satisfy the condition of the following equation (1) one by one.

X _n mod R = R / 2 and Y _n mod R = R / 2 ‥‥ (1) where L _n , a _n , and b _n are three components of the pixel color vector, and in the above case, L ^* , A ^* , and b ^* components, and _Xn , Y
_n is the X, Y coordinates of the pixel on the image plane. In addition, mod is an operator for obtaining a remainder, and the result of the division in the expression (1) is rounded down to a decimal point to obtain an integer value.

Next, in step S203, the feature vector P _n of the selected pixel and the center of each area V _m = (L _m , a _m , b
_m , X _m , Y _m ), and search for the center of the area having the distance d _min closest to the selected pixel. The distance d is calculated by, for example, the following equation (2).

[0037] _{d = {k L (L n} -L m) + k a (a n -a m) + k b (b n -b m) + k X (X n -X m) + k Y (Y n -Y m )} ^1/2 ‥‥ (2) _{where, L m, a m, b} m are L ^* color vector area centered V _m, a ^*, a b ^* components, X _m, Y _m is The X and Y coordinates of the region center _{Vm on} the image plane. Also, k _L , k _a ,
k _b , k _X , and k _Y are coefficients for scaling the feature axes corresponding to the L ^* component, a ^* component, b ^* component, X coordinate, and Y coordinate, respectively. In this example, since the color vector information is more important than the coordinates, 0 is set for k _L , k _a , and k _b = 1.
Set <k _X , k _Y <1.

However, in this example, in order to make the search for the center of the area more efficient, only the center of the area whose distance d from the selected pixel is smaller than a predetermined threshold value d _{th is} set as a search candidate, and the search candidate The center of the area having the distance d _min closest to the selected pixel is searched from the inside.

Next, in step S204, the closest distance d _min is set to a predetermined threshold T _diff (T _diff
It is determined whether or not the selected pixel is smaller than <d _th ) to determine whether or not the selected pixel can be relocated to the area at the center of the area having the closest distance d _min .

That is, when the distance d _min is smaller than the threshold T _diff , it is determined that the selected pixel can be rearranged in the center of the area having the distance d _min, and the process proceeds to step S205, where the distance d _min is _{equal to} or _larger than the threshold T _diff. In some cases, it is determined that the selected pixel cannot be rearranged in the region at the center of the region having the distance d _min , and the process proceeds to step S206.

In step S203, even when there is no area center of the search candidate whose distance d from the selected pixel is smaller than the threshold value d _{th, an} appropriate area center to which the selected pixel is to be rearranged exists. If not, step S20
The process proceeds from Step 4 to Step S206.

In step S205, the selected pixel is distributed to the region at the center of the region having the closest distance d _min within a range smaller than the threshold value T _diff .

In step S206, an area composed of the selected one pixel is newly generated. Specifically, a feature vector composed of the L ^* , a ^* , and b ^* components of the color vector of one selected pixel and the XY coordinates on the image plane of the pixel is generated, and this is set as the center of the area. One selected pixel is allocated to the region at the center of the region. As a result, the area center or the total number of areas increases by one.

After rearranging the pixels in step S205 or newly generating an area consisting of one pixel in step S206, next in step S207, step S20 is performed.
It is determined whether all the pixels to be selected in Step 2 have been selected,
If not, the process returns to step S202 to select the next pixel to be selected, and repeats the processing from step S203 on for the selected pixel.

When it is determined in step S207 that all the pixels to be selected in step S202 have been selected,
Next, the process proceeds to step S208, and the average of the feature vectors of the pixels rearranged in the region center of each region in step S205 is calculated as a new region center for each region.

Next, in step S209, the calculated new area center is replaced with the previous area center. At this time, the distance in the feature space between the center of the new region and the center of the previous region is calculated for each region, and if the distance is equal to or greater than a predetermined threshold, it is determined that the region has not converged yet. Set the flag to indicate

Next, in step S210, it is determined whether or not the flag has been set, and it is determined whether or not the area has converged. If the flag has been set and the area has not converged, the process proceeds to step S211. Proceeding, after removing the pixels rearranged from the region at the center of each region, the process returns to step S202, and the above processing is repeated from the selection of the pixel.

In step S210, when it is determined that the area has converged because the flag is not set, the area estimation processing ends.

In step S30, as shown in FIG. 7 (however, the target partial image in step S302 is replaced with the input image), first, in step S301, the region center calculated in the region estimation step S20 is set as the initial region center. Set. However, the pixels rearranged in the region at the center of each region are deleted.

Next, in step S302, one pixel of the input image is thinned out at a thinning rate R = Rb (Rb <Ra), and one is selected. The selection of the pixels is performed, for example, by sequentially selecting the pixels satisfying the condition of the above equation (1) one by one, similarly to the selection of the pixels in the step S202 of the area estimation step S20 in FIG.

Step S303 and subsequent steps are the same as step S203 and subsequent steps of the area estimation step S20 in FIG.

In step S40, as shown in FIG. 8 (however, the target partial image in step S402 is replaced with the input image), first in step S401, the area center calculated in the above-described area generation step S30 is set as the initial area center. Set. However, the pixels rearranged in the region at the center of each region are deleted.

Next, in step S402, all the pixels included in the input image are selected one by one. This can be achieved by setting the thinning rate R to 1. However, it is useless to perform the condition determination of the above equation (1) despite selecting all the pixels included in the input image in order. Therefore, in the area determination step S40, a flag for performing only the rearrangement is created, and if this flag is set, the pixels of the input image can be selected unconditionally in order.

Next, in step S403, unlike step S203 of region estimation step S20 and step S303 of region generation step S30, the region having the closest distance to the selected pixel is unconditionally different from step S203 of region estimation step S20 and step S303 of region generation step S30. The center is searched, and then, in step S405, in order to improve the processing efficiency, step S20 of region estimation step S20 is performed.
4 and the region estimation step S2 without making the determination as in the step S304 of the region generation step S30.
The selected pixel is distributed to the region at the center of the region having the closest distance without newly generating a region consisting of one selected pixel as in step S206 of step S206 and step S306 of the region generation step S30. .

