JP2000306104A - Method and device for picture area division - Google Patents

Abstract

PROBLEM TO BE SOLVED: To divide digital picture data into plural areas without requiring initial setting of a division number so that outlines of areas may coincide with edges of the picture. SOLUTION: This picture area division method is provided with an area division fundamental processing means 101 which divides the area of digital picture data based on the luminance level and pixel positions, an area division discrimination means 102 which discriminates whether each divided area should be divided furthermore or not, and a divided area integrating means 103 which integrates adjacent areas into one area in accordance with the color difference between adjacent areas and the edge strength of the area boundary, and the area division fundamental processing means 101 is provided with a binarization and division means 111, which binarizes picture data to divide it into two areas, and a non-linked area division means 112 which discriminates whether each divided area divided by this means 111 is linked area or not and divides it into non-linked area units.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image area dividing apparatus, and more particularly to image recognition, image understanding, image compression, voice recognition, and automatic index assignment to images.

[0002]

2. Description of the Related Art In recent years, with the development of digital technology, a larger amount of digital image data has been generated, transmitted, and stored as compared with the related art. Therefore, a technique for efficiently using image data is desired. For example, in the process of image recognition and image understanding, it is necessary to extract a region to be recognized, and at this time, a region division technique is required. In the image compression processing, not only the compression rate is improved but also editing and synthesis are easily performed by a compression method (object-based encoding) that divides an image into small regions and performs encoding suitable for each region. It is considered possible. Furthermore, a technique for extracting an appropriate area is required when measuring the movement of the lips and face to improve the voice recognition rate or when automatically adding an index to an image.

[0003] The image area dividing method that has been proposed so far can be roughly classified into a method of integrating pixels based on the similarity of luminance between adjacent pixels as in an area growing method (hereinafter referred to as a pixel integration method). And a method of performing clustering based on the feature amount of pixels as in the K-means algorithm (hereinafter, clustering method). In the pixel integration method, since the result greatly differs depending on how to determine a threshold for determining whether or not the pixels are similar, it is difficult to set an optimal threshold.
Further, the amount of calculation becomes enormous as compared with the clustering method.
For this reason, the clustering method has become the mainstream as a recent image division method. On the other hand, in the K-means algorithm, which is a typical method of the clustering method, it is difficult to set an optimal initial number of clusters because the final number of divided areas increases or decreases according to the increase or decrease of the number of clusters initially given. Furthermore, there is a problem that a boundary line other than the outline of the object is generated due to the influence of the shape of the divided region initially given, a problem that a small granular region (noise) is generated around a large region, and a granular region that is not connected to a large region. There is also the problem that small areas are in the same cluster.

In order to overcome such problems, for example, (1) the method described in JP-A-8-263647 and (2) NHK Giken R & D (No. 53 1998/11 p.42-5)
The method described in 5) is known. In the method of (1), three components of the L * a * b * color space and x
The K-means algorithm is applied in a total of six dimensions of two components of the y coordinate and the direction component d of the pixel (horizontal component d1 and vertical component d2). Since the directional component represents the characteristic amount of the texture (pattern) around the pixel, by adding the directional component of the pixel, texture information that cannot be obtained only from the color and position information of the pixel can be reflected in the clustering. This is a feature of this method. That is, the method (1) has an advantage that pixels having similar textures are likely to be in the same cluster. In addition, processing for integrating the granular noise into a large area is also performed. In the method (2), the K-average algorithm is applied to three components of the pixel level RGB and two components of the pixel position xy as the feature amount of the pixel.
The feature of this method is that a method for dealing with the problem of the K-means algorithm is added. That is, a granular region generated around a large region is removed using a majority filter, small regions of 1000 pixels or less are integrated into an adjacent large region, and a boundary other than the outline of the object is obtained using a plurality of initial cluster shapes. Lines are identified and removed. (1),
Each of the methods (2) is improved based on the K-means algorithm.

