JP2004258752A - Image dividing method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To carry out more appropriate area division of an image, in an image division method and device, for each of object areas on the basis of various feature values extracted from image data such as a photographic image. <P>SOLUTION: A feature value extracting means calculates luminance color information and texture information for each of the pixels of an image, an area dividing mens integrates the pixels wherein the combination of the luminance color information and texture information is equivalent, and divides the image into a plurality of areas I to VI constituted of the integrated pixel group, and an area integrating means integrates the area III under consideration and the area V under consideration among those areas with another adjacent area VI whose luminance color information is equivalent, thereby dividing the image into a plurality of areas 1 to 4. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an image dividing method and apparatus for dividing an image into a plurality of regions based on a plurality of types of feature amounts extracted from image data.
[0002]
[Prior art]
An image captured by a digital camera or the like includes, for example, a person, the sea, the sky, a building, and other backgrounds. If image data such as the person and background is extracted from the image as one object, different image processing can be performed for each object, and the meaning of each object can be determined and presented to the user. it can. For this purpose, it is necessary to divide the object area for each object included in the image.
[0003]
Therefore, the image data constituting one object is extracted from the image data by using the property that the feature quantities such as color information, texture information, and wavelet coefficients are similar, and this feature quantity is extracted. A method of dividing an image into a plurality of object regions based on the above has been proposed. For example, in Patent Document 1, a plurality of types of feature amounts are extracted for each pixel of image data, a plurality of feature vectors having the plurality of types of feature amounts as elements are generated, and the feature vectors are similar to each feature vector. There has been proposed a method of classifying and classifying an image into a plurality of object regions based on the clustered result.
[0004]
[Patent Document 1]
Japanese Patent Application No. 2002-246395 [0005]
[Problems to be solved by the invention]
However, when an image is divided based on a feature vector composed of a plurality of types of feature amounts as described above, the dynamic range differs between luminance information, color information, and texture information. Weighting is performed by generating a feature vector by multiplying a predetermined coefficient. Therefore, the result of region division differs depending on this weighting, so it is necessary to optimize the weighting. However, since the optimum weighting is different for each image object, it is not easy, and the image object is associated with the weight. A database is required, which increases costs.
[0006]
The present invention provides an image segmentation method and apparatus for segmenting an image into a plurality of regions based on a plurality of types of feature amounts extracted from image data. It is an object of the present invention to provide an image dividing method and apparatus capable of dividing each area.
[0007]
[Means for Solving the Problems]
The image dividing method of the present invention obtains a plurality of types of feature amounts for each pixel of an image, integrates pixels having the same combination of the plurality of types of feature amounts, and forms an image by the integrated pixel group. Divided into a plurality of regions, and at least one region of the region is set as a region of interest, and the region of interest is a region adjacent to the region of interest having at least one type of feature amount equivalent among the plurality of types of feature amounts The image is divided into a plurality of regions by integration.
[0008]
Here, examples of the “plural types of feature amounts” include luminance information, color information, and texture information. The texture information is information related to edges in the image, and for example, wavelet transform coefficients, DCT coefficients, primary differential filter calculation processing results, secondary differential filter processing results, and the like can be used.
[0009]
The phrase “a combination of a plurality of types of feature values is equivalent” means that each of the feature values of a plurality of feature values of a pixel belongs to the same classification. Belonging to the same classification means, for example, that the feature value is within a predetermined value range. In addition, when a feature amount within the range of such a predetermined value is replaced with another value, the combination of the values is the same.
[0010]
In the image dividing method, when there are a plurality of other regions, the distance in the feature amount space of any one of the plurality of types of feature amounts with the attention region among the other regions is minimized. Other regions can be integrated with the region of interest to divide the image into a plurality of regions.
[0011]
Here, the “distance of the feature amount in the feature amount space” is, for example, the Euclidean distance between the average value of the feature amount of all the pixels in the region of interest and the average value of the feature amount of all the pixels in the other region. Say.
[0012]
The image dividing apparatus according to the present invention integrates a feature amount extraction unit that obtains a plurality of types of feature amounts for each pixel of an image, and a pixel group in which a combination of a plurality of types of feature amounts is equivalent, and combines the images Region dividing means for dividing the region into a plurality of regions, and at least one of the regions as a region of interest, and the region of interest is equivalent to at least one type of feature amount among a plurality of types of feature amounts And an area integration means for dividing the image into a plurality of areas by integrating with other areas adjacent to the area.