There is a trade-off between the efficiency of this processing and the accuracy of the generated area, and the thinning rate Rb in the area generating step S30 is adjusted according to the required division content.

Next, in step S407, step S40
It is determined whether or not all the pixels of the input image have been selected in step 2, and if not all the pixels have been selected, step S4
Returning to step 02, the next pixel to be selected is selected, and the processing from step S403 is repeated for the selected pixel.

Step S408 and subsequent steps are the same as step S208 and subsequent steps of the area estimation step S20 and step S308 and subsequent steps of the area generation step S30.

The above example is the case where simple thinning is used in the area estimation step S20 and the area generation step S30.
Any thinning method may be used as long as the number of pixels is smaller than that of the input image. For example, an average value or a median value of R ² (Ra ² or Rb ² ) pixels of the input image may be used as the thinned image. Can be.

According to the above-described image area division, the number and center of the areas of the input image are calculated from the thinned image having the predetermined thinning rate Rb, and all the pixels included in the input image are allocated to the calculated area. The time required for convergence of the region can be greatly reduced, the input image can be divided into regions at high speed, and in calculating the number and the center of the region, the region of the input image can be reduced by the thinned image having a larger thinning rate Ra. Estimate the approximate number and center of,
Using the estimation result, the number and center of the areas of the input image are calculated from the thinned image of the predetermined thinning rate Rb, so that the number and center of the areas of the input image are calculated from the thinned image of the predetermined thinning rate Rb from the beginning. Compared to when
The time required to calculate the number and center of the regions can be further reduced.

FIG. 3 shows an image area dividing apparatus which divides an input image into areas by utilizing image blocking. The image area dividing device includes an image input unit 10, a target partial image extracting unit 50, and a partial image area dividing unit 60. The image input unit 10 is the same as that of FIG.

The target partial image extracting unit 50 is provided with the image input unit 1
When the input image obtained as input image data at 0 is divided into a plurality of block images, a target partial image including pixels included in a calculated block adjacent to the target block image and the target block image is divided into a plurality of block images. Different block images are sequentially extracted as a block image of interest. However, it is assumed that the target partial image holds the three components of the color vector of the pixel and the XY coordinates of the pixel on the image plane. The target partial image extracting unit 50 includes a storage unit for temporarily storing the extracted target partial image.

The partial image area dividing section 60 sequentially converts the target partial images sequentially extracted by the target partial image extracting section 50,
Divide the area. In addition, the partial image region dividing unit 60 stores the region data calculated by the region division in the internal storage unit, determines whether or not the processing for all the block images is completed, and outputs the determination output to the target region. The image is sent to the image extracting unit 50.

FIG. 4 shows an image area dividing method corresponding to the image area dividing apparatus shown in FIG. 3. First, in step S50, a target partial image is extracted by using one of a plurality of block images as a target block image. Next, in step S60,
The target partial image is divided into regions, and then, in step S70
Then, it is determined whether or not the final partial image including the last block image has been divided into regions. If the final partial image has not been divided into regions, the process returns to step S50, and the next block image is set as the target block image. The target partial image is extracted, and in step S60, the target partial image is divided into regions. When it is determined in step S70 that the final partial image has been divided into regions, the region division of the input image ends.

For example, as shown in FIG. 10B, the input image is divided into six parts in the X direction and four parts in the Y direction.
The image is divided into block images f11, f12 ‥‥ f45, f46. Then, for example, as shown in FIG. 10C, f11, f12... F16, f21, f22.
26, f31, f32 {f36, f41, f42}
The block image of interest is set in the order of f46. However, the order of setting the block image of interest is not limited to this, and may be arbitrary. It is also possible to simultaneously process two or more block images as a block image of interest in parallel.

The block image of interest is extracted, for example, as follows. First, the width (number of pixels) W and the height (number of pixels) H of the input image Ii as shown in FIG.
Approximate size (number of pixels) in the X and Y directions of one predetermined block image, respectively
Then, the number of divisions Nx and Ny in the X and Y directions when the input image Ii is divided in the X and Y directions is calculated by dividing the input image Ii in the X and Y directions.

Next, according to the following equation (3), the upper left pixel of the block image f _ij (1 ≦ i ≦ Nx, 1 ≦ j ≦ Ny) as shown in FIG. the coordinates (x _i, y _j) is calculated, the equation (4), and calculates the block image f _ij width (number of pixels) w _ij and height (number of pixels) h _ij.

[0067] _{(x i, y j) =} (W (i-1) / Nx, H (j-1) / Ny) ‥ (3) (w ij, h ij) = (x i + 1 -x i , Y _{i + 1} −y _i ) (4) where the XY coordinates of the pixel are, as shown in FIG. 10B, the coordinates of the upper left pixel of the input image Ii (0,
0), it increases as the horizontal direction goes to the right and the vertical direction goes down. Also, the result of the division in equation (3) is rounded down to the nearest whole number to obtain an integer value.

As shown in FIG. 12B, x in the equation (4)
_{i + 1} is the block image f on the right of the block image of interest f _{ij
(i + 1) j} is the X coordinate of the upper left pixel, and y _{i + 1} is the Y coordinate of the upper left pixel of the block image f _{i (j + 1) immediately} below the block image of interest f _ij .

The target partial image extracting section 50 extracts the block image of interest in this way, and uses the area data obtained as a result of the area division of the partial image previously performed by the partial image area dividing section 60 as described later. Then, a calculated area adjacent to the target block image is extracted, and pixels included in the target block image and the calculated area adjacent thereto are set as a target partial image. However, when the first block image, for example, the block image f11 in FIG. 10B is extracted as the block image of interest, the partial image region dividing unit 60 has not divided the partial image yet, and has calculated the region. Does not exist, only the pixels included in the first block image are extracted as the target partial image.