With respect to the above-mentioned conventional method, a method of calculating the K-means algorithm will be described. FIG. 10 is a flowchart illustrating the calculation processing operation of the K-means algorithm used in the conventional method. In this figure, as an initial state, for example, as shown in FIG. 11, an image is divided into 5 × 4 = 20 (areas) and rectangular clusters are set (step 1). Next, the average value of the feature amount is calculated for each cluster (step 2). If it is not the initial state, the average value of the cluster before the reconfiguration and the average value of the cluster after the reconfiguration are compared to determine whether or not the average has changed. When there is a change from the initial state, the process proceeds to step 4. If there is no change, the process ends (step 3). Next, the distance between the feature value of a certain pixel and the average value of the feature values of each cluster is calculated, and the cluster having the closest distance is checked (step 4). Integrate with the cluster and reconfigure the cluster (step 5). It is determined whether or not the processing of steps 4 and 5 has been performed on the pixels of the entire screen. If there is an unprocessed pixel, the processing returns to the processing of step 4 and continues. If there is no unprocessed pixel, the process returns to step 2 (step 6). By performing the above processing, it is possible to perform clustering according to the feature amount of the pixel.

[0006] In this manner, clustering is performed using the feature amounts extracted from the image, and the image can be divided into small areas.

[0007]

However, the method (1) has a problem that the number of divisions increases or decreases under the influence of the initial number of clusters. In the method (2), the area of the fine pattern (texture) is integrated into a large area, and the area of the fine pattern disappears. As a result, there is a problem that a difference occurs between the edge of the image and the divided area. Either is a problem caused by K-means clustering or a new problem caused by dealing with the problem.

The present invention has been made in view of such a conventional problem. In an image area dividing apparatus that divides digital image data into a plurality of areas, it is not necessary to initially set the number of divisions. It is intended to divide the image so that the contour and the edge of the image match.

[0009]

According to the present invention, there is provided an image area dividing apparatus for dividing digital image data into a plurality of areas. Area division basic processing means for dividing the area, area division determination means for determining whether or not each area divided by the area division basic processing means should be further divided, position, color, and area boundary of the divided area Divided region integrating means for combining a plurality of divided regions into one according to the edge strength of the divided region.

The image region dividing device is provided for each region divided image, and the region dividing basic processing means binarizes the digital image data and divides the digital image data into two regions; Non-connected area separating means for determining whether or not each of the divided areas divided by the means is a connected area and dividing the area into unconnected area units.

According to the present invention, in an image area dividing apparatus for dividing digital image data into a plurality of areas, there is no need to initially set the number of divisions, and the image is divided so that the outline of the area coincides with the edge of the image. be able to.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The invention according to claim 1 of the present invention is directed to a method of dividing an image area into a plurality of areas, the first area being used to divide the area of the digital image data. A dividing step, an area division determining step of determining whether or not to perform further dividing processing on each of the divided areas divided in the primary area dividing step, and further performing the dividing processing in the area dividing determining step. A secondary area division step of dividing the image data area of the determined division area, and further performing image division while limiting the area to be divided based on the determination result in the area division determination step. Thus, it is possible to divide a region at high speed only by setting a condition for terminating the region division processing without setting an initial number of divisions. It has the effect of kill.

According to a second aspect of the present invention, in the image area dividing method according to the first aspect, when there are a plurality of image division areas to be subjected to further division processing in the second and subsequent image divisions, The division condition is determined for each image division area, and the area of the digital image data is divided, and the condition for terminating the area division processing can be individually set for each divided image unit. Has an action.

According to a third aspect of the present invention, there is provided an image area dividing apparatus for dividing digital image data into a plurality of areas, the area dividing basic processing means for dividing an area of digital image data, and the area dividing basic processing means. And a region division determining unit that determines whether to perform a further division process on each of the regions divided by the processing unit, and terminates the region division process without setting an initial number of divisions. This has the effect that the area can be divided only by setting the conditions.

According to a fourth aspect of the present invention, in the image area dividing apparatus according to the third aspect, when there are a plurality of image division areas to be further divided, the area division basic processing means is used for each image division area. It has an area division determining unit, and has an effect that a condition for ending the area division processing can be set for each divided image.

According to a fifth aspect of the present invention, in the image area dividing apparatus according to the third or fourth aspect, the area dividing basic processing means performs the area dividing based on the luminance level and the position of the pixel. And has the effect that the edge and the region boundary coincide.

According to a sixth aspect of the present invention, in the image area dividing device according to any one of the third to fifth aspects, the area dividing basic processing means converts the digital image data into two.
A binarizing division unit that divides the image into two regions by binarizing, and determines whether each of the divided regions divided by the binarizing division unit is a connected region, and divides the region into unconnected region units It is characterized by comprising a non-connected area separating means, and has an effect that an image to be divided is optimally divided into two, and a non-connected area can be treated as another area.