[0013]
In the image dividing apparatus, when there are a plurality of other regions, the region integration unit is configured in a feature amount space of any one feature amount among a plurality of types of feature amounts with the attention region among the plurality of other regions. It is possible to divide the image into a plurality of regions by integrating another region having the smallest distance with the region of interest.
[0014]
【The invention's effect】
According to the image dividing method and apparatus of the present invention, a plurality of types of feature amounts are obtained for each pixel of an image, pixels having the same combination of the plurality of types of feature amounts are integrated, and an image is integrated into the pixel group. Is divided into a plurality of regions, and at least one of the regions is set as a region of interest, and the region of interest is adjacent to the region of interest having at least one type of feature amount equivalent among the plurality of types of feature amounts. Since the image is divided into a plurality of areas by being integrated with the area, it is possible to perform appropriate area division for each object area without the need for a database as described above.
[0015]
In the image dividing method and apparatus, when there are a plurality of other regions, the distance in the feature amount space of any one of the plurality of types of feature amounts with the attention region among the other regions is the minimum. If the image is divided into a plurality of areas by integrating the other areas with the attention area, optimization of the other areas to be integrated with the attention area can be achieved. Region division can be performed.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of an image dividing apparatus that performs an image dividing method of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the above embodiment. The image segmentation apparatus according to the present embodiment extracts luminance color information and texture information for each pixel of the image, and integrates pixels having the same combination of the luminance color information and texture information into a plurality of regions. The image is divided, and at least one of the regions is integrated with an adjacent region having the same luminance color information, and the image is divided into a plurality of regions.
[0017]
As shown in FIG. 1, the image segmentation apparatus according to the present embodiment is a feature amount extraction unit 10 that extracts luminance color information and texture information from image data, and the luminance color information and texture information extracted by the feature amount extraction unit 10. Clustering processing means 12 that performs clustering processing for each, labeling processing means 14 that performs labeling processing for each of the luminance color information and texture information subjected to clustering processing by the clustering processing means 12, and labeling processing performed by the labeling processing means 14. For each of the luminance color information and the texture information, the micro area integration processing means 16 for performing the micro area integration processing, and for each of the luminance color information and the texture information subjected to the micro area integration processing by the micro area integration processing means 16, Luminance color information is obtained by integrating the pixels with the same combination of ring data to divide the image into a plurality of areas, and the area having a predetermined number of pixels or less of the plurality of areas divided by the area dividing means 20 And an area integration means 22 for dividing the image into a plurality of areas by integrating the areas into the area adjacent to the area and outputting the area divided image.
[0018]
As shown in FIG. 1, the feature quantity extraction unit 10 includes an RGB-YCC conversion processing unit 11 that extracts luminance color information from image data and a texture information extraction processing unit 13 that extracts texture information from image data. Yes.
[0019]
The labeling processing unit 14 performs the labeling process by attaching the same label to a region in which pixels having numerical values indicating the same cluster are grouped in the real space. This labeling process is performed because the result of clustering in the clustering processing means 12 is that pixels belonging to the same cluster are grouped in the vector space but may be discretely distributed in the real space. .
[0020]
When the micro area integration processing means 16 divides the area based on the result of the labeling process performed by the labeling processing means 14, a micro area consisting of several pixels to several tens of pixels is actually an unnecessary area, but one area. Therefore, a process for integrating such a minute area into an adjacent area is performed.
[0021]
Next, the operation of one embodiment of the image dividing apparatus of the present invention will be described.
[0022]
First, image data is input to the RGB-YCC conversion processing unit 11 and the texture information extraction processing unit 13 of the feature amount extraction unit 10. The RGB-YCC conversion processing unit 11 generates an RGB component for each pixel data of the image data, performs an RGB-YCC conversion process on the RGB components, and generates a YCC component for each pixel data of the image data. . In this embodiment, the YCC component is generated as the luminance color information as described above, but the RGB component for each pixel data may be used as it is. Further, the luminance color information may be extracted by using HSV conversion, L * a * b * conversion, or the like without being limited to the conversion to the YCC component as described above. Then, the RGB-YCC conversion processing unit 11 outputs the YCC component for each pixel data to the clustering processing unit 12.