Therefore, for example, when the block images f11 ‥‥ f46 shown in FIG. 10B are to be used as the block images of interest in the order shown in FIG. 10C, first, the target partial image extracting unit 50 Then, only the block image f11 is extracted as the target partial image, and the partial image region dividing unit 60
In FIG. 11, the block image f11 is divided into regions, for example, as shown in FIG. 11A, four regions Aa, A
b, Ac, and Ad are calculated.

Next, in the target partial image extracting unit 50, FIG.
As shown by hatching in FIG. 1 (B), the block image f1
2 is extracted as a block image of interest, the regions Ab and Ac are extracted as calculated regions adjacent to the block image of interest f12, and the block image f12 and the regions Ab,
The partial image G12 of Ac is extracted as a target partial image, and the partial image region dividing unit 60 divides the target partial image G12 into regions, for example, as shown in FIG. Areas Ab and Ac and a new 4
The areas Ae, Af, Ag, and Ah are calculated.

FIG. 12A shows a state in which the region has been divided up to the target partial image having the block image f21 adjacent to the left of the block image f22 as the target block image in this manner. When the block image f21 is set as the target block image, the calculated area adjacent thereto is the area A, as is clear from FIG.
a, Ac, and Ad.

In the following, a set of areas calculated by area division of a target partial image including a certain block image f _ij and a calculated area adjacent thereto is referred to as an area set C _ij.
And The region set C _ij is stored in the partial image region dividing unit 60 in association with the block image f _ij , and is used for subsequent extraction of a calculated region adjacent to the target block image by the target partial image extracting unit 50.

The extraction of the calculated area adjacent to the target block image is performed as follows. For example, when each block image is set as a target block image in the order as shown in FIG. 10C, as shown in FIG. 12B, the calculated area adjacent to the target block image f _ij A region including a pixel whose X coordinate is equal to x _i -1 in a region set C _{(i-1) j} corresponding to a block image f _{(i-1) j} adjacent to the left of the block image f _ij ; The block images f _{i (j−1)} and f adjacent immediately above and right above the image f _ij
The region sets C _{i (j−1)} and C _{i (j−1)} corresponding to _{(i + 1) (j−1)
(i + 1) (j-} 1) in, X coordinate x _i or more, less than x i _{+ 1,} and Y coordinate is an area including pixels equal to y _i -1.

That is, a pixel whose XY coordinates are in (x _i -1, y _i ) or a hatched portion below it, or (x _i , y _i -1) or a hatched portion to the right thereof The area including the pixel inside is a calculated area adjacent to the block image of interest f _ij , and the area is searched for from the XY coordinates of the pixel, and the pixels included in the searched area are extracted. , A calculated area adjacent to the block image of interest f _ij can be extracted.

However, the block image of interest is shown in FIG.
Are located at the upper end of the input image as in block images f11 to f16, or block images f11 and f2.
When located at the left end of the input image as in 1, f31, and f41, there is no area adjacent to the upper or left side of the target block image, and therefore, pixels are extracted from the area adjacent to the upper side or left side. Is not done.

When the block image f22 in FIG. 10B is used as the target block image, the block image f22 shown in FIG.
If the region division as shown in (1) is performed, the pixels included in the regions Ac, Ag, Ah, and Aq are extracted as the calculated regions adjacent to the block image f22.

Step S6 in the partial image area dividing section 60
At 0, the target partial image is region-divided by a clustering method obtained by improving the K-means algorithm as described below. This is the same as the case where the input image is replaced with the above-mentioned target partial image, which will be described later with reference to FIG.

First, in step S101, the center of the initial area is set for the target partial image. The center of the initial region can be arbitrarily given to the block image of interest. For example, a feature vector of a pixel located at the center of gravity of the block image can be set as the center of the initial region. Four initial region centers are set based on feature vectors of pixels located at the center of gravity of a rectangular region obtained by dividing the image into four regions. For a calculated area adjacent to the block image of interest, the center of each calculated area is set as the initial area center.

Next, in step S102, one pixel included in the target partial image is selected, and then in step S103,
The feature vector P _n = (L _n , a _n ,
b _n , X _n , Y _n ) (where 1 ≦ n ≦ the total number of pixels of the target partial image) and each region center V _m = (L _m , a _m , b _m , X _m ,
Y _m ) is calculated, and the center of the area having the distance d _min closest to the selected pixel is searched. The distance d is
For example, it is calculated by the aforementioned equation (2). Each component of the feature vector P _n and the area center V _m is the same as in the above-described embodiment.

However, also in this example, in order to make the search for the center of the area more efficient, only the center of the area whose distance d from the selected pixel is smaller than a predetermined threshold d _{th is} set as a search candidate, and the search candidate The center of the area having the distance d _min closest to the selected pixel is searched from the inside.

In the following steps S104 and subsequent steps, step S204 of the area estimation step S20 shown in FIG.
This is the same as that after step S304 of the area generation step S30 shown in FIG. However, step S
At 107, it is determined whether or not all the pixels of the target partial image have been selected at step S102.

As described above with reference to FIG.
After the target partial image is divided into regions at 60, in step S70, including the last block image, for example,
When the block images f11 ‥‥ f46 shown in FIG. 10B are used as the block images of interest in the order shown in FIG. 10C, it is determined whether or not the final partial image including the block image f46 is divided into regions. If it is determined that the final partial image has not been divided into regions, the process returns to step S50 to extract the next target partial image using the next block image as the target block image, and in step S60, as shown in FIG. Next, the target partial image is divided into regions. When it is determined in step S70 that the final partial image has been divided into regions, the region division of the input image ends.

According to the above-described image region division, by dividing the image into blocks, each region division can be performed at high speed by dividing a small number of pixels into a small number of regions.
As a whole, the input image can be divided into regions at high speed. In addition, since the regions adjacent to the block images are processed in an overlapping manner, there is no inconsistency between the block images, and no unnatural region division occurs.