According to a seventh aspect of the present invention, in the image area dividing apparatus according to any one of the third to sixth aspects, the area division judging means comprises a color distribution or a color variance or an area size. Alternatively, it is characterized in that it is determined whether or not to perform the division processing further based on the edge strength of the region boundary, and has an effect that an end condition of the division processing can be set.

According to an eighth aspect of the present invention, there is provided an image area dividing apparatus according to any one of the third to fifth aspects, wherein the digital image data is divided into a plurality of areas. It is characterized in that it is equipped with divided area integration means that combines multiple divided areas into one according to the position, color, and edge strength of the area boundary, and integrates excessively divided areas into an appropriate number of areas. Has the effect of being able to.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. (Embodiment 1) FIG. 1 is a system configuration diagram of an image area dividing apparatus according to Embodiment 1 of the present invention. The image region dividing apparatus according to the present embodiment includes a region dividing basic processing unit 101 that divides a region of digital image data based on a luminance level and a position of a pixel, and whether to further perform a dividing process on each divided region. Division determination means 10 for determining
2 and a divided area integrating means 103 for integrating adjacent areas into one according to the color difference between adjacent areas and the edge strength of the area boundary. Further, the region division basic processing unit 101 binarizes the digital image data and divides the image data into two regions, and determines whether or not each of the divided regions divided by the binarization division unit is connected to each other. And a non-connected area dividing unit 112 for dividing the area into unconnected area units.

The operation of executing the image area dividing method of the present invention by the image area dividing apparatus having the above configuration will be described. FIG. 2 is a flowchart illustrating the basic operation of image area division according to the present invention. As shown in FIG. 2, the digital image is first converted by the binarizing division unit 111
The digital image is binarized using a luminance value, which is one of the features of the digital image, to divide the digital image into two regions. Next, the two divided areas are sequentially connected to the non-connected area dividing means 112.
And the region that is not connected is further divided as another region. By these processes, the primary region division basic process is executed (Step 201).

Next, a plurality of areas (2-1, 2-2,..., 2-) divided in the above-described primary area division basic processing.
N) are sequentially input to the area division determination means 102, and it is determined whether or not the image should be further divided using the color feature amount and the size of the area (steps 211, 212,..., 21).
N).

The areas determined to be further divided in the above-described area division determination processing are sequentially input to the binarization division means 111, and are subjected to the same secondary processing as in the above-described primary area division basic processing (step 201). The area division basic processing is executed (steps 221,... 22N). At this time, in the secondary area division basic processing (step 221) and the secondary area division basic processing (step 22N), it is determined that the area is divided in the area division determination processing (steps 211 and 21N). A division condition is determined for each image division area, and the area of the digital image data is divided. Further, a plurality of areas (3-1, 3-2,..., 3-N, or 4-1 and 4-N) divided by the above-described secondary area division basic processing (steps 221,... 22N).
2,..., 4-N) are sequentially input to the area division determination unit 102, and it is determined whether or not the image should be further divided using the color feature amount and the area size (steps 231 and 23).
2, 23N; 241, 242, 24N).

On the other hand, the area determined not to be divided is input to the divided area integrating means 103 to perform divided area integration processing (step 202), and thereafter stored in a predetermined storage medium. Further, the region division determining unit 102 determines that there is no more region to be divided by the divided region integrating unit 10.
Notify 3. Upon receiving this notification, the divided area integrating means 10
Step 3 integrates the divided areas that match the integration condition, and repeats the processing until there is no more divided area that matches the integration condition. The integrated divided area is output as a division result.

Hereinafter, the binarization division processing, the non-connected area division processing, the area division determination processing, and the divided area integration processing will be specifically described.

FIG. 3 shows a processing flow of the binarization division processing. This processing method is based on the method described in the document “Transactions of the Institute of Electronics, Information and Communication Engineers '80 / 4 Vol.J63-D No.4 p.349-356” applied to color images. First, a histogram of L gradations (for example, 256 gradations) is created using a luminance value, which is one of the feature values of a digital image (step 501). An initial value K = 1 is set using the level K as a threshold (step 502). If the value of the level K is larger than L, the process is terminated, and the level Kmax is set as a threshold for binarization (step 503). If the value of level K is L or less,
Using the equation (1), the 0th moment and the 1st moment up to the K level are calculated (step 504).