FIG. 5 and FIGS. 6 to 8 show a case where the image segmentation is combined with the image blocking in the image region dividing apparatus and the image region dividing method described above.

The image area dividing device shown in FIG.
0, image storage unit 15, target partial image extraction unit 50, region set storage unit 70, partial image storage unit 80, region estimation unit 20,
It is configured by an area generation unit 30, an area determination unit 40, an end determination unit 90, and a statistical information calculation unit 100.

The image input unit 10 is the same as that of FIGS. 1 and 3. Also in this example, the input image data is RGB
It is assumed that the image data is obtained as image data, converted into L ^* a ^* b ^* image data, and represented by a two-dimensional map of a color vector having L ^* , a ^* , and b ^* components.

The image storage unit 15 is a storage device such as a memory for storing a part or all of the input image data, which is omitted in FIG. 1 and FIG. 3 because it is included in the image input unit 10. A part of the input image data from which the target partial image can be extracted by the image extracting unit 50 is stored. Therefore, the image input unit 10 generates image data of a part necessary for the processing in synchronization with the processing in the target partial image extracting unit 50 and writes the image data in the image storage unit 15. However, when a sufficient capacity can be used as the image storage unit 15, the entire input image data can be read at a time and written to the image storage unit 15.

The area set storage unit 70 stores the area data calculated by the area determination unit 40 as described later. This area data can be represented by, for example, general mask information. In this example, a list of XY coordinates of pixels constituting an area on an image plane is used as area data, and a set of the area data is used as an area set. And

The target partial image extraction unit 50 uses the area data stored in the area storage unit 70 and the area estimation unit 2 based on the image data stored in the image storage unit 15.
0, a target partial image composed of the above-described block image of interest and a calculated region adjacent thereto which is to be processed by the region generation unit 30 and the region determination unit 40 is extracted. The target partial image in this case also holds the three components of the color vector of the pixel and the XY coordinates of the pixel on the image plane.

As a method of extracting the target partial image in the target partial image extracting section 50, the same method as described above can be used. When the pixels included in the calculated area adjacent to the target block image are extracted by the target partial image extracting unit 50 in addition to the pixels included in the target block image, the area of the extracted calculated area is determined. The data is deleted from the area set storage unit 70.

The partial image storage unit 80 temporarily stores the target partial image extracted by the target partial image extraction unit 50,
Area estimation section 20, area generation section 30, and area determination section 40
Provide for processing at When the processing is completed, the target partial image can be discarded.

The area estimating section 20 divides the target partial image stored in the partial image storage section 80 into thinned-out partial images at a predetermined thinning rate Ra to divide the target partial image into areas. Estimate the number and center of.

Using the number and center of the areas estimated by the area estimating section 20, the area generating section 30 thins out the target partial image stored in the partial image storing section 80 into a thinning rate smaller than the above thinning rate Ra. The thinned partial image thinned out at the rate Rb is divided into regions, and the number and center of the regions of the target partial image are calculated.

The area determination section 40 distributes all the pixels included in the target partial image stored in the partial image storage section 80 to the most appropriate area among the areas calculated by the area generation section 30. , An area set constituted by the target partial image is calculated. The area set calculated by the area determination unit 40 is stored in the area set storage unit 70 as described above.

The region estimating unit 20, the region generating unit 30, and the region determining unit 40 perform the region estimating step S20 and the region estimating step S20 in FIGS. 6 to 8 except that each process is performed on the target partial image instead of the input image. The same can be applied to the generation step S30 and the area determination step S40.

[0097] The target partial image is extracted by the target partial image extraction unit 50, the extracted target partial image is written into the partial image storage unit 80, and the area is estimated for the target partial image written into the partial image storage unit 80. The processing in the section 20, the area generation section 30 and the area determination section 40, and the writing of the area set calculated in the area determination section 40 to the area set storage section 70 are performed by sequentially changing the block image of interest in the target partial image. , And so on.

The end determination section 90 determines whether or not processing has been completed for all block images. If there is an unprocessed block image, the processing after the target partial image extraction section 50 is repeated.

The statistical information calculation section 100 includes an end determination section 90
When it is determined that the processing for all the block images has been completed, statistical information such as the number of pixels constituting the region and the variance of the color vector is calculated from the region set stored in the region set storage unit 70. Integrate or delete areas as needed. This processing depends on how the area obtained by the area division is used, and is not the gist of the present invention, so that a specific example is omitted here.

In the area estimating section 20, as shown in FIG.
Then, the center of the initial region is set for the target partial image. In the same manner as described above with reference to FIG. 13, for the target block image, in this example, the feature vector of the pixel located at the center of gravity of the rectangular area obtained by dividing the target block image into four is used.
One initial region center is set, and for the calculated regions adjacent to the block image of interest, each calculated region center is set as the initial region center.

Next, in step S202, one pixel of the target partial image is thinned out at a thinning rate R = Ra to select one. The selection of the pixel, for example, feature vectors of the pixels of the target partial image _{P n = (L n, a} n, b n, X n, Y n) ( provided that, 1 ≦
(n ≦ the total number of pixels of the target partial image) by sequentially selecting pixels that satisfy the condition of the above equation (1) one by one.

The subsequent steps S203 and subsequent steps are the same as those in the above-described embodiment.

In the area generating section 30, as shown in FIG. 7 as an area generating step S30, first, in step S301.
Then, as the initial region center, the above-described region estimation step S2
The center of the area calculated at 0 is set. However, the pixels rearranged in the region at the center of each region are deleted.

Next, in step S302, one pixel of the target partial image is thinned out at a thinning rate R = Rb (Rb <Ra), and one is selected. The selection of the pixel is performed, for example, by sequentially selecting the pixels satisfying the condition of the above-described equation (1) one by one, similarly to the selection of the pixel in step S202 of the area estimation step S20 in FIG.

The following step S303 and subsequent steps are the same as step S203 and subsequent steps of the area estimation step S20 in FIG.

In the area determination section 40, as shown as an area determination step S40 in FIG.
Then, as the initial region center, the region generation step S3
The center of the area calculated at 0 is set. However, the pixels rearranged in the region at the center of each region are deleted.

Next, in step S402, all the pixels included in the target partial image are selected one by one. This can be achieved by setting the thinning rate R to 1. However, it is useless to perform the condition determination of the above expression (1) despite selecting all the pixels included in the target partial image in order. Therefore, in the region determination step S40, a flag for performing only the rearrangement is created, and if this flag is set, the pixels of the target partial image can be unconditionally selected in order.

The following steps S403 and subsequent steps are the same as those described in the above embodiment. However, in step S407, it is determined whether all the pixels of the target partial image have been selected.

The above example is a case where simple thinning is used in the area estimation step S20 and the area generation step S30. The feature of the color distribution of the target partial image is retained as much as possible, and the total number of pixels is smaller than that of the target partial image. Any thinning method may be used as long as it is a thinning method. For example,
The average value or median value of R2 (Ra2 or Rb2) pixels of the target partial image can be used as the thinned partial image.

According to the above-described embodiment, the area division can be performed at a much higher speed, and no inconsistency between the block images occurs, and no unnatural area division occurs.

FIG. 13 shows another embodiment of the image area dividing method.
This is a case where an input image is divided into regions by a clustering method obtained by improving the average algorithm.

In this image area dividing method, first, in step S101, the number of clusters K is appropriately selected, an initial area center is set for an input image, and then step S102
, One pixel included in the input image is selected.

[0113] Then, in step S103, the feature vector P _n of the selected pixel _{= (L n, a n,} b n, X n, Y n)
(Where 1 ≦ n ≦ the total number of pixels of the input image) and the distance d between each region center V _m = (L _m , a _m , b _m , X _m , Y _m ) are calculated, and Search for the center of the area with a close distance d _min . The distance d is calculated by, for example, the above-described equation (2). Each component of the feature vector P _n and the area center V _m is the same as in the above-described example.

However, in this case, only the center of the area whose distance d from the selected pixel is smaller than a predetermined threshold value d _{th is} set as a search candidate, and the distance d closest to the selected pixel is selected from the search candidates. Search for the region center with _min .

Next, in step S104, the closest distance d _min is set to a predetermined threshold T _diff (T _diff
It is determined whether or not the selected pixel is smaller than <d _th ) to determine whether or not the selected pixel can be relocated to the area at the center of the area having the closest distance d _min .

That is, when the distance d _min is smaller than the threshold T _diff , it is determined that the selected pixel can be rearranged in the center of the area having the distance d _min , and the process proceeds to step S105, where the distance d _min is _{equal to} or _larger than the threshold T _diff. In some cases, it is determined that the selected pixel cannot be rearranged in the center area of the area having the distance d _min , and the process proceeds to step S106.

In step S103, even when there is no area center of the search candidate whose distance d from the selected pixel is smaller than the threshold value d _{th, an} appropriate area center to which the selected pixel should be relocated exists. If not, step S10
The process proceeds from Step 4 to Step S106.

In step S105, the selected pixel is distributed to the region at the center of the region having the closest distance d _min within a range smaller than the threshold value T _diff .

In step S106, a new area consisting of the selected one pixel is newly generated. Specifically, a feature vector composed of the L ^* , a ^* , and b ^* components of the color vector of one selected pixel and the XY coordinates on the image plane of the pixel is generated, and this is set as the center of the area. One selected pixel is allocated to the region at the center of the region. As a result, the area center or the total number of areas increases by one.

Step S107 and subsequent steps after step S105 or S106 are the same as step S207 and subsequent steps of the area estimation step S20 shown in FIG. 6 or steps S307 and subsequent steps of the area generation step S30 shown in FIG. However, in step S107, it is determined whether or not all the pixels of the input image have been selected in step S102.

According to the above-described image area dividing method, only the center of an area whose distance d from the target pixel is smaller than a predetermined threshold value d _th is set as a search candidate, and the distance d closest to the target pixel is selected from the search candidates. _{Since the} center of the area having _min is searched, the search for the area where the target pixel should be rearranged can be made more efficient,
Region division can be speeded up.

When the center of the area of the search candidate does not exist, or when the distance d _min between the center of the area having the distance d _min closest to the target pixel and the target pixel is _equal to a predetermined threshold T _diff.
In the above case, since the region including one target pixel is generated before the region converges, the time required for the region to converge can be reduced.

FIG. 14 is a block diagram showing a configuration of an area dividing apparatus which divides an area by combining image blocking and image thinning in the image area dividing apparatus and the image area dividing method.

The image area dividing apparatus of this embodiment includes an image input section 10, an image storage section 15, a target partial image extraction section 50, an area set storage section 70, a partial image storage section 80, and an area generation section 3.
5, an area determination unit 45, an end determination unit 90, and a statistical information calculation unit 100.

The image input unit 10, the image storage unit 15, the target partial image extraction unit 50, the area set storage unit 70, the partial image storage unit 80, the end determination unit 90, and the statistical information calculation unit 100 are the same as those shown in FIG. Can be configured in exactly the same manner as the one having the same number in the image area dividing device.

The area generating section 35 divides the target partial image stored in the partial image storage section 80 into thinned-out partial images at a predetermined thinning-out rate Ra, and divides the target partial image by the number of areas of the target partial image. And calculate the center.

The area determination section 45 allocates all pixels included in the target partial image stored in the partial image storage section 80 to the most appropriate area among the areas calculated by the area generation section 35. The area calculated by the area determining unit 45 that calculates the area set formed by the target partial image is stored in the area set storage unit 70.