(Equation 1) ... (1)

In equation (1), if n _i is the number of pixels of level L and N is the total number of pixels, the normalized histogram p _i can be regarded as a probability distribution of luminance.

Here, assuming that the level K is a threshold value, S = [1, 2,..., K] S = [K + 1, K +
2,..., L] The pixels belonging to the level are classified into _two classes C ₁ and C ₂ . At this time, the 0th and 1st moments of the luminance distribution up to the k level are calculated using the following equations (2) and (3).

(Equation 2) .... (2)

(Equation 3) ... (3)

Next, the occurrence probabilities of the _two classes C ₁ and C ₂ are calculated using the following equation (4) (step 50).
5). Further, the average level of each class is calculated using equation (5) (step 506). The inter-class variance is calculated using the occurrence probability and the average level of each class by using Expression (6) (Step 507).

(Equation 4) ... (4)

(Equation 5) ... (5)

(Equation 6) ... (6)

Thereafter, the maximum value of the interclass variance (σ _max )
² is detected (step 508), and the level K at which the inter-class variance is maximized is set to Kmax (step 509). Thereafter, the value of the level K is increased by one and the process returns to step 503 (step 510).

FIG. 4 is a flowchart for explaining the entire processing operation of the non-connected area dividing process, and FIG. 5 is a flowchart for explaining the processing operation for dividing the parent list. First, in the flowchart of FIG. 4, a list of classes 1 and 2 is created using the pixel position information, which is one of the feature amounts of the digital image, and the Kmax output in the binarization division process (step 601). . Next, the class 1 list is divided as a parent list (step 602), and then the class 2 list is divided as a parent list (step 60).
3).

Next, in the flowchart of FIG. 5, the head pixel of the parent list is moved to the tmp1 list (step 611). Next, all pixels adjacent to the pixels in the tmp1 list are found from the parent list and moved to the tmp2 list (step 612). It is determined whether or not the tmp2 list is empty (step 613). If not, all the pixels in the tmp2 list are moved to the tmp1 list. At this time, t
Among the pixels in the mp1 list, a pixel for which an adjacent pixel has been found in the parent list is distinguished from a pixel for which an adjacent pixel has not been found (step 614). If the tmp2 list is empty, tmp1
All pixels in the list are moved to the divided area pixel list i (step 616). The value of the divided area pixel list number i is increased by one (step 617). After step 614 or 617, it is checked whether the parent list is empty (step 615). If the parent list is not empty, the process returns to step 611. If the parent list is empty, the process ends.

FIG. 6 shows a processing flow of the area division determination processing. Calculate the color variance σ ² for each divided region (step 7)
01). If the color variance σ ² is smaller than the threshold α or the divided area size (the number of pixels included in the divided area pixel list) is smaller than the threshold β, it is determined that the divided area is not to be further divided. Conversely, when the color variance σ ² is equal to or larger than the threshold α and the divided area size (the number of pixels included in the divided area pixel list) is equal to or larger than the threshold β, it is determined that the image is further divided (step 701).

FIGS. 7 and 8 are flow charts for explaining the divided area integration processing operation. In the flowchart of FIG. 7, first, a set of all divided area pixel lists (divided area list) is created (step 801). The divided area list is rearranged in ascending order of the divided area size (step 802), and the head list of the rearranged divided area list (the divided area pixel list having the smallest area size)
Is moved to the tmp1 list (step 803). Next, a region adjacent to the divided region of the tmp1 list (the divided region pixel list means one divided region) is found from the divided region list and moved to the tmp2 list (step 8).
04). One divided area pixel list is selected from the tmp2 list, and this is set as the divided area S (step 805).
The average value V _ave of the edge strength at the boundary between the divided area of the tmp1 list and the divided area S is calculated (step 80).
6). The average value V _ave of the edge strength is compared with a preset threshold γ (step 807), and if it is larger than the threshold γ, the divided area S is moved to the tmp3 list (step 808). Otherwise, the divided area S is set to t
Move to the mp4 list (step 809). Also, t
It is checked whether or not all the areas in the mp2 list have been selected (step 810). If there is any area that has not been selected, the process returns to step 805.