As shown in FIG. 15, the area generating section 35
First, in step S501, an initial region center is set for a target partial image. In this example, for the block image of interest, the feature vector of the pixel located at the center of gravity of the block image of interest is calculated, and for the calculated region adjacent to the block image of interest, the center of each calculated region is set to the region center V _m = (L _m , a _m , b _m , X _m , Y _m ) (where 1 ≦ m ≦
(Total number of area centers). Set the zero vector as a new region center V _'m corresponding to each region center V _m at the same time, the total number N _m of rearranged pixels keep the 0.

Next, in step S502, one of the elements of the target partial image is thinned out at a thinning rate R = Ra and one is selected. The selection of the pixel, for example, feature vectors of the pixels of the target partial image _{P n = (L n, a} n, b n, X n, Y n) ( provided that, 1 ≦
n ≦ total number of pixels of the target partial image), the above equation (1)
Is performed by sequentially selecting pixels that satisfy the condition (1) one by one.

Next, in step S503, the feature vector P _n of the selected pixel and each region center V _m = (L _m , a _m , b
_m , X _m , Y _m ), and search for the center of the area having the distance d _min closest to the selected pixel. The distance d is calculated by the aforementioned equation (2). At this time, the comparison may be performed using the value of the square of the distance, without the trouble of calculating the square root.

Next, in step S504, it is determined whether or not the closest distance d _min is smaller than a predetermined threshold value T _diff , and the selected pixel is determined as the closest distance d _diff.
It is determined whether it is appropriate to relocate to the center of the area having _min .

That is, when the distance d _min is smaller than the threshold T _diff , it is determined that the selected pixel can be rearranged at the center of the area having the distance d _min, and the process proceeds to step S505. When the distance d _min is _{equal to} or _larger than the threshold T _diff , Assuming that there is no area center suitable for rearranging the selected pixel,
Proceed to 506.

[0133] At step S505, it calculates a new region center in the case of relocating the selected pixel in the region center V _m with the closest distance d _min. In this example, first, the total number N _m of relocated pixel region center V _m with the closest distance d _min is increased by one. Then, the new region center V ′ _m corresponding to the region center V _m having the closest distance d _min is updated by the feature vector P _n of the selected pixel. This update can be performed using equation (5). V ′ _m = V ′ _m + (P _n −V ′ _m ) / N _m ‥ (5)

In step S506, a new area center corresponding to the area including the selected one pixel is newly generated. Specifically, L of the color vector of one selected pixel is
^* , A ^* , b ^* components and XY of the pixel in the image plane
To generate a 5-dimensional feature vectors consisting of the coordinates, this is a region centered V _m. The new region center V _'m to set the same vector as the region center, the total number N _m of rearranged pixel in this area and 1. Of course, the total number of region centers will increase by one.

After updating the center of the new region in step S505 or newly generating the center of one pixel in step S506, the process proceeds to step S507.
Then, it is determined whether or not all the pixels to be selected have been selected in step S502. If not all the pixels have been selected, the process returns to step S502 to select the next pixel to be selected.
The process from step S503 is repeated for the selected pixel.

If it is determined in step S507 that all the pixels to be selected in step S502 have been selected, the flow advances to the next step S508.

In step S508, the new area center V ′ _m
The set as a new region centered V _m, zero vectors in the new region center V _'m, the total number of relocated pixels N _m is set to 0. Here, if a flag indicating that the region has not converged has not been set before these settings, the distance in the feature space between the new region center V ′ _m and the region center V _m is calculated, and this distance is calculated. Is greater than or equal to a predetermined threshold, a flag is set to indicate that the area has not yet converged.

Next, in step S509, it is determined whether or not the flag has been set, and it is determined whether or not the area has converged. When the flag has been set and the area has not converged, the flow returns to step S502. Then, the above processing is repeated from the selection of the pixel.

In step S509, when it is determined that the area has converged due to the flag not being set, the area generation processing in the area generation unit 35 ends.

The area determination section 45 takes over the area center calculated by the area generation section 35 and the total number of area centers as it is, and as shown in FIG. 16, first, in step S601,
All pixels included in the target partial image are sequentially selected one by one.

Next, in step S602, the center of the area having the closest distance to the selected pixel is searched for, and step S6 is executed.
At 03, the selected pixel is arranged in the center area of the searched area.

Next, in step S604, it is determined whether all the pixels included in the target partial image have been selected. If all the pixels have not been selected yet, step S60 is performed.
Returning to step 1, the next pixel to be selected is selected, and the processing from step S602 is repeated for the selected pixel.

If it is determined in step S604 that all the pixels have been selected, the area determination processing in the area determination section 45 ends.

According to the above-described image region division mode, by dividing the image into blocks, the region division of each block image can be performed at high speed because a small number of pixels are divided into a small number of regions. Since the number and center of the regions of the input image are calculated from the partial images, the calculation of the region center can be performed at high speed. Further, all the pixels included in the input image are calculated only once with respect to the calculated region center. Since the divided regions are generated by the distribution, higher-speed region division can be realized. Also, in the region generation processing, the pixels are rearranged or rearranged by performing only the update of the region center on the assumption that the pixels are rearranged instead of actually rearranging the pixels with respect to the region center. It is possible to save the time and storage area required for deleting a file.

FIG. 17 is a block diagram showing an image area dividing apparatus having a small area integrating section. This figure 1
In the image area dividing device indicated by 7, the area is divided by combining the image blocking with the image thinning, and the minute area is integrated for each block image. The image area dividing apparatus of this embodiment includes an image input unit 10, a target partial image extracting unit 50,
It is composed of an area set storage section 70, a partial image storage section 80, an area generation section 35, an area determination section 45, a small area integration section 110, an end determination section 90, and a statistical information calculation section 100.

Image input section 10, image storage section 15, target partial image extraction section 50, area set storage section 70, partial image storage section 80, area generation section 35, area determination section 45, end determination section 90, statistical information calculation The unit 100 can be configured in exactly the same way as with the same number in FIG.

The micro area integration section 110 is provided with an area determination section 45.
Using the information stored in the area set storage unit 70 as the set of the coordinate values of the pixels belonging to the area m, the area smaller than the threshold value specified in advance, Integrate into the most appropriate areas. The area set corrected by the integration processing is reflected in the area set storage unit 70.

As shown in FIG. 18, first, in step S701, the micro area integration unit 110 labels the area generated from the target partial image, thereby obtaining the pixels belonging to one area that is not adjacent on the image plane. The set is divided so that any set of pixels belonging to one region is always adjacent. FIG. 19 shows an example of a pixel set belonging to one non-adjacent area on the image plane. In the figure, all the shaded areas are the same area, and the broken lines indicate the block images. The number assigned to the area is the area number. In this example, the shadowed area of area number 5 is divided into two areas on the image plane. Therefore, this shaded area is divided into two different areas by the labeling process.

Next, in step S702, a map of the area number of the area to which all the pixels included in the target partial image belong is created. For example, an area number map is created by creating a two-dimensional map having a minimum rectangular size including all the pixels in the target partial image and recording the area numbers of the area to which each pixel belongs on the corresponding two-dimensional map. Can be created. FIG. 20A is a diagram showing this area number map, in which the broken line represents the minimum rectangle including all the pixels in the target partial image. The part surrounded by the solid line is one area, and the value on the two-dimensional map corresponding to this area is the area number described in the area in the figure. However, an area number corresponding to a pixel other than the target partial image existing on the two-dimensional map is set to 0. FIG. 20B shows an example of an actual area number map of a shaded portion in the figure.

Next, in step S703, the area number map is scanned, and a list of areas having a size equal to or larger than the predetermined threshold value adjacent to an area smaller than the predetermined threshold value is created. For example, for all the elements of the area number map, the element of interest in the raster order is compared with the elements adjacent to the right and downward, and only when the elements are different, the area of the area corresponding to the two elements is compared. Checking the size, if the size of one area is smaller than the pre-specified threshold and the size of the other area is larger than the pre-specified threshold, the area smaller than the pre-specified threshold This can be performed by associating a number with an area number larger than a predetermined threshold value. Assuming that the regions of the region numbers 8 and 9 in the region number map shown in FIG. 20 are smaller than a predetermined threshold, the list of the regions adjacent to the minute regions 8 and 9 to be created is shown in FIG. Become like

Next, in step S704, the minute regions are integrated into an appropriate region adjacent to each minute region. This process
Looking at the list of areas adjacent to the small area created in S703, if there is only one adjacent area, the target small area is integrated with the adjacent area. When there are a plurality of areas in this list, the areas are integrated into the area having the smallest distance between the centers of the two areas. Alternatively, there is a method of integrating into a larger area. FIG. 22 shows an example of a result obtained by integrating the area number maps shown in FIG.

According to the above-described region division mode, isolated minute regions smaller than the effective size designated in advance can be integrated, and as a result, the number of regions in the region set obtained as a region division result can be drastically reduced. Can be. In addition, since the integration of the minute areas is performed for each target partial image, the size of the work area required for the integration can be reduced.

Note that the above-described various image area dividing apparatuses can be realized as an image area dividing apparatus that is appropriately combined, and the above-described image area dividing methods can be appropriately combined and executed. For example, in the image region dividing device shown in FIG. 14 or FIG. 17, the region estimating unit described in FIG. 5 is provided, and the target partial image extracted by the target partial image extracting unit 50 is decimated at the first decimation rate. The thinned partial image is divided into regions, and the approximate number and center of the regions of the target partial image extracted by the target partial image extraction unit 50 are estimated, and the number and center of the regions estimated by the region estimation unit are used. The target sub-image extracted by the target sub-image extracting unit is subdivided at a second sub-thinning ratio smaller than the first sub-thinning ratio, and the second sub-sub-image is segmented and extracted by the target sub-image extracting unit. A configuration in which the number and center of the regions of the target partial image are calculated by the region generation unit 35 may be employed.

The image area division of the present invention described above
A program for executing the above-described image area dividing method is recorded on various recording media such as various types of magnetic recording, optical recording, and magneto-optical recording, and the recording program is configured to be executed by the above-described image area dividing apparatus. It is possible.

[0155]

As described above, according to the image area dividing method and the image area dividing apparatus of the present invention, by dividing an input image into block images, each time of area division, a small number of pixels are reduced to a small number of areas. The input image can be divided into regions at high speed as a whole, and no inconsistency between the block images occurs, and unnatural region division does not occur. It is possible to drastically reduce the number of minute areas that have little meaning in the area of the generated area data. In addition, the integration processing of the minute areas is performed by using an adjacent area for each block image as a processing unit. Therefore, the storage area required for processing can be reduced.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a configuration of an area dividing device that divides an input image using image thinning.

FIG. 2 is a diagram illustrating a region dividing method for dividing an input image into regions by using image thinning.

FIG. 3 is a diagram illustrating a configuration of an image area dividing device that divides an input image into areas using image blocking.

FIG. 4 is a diagram illustrating an image area dividing method for dividing an input image into areas using image blocking.

FIG. 5 is a diagram illustrating a configuration of an image area dividing apparatus that divides an input image into areas by combining image blocking and image thinning.

FIG. 6 is a diagram illustrating an example of a region estimation step of the image region dividing device illustrated in FIG. 5;

FIG. 7 is a diagram showing an example of an area generating step of the image area dividing device shown in FIG. 5;

8 is a diagram showing an example of an area determination step of the image area dividing device shown in FIG.

FIG. 9 is a diagram provided to explain an input image and its thinning.

FIG. 10 is a diagram provided for explanation of blocking of an input image.

FIG. 11 is a diagram provided for explanation of area division in the case of blocking.

FIG. 12 is a diagram provided for describing extraction of a target partial image in the case of blocking.

FIG. 13 is a diagram illustrating a configuration of an image region dividing apparatus that divides an input image into regions by a clustering method obtained by improving a K-means algorithm.

FIG. 14 is a diagram illustrating a configuration of an image area dividing apparatus that divides an area by combining image blocking and image thinning.