In the flowchart of FIG. 8, it is checked whether or not all the areas in the tmp2 list have been selected in step 810. As a result, if all the areas in the tmp2 list have been selected, it is determined whether the tmp4 list is empty. Check (step 811). If there is a divided area in the tmp4 list, it is integrated with the area in the tmp1 list (step 812). If the tmp4 list is empty, tmp1
It is determined that there is no need to further integrate the divided areas of the list, and the list is moved to the integrated area pixel list i (step 81).
4). One is added to the number of the integrated area pixel list (step 815). If there is a divided area in the unintegrated tmp4 list, the list is returned to the divided area list (step 816).
It is checked whether the divided area list is empty (step 817).
If not empty, or return to step 802 as a process following step 813. If the split area list is empty,
The integration processing ends.

To clarify the distinction between the divided area pixel list and the divided area list used in the above description, FIG.
This will be described with reference to FIG. The divided area pixel list represents one of the areas of the divided image, and the feature amount of the pixels included in the divided area is stored in pixel units. On the other hand, the divided area list is a list storing divided areas constituting the entire image, and is stored in divided area pixel list units.

[0037]

As described above, the image area dividing apparatus according to the present invention judges whether or not to divide based on the preset color variance and the area size, so that it is not necessary to initially set the number of divisions. Since the area is divided based on the luminance level and the position of the pixel, and further integrated by the edge strength at the boundary between the adjacent divided areas, the effect of dividing the area so that the contour of the area coincides with the edge of the image can be obtained. is there.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a system configuration of an image area dividing device according to an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a basic operation of image area division in the embodiment.

FIG. 3 is a flowchart showing a binarization division process in the embodiment.

FIG. 4 is a flowchart for explaining an entire non-connected area dividing process according to the embodiment;

FIG. 5 is a flowchart showing a process of dividing a parent list in the non-connected region dividing process according to the embodiment.

FIG. 6 is a flowchart for explaining an area division determination process in the embodiment.

FIG. 7 is a flowchart for explaining divided region integration processing in the embodiment.

FIG. 8 is an explanatory diagram of a divided region pixel list and a divided region list in the embodiment.

FIG. 9 is a flowchart for explaining divided region integration processing in the embodiment.

FIG. 10 is a flowchart showing a method of calculating a K-means algorithm in a conventional image region dividing apparatus.

FIG. 11 is a diagram showing an example of area division of an image set as an initial state in the conventional example.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 101 Area division basic processing means 102 Area division determination means 103 Division area integration means 111 Binarization division means 112 Non-connected area division means

Claims (8)

[Claims] When dividing digital image data into a plurality of areas, a first area dividing step of dividing the area of the digital image data; and a plurality of divided areas divided in the first area dividing step. A region division determining step of determining whether or not to perform a further division process; and a secondary region division step of dividing an image data region of the divided region determined to further perform the division process in the region division determination step. An image area dividing method, further comprising: performing image division while restricting an area to be divided based on a determination result in the area division determining step. 2. In the second and subsequent image divisions, when there are a plurality of image division areas to be subjected to division processing, a division condition is determined for each image division area unit to divide the digital image data area. 2. The image area dividing method according to claim 1, wherein: 3. An image area dividing apparatus for dividing digital image data into a plurality of areas, wherein the area dividing basic processing means for dividing the area of the digital image data, and each of the areas divided by the area dividing basic processing means. A region division determining means for determining whether or not to perform further division processing, and for a division region determined to perform further division processing, an image division region is further divided. Splitting device. 4. In an image area dividing apparatus for dividing digital image data into a plurality of areas, when there are a plurality of image divided areas, said area dividing basic processing means and area dividing judging means for each image divided area. The image area dividing apparatus according to claim 3, comprising: 5. The image area according to claim 3, wherein the area division basic processing means for dividing the area of the digital image data performs the area division based on the luminance level and the position of the pixel. Splitting device. 6. A region division basic processing unit binarizes digital image data and divides the image data into two regions, and connects the respective divided regions divided by the binarization division unit. 6. The image area dividing apparatus according to claim 1, further comprising a non-connected area separating unit that determines whether the area is an area and divides the area into unconnected area units. 7. The method according to claim 3, wherein the area division determining means determines whether or not to perform further division processing based on color distribution, color variance, area size, or edge strength of an area boundary. The image region dividing device according to any one of the above. 8. An image area dividing apparatus for dividing digital image data into a plurality of areas, the position, color,
8. The image area dividing device according to claim 3, further comprising a divided area integrating unit that combines a plurality of divided areas into one according to an edge strength of an area boundary.