FIG. 15 is a diagram illustrating an example of an area generating step of the image area dividing apparatus illustrated in FIG. 14;

FIG. 16 is a diagram showing an example of an area determination step of the image area dividing device shown in FIG. 14;

FIG. 17 is a diagram illustrating a configuration of an image area dividing apparatus that divides an area by combining image blocking with image thinning and integrates minute areas for each block image.

18 is a diagram illustrating an example of an area integrating step of the image area dividing apparatus illustrated in FIG. 17;

19 is a diagram illustrating an example of an area generated in a target partial image of the image area dividing device illustrated in FIG. 17;

20 is a diagram showing an area number map in a micro area integration step of the image area dividing device shown in FIG. 17;

21 is a diagram showing an area number list adjacent to a minute area in the minute area integrating step of the image area dividing apparatus shown in FIG. 17;

FIG. 22 is a diagram illustrating an area obtained as a result of integrating small areas in the small area integrating step of the image area dividing apparatus shown in FIG. 17;

[Explanation of symbols]

 Reference Signs List 10 image input unit 15 image storage unit 20 region estimation unit 30, 35 region generation unit 40, 45 region determination unit 50 target partial image extraction unit 60 partial image region division unit 70 region set storage unit 80 partial image storage unit 90 end determination unit 100 Statistical information calculation unit 110 Micro area integration unit

Claims (11)

[Claims] 1. A target block image included in a target block image and a calculated area adjacent to the target block image when a target input image represented by quantized multi-tone input image data is divided into a plurality of block images. Image segmentation for segmenting the input image by repeatedly performing a partial image region segmentation step of segmenting a target partial image composed of pixels to be executed, wherein different block images in the plurality of block images are sequentially and repeatedly used as the target block image. The method, wherein the partial image region dividing step is a region calculating step of dividing the target partial image at a predetermined thinning rate into regions, and calculating the number and center of the regions of the target partial image. An area determination step of allocating all pixels included in the target partial image to the area calculated in the area calculation step; Image Segmentation method characterized by comprising the micro region integrating step of integrating a small area in the neighboring region, the. 2. The image area dividing method according to claim 1, wherein, in the minute area integrating step, an area smaller than a predetermined threshold value is selected as a minute area, and the selected minute area is selected as the selected minute area. And integrating the image into an adjacent area having an area larger than a predetermined threshold value. 3. The image area dividing method according to claim 1, wherein the area calculating step divides the first partial image obtained by thinning the target partial image at a first thinning rate to obtain a target image. A region estimating step of estimating an approximate number and a center of the regions of the partial image, and using the number and the center of the regions estimated in the region estimating step, the target partial image is reduced to a second smaller than the first thinning rate. An area dividing step of dividing the area of the second thinned partial image thinned at the thinning rate to calculate the number and center of the areas of the target partial image. 4. The image area dividing method according to claim 1, 2 or 3, wherein the area calculating step, or the area estimating step,
Alternatively, the area generating step or the partial image area dividing step includes: performing a K-means algorithm on the five-dimensional feature space based on three components of a color vector of a pixel and coordinates on an image plane; Alternatively, an image area dividing method characterized by dividing the thinned partial image into areas. 5. The image area dividing method according to claim 1, 2, 3, or 4, wherein the area determining step includes determining a weighted distance in a five-dimensional feature space based on three components of a color vector of a pixel and coordinates on an image plane. An image area dividing method, wherein pixels are distributed to the smallest area. 6. An image input unit for obtaining quantized multi-tone input image data representing an input image to be divided, wherein the image input unit divides the input image into a plurality of block images.
A target partial image extraction unit for sequentially extracting a target partial image composed of pixels included in a target block image and a pixel included in a calculated area adjacent thereto, sequentially as different block images among the plurality of block images as the target block image A partial image area dividing section for sequentially dividing the target partial images sequentially extracted by the target partial image extracting section into regions, and thinning out the target partial images extracted by the target partial image extracting section at a predetermined thinning rate. An area calculation unit that divides the thinned-out partial image into regions and calculates the number and center of the regions of the target partial image extracted by the target partial image extraction unit; and stores the target portion in the region calculated by the region calculation unit. An area determination unit that distributes all pixels included in the target partial image extracted by the image extraction unit, and a micro area integration that integrates isolated micro areas into adjacent areas And a partial image area dividing section having a section. 7. The image area dividing device according to claim 6, wherein the small area part selects an area smaller than a predetermined threshold value as a small area, and selects the selected small area.
An image area dividing apparatus, which is means for integrating into an adjacent area adjacent to the selected minute area and having an area larger than a predetermined threshold. 8. The image area dividing device according to claim 6, wherein the area calculating section thins out the target partial image extracted by the target partial image extracting section at a first thinning rate. An area estimating unit that divides an image into regions and estimates an approximate number and a center of the regions of the target partial image extracted by the target partial image extracting unit; and a number and a center of the regions estimated by the region estimating unit. The target partial image extracted by the target partial image extracting unit is divided into a second thinned partial image obtained by thinning the target partial image at a second thinning rate smaller than the first thinning rate. A region generation unit that calculates the number and center of regions of the target partial image extracted by the unit. 9. The image area dividing device according to claim 6, wherein the area calculating unit, the area estimating unit, the area generating unit, or the partial image area dividing unit calculates a color vector of a pixel. An image region dividing apparatus, wherein the thinned image, the target partial image, or the thinned partial image is region-divided by a K-means algorithm on a five-dimensional feature space based on coordinates on three components and an image plane. 10. The image area dividing apparatus according to claim 6, wherein the area determining unit determines a weighted distance in a five-dimensional feature space based on three components of a color vector of a pixel and coordinates on an image plane. An image area dividing device, wherein pixels are distributed to the smallest area. 11. A program recording medium on which a program for executing the image area dividing method according to claim 1, 2, 3, 4, or 5 is recorded.