[0023]
The clustering processing means 12 uses each pixel data on a vector space (a three-dimensional space when using the YCC component as luminance color information) having each component of the YCC component (Y component, Cb component and Cr component) as elements. A vector whose elements are the components of the YCC component for is mapped. In the mapped vector space, vectors of pixel data having similar YCC components are gathered to form a cluster, and thus each pixel data is collected for each cluster. The pixel data assigned to each cluster is assigned a different numerical value for each cluster, and the result is output to the labeling processing means 14.
[0024]
The labeling processing means 14 maps the clustering processing result in the vector space obtained as described above to the real space, and the same name (for example, for example) the pixel of the region where the pixels to which the same cluster belongs are grouped in the real space. Numerical values 0, 1, 2,. Then, the labeling processing unit 14 outputs the labeling processing result to the minute region integration processing unit 16.
[0025]
The micro area integration processing means 16 extracts the minimum area having the smallest number of pixels from the areas divided based on the input labeling processing result. When the number of pixels in the minimum area is equal to or less than a predetermined threshold (for example, the predetermined threshold is about 1/1000 of the total number of pixels), this area is set as a processing target area, and the processing target area is circumscribed. An adjacent area (see FIG. 2) existing in the rectangular area is extracted. Then, average values of the Y component, the Cb component, and the Cr component of the pixel data in the processing target area are obtained as representative values of the processing target area. Similarly, the representative value is obtained for each adjacent region. From the representative value of the processing target region to the representative value of the adjacent region when the representative value of the processing target region and the representative value of the adjacent region are plotted on the three-dimensional coordinates with the Y component, the Cr component, and the Cb component as axes. Measure the Euclidean distance. Then, an adjacent area having the shortest distance is extracted from the area, the processing target area and the adjacent area are integrated, and the label of the pixel of the integrated processing target area is re-attached to the label of the pixel of the adjacent area. Then, the minimum area with the smallest number of pixels is extracted again, and when the number of pixels in the minimum area is equal to or less than a predetermined threshold value, it is integrated with the adjacent area in the same manner as described above. Then, the above process is repeated until there is no area having the number of pixels equal to or smaller than the predetermined threshold value, and the result is output to the area dividing means 18.
[0026]
On the other hand, the texture information extraction processing unit 13 to which the image data is input detects an edge signal from the image data, and outputs the detection result of the edge signal to the clustering processing unit 12 as texture information of each pixel data. The detection of the edge signal may be performed using, for example, a primary differential filter process or a Laplacian filter process, or a wavelet transform coefficient or a DCT coefficient may be used as texture information. Further, the results of the vertical, horizontal and diagonal first-order differential filter processing may be used as three-dimensional information, or the image data S0 is subjected to multi-resolution processing, and the texture information and the low-resolution image in the high-resolution image. The texture information may be used as two-dimensional information.
[0027]
Based on the input edge signal detection result, the clustering processing unit 12 performs clustering on the portion where the edge signals are gathered and the portion where the edge signals are small, and outputs the result to the labeling processing means 14. Specifically, clustering may be performed using a result of performing a morphology operation process on the edge signal.
[0028]
The labeling processing unit 14 performs labeling processing in the same manner as described above based on the input clustering processing result, and the micro region integration processing unit 16 performs micro region integration processing on the labeling processing result in the same manner as described above. The result is output to the area dividing means 18.
[0029]
As described above, the result obtained by dividing the image data based on the luminance color information and the result obtained by dividing the area based on the texture information are input to the region dividing unit 18. Then, for example, the result of the area division based on the luminance color information is the area A with the label 0, the area B with the label 1, and the label 2 as shown in FIG. The area C is divided into three areas, and the result of the area division based on the texture information is the result of the area a and the label 1 labeled 0 as shown in FIG. Assuming that the area is divided into two areas of the attached area b, the area dividing means 18 applies both the label attached in the luminance color information area division result and the label attached in the texture information division result. Label each pixel again based on it. Specifically, for example, when the label of the luminance color information area division result is 0 and the label of the texture information area division result is 0 for a predetermined pixel, the pixel is labeled 0. For a given pixel, if the label of the luminance color information area division result is 1 and the label of the texture information area division result is 0, the target pixel is assigned an integrated label 1, and so on. The same combination label is attached to the same combination of the label of the luminance color information area division result and the label of the texture information area division result. The integrated label is attached according to the table of FIG. 4, and the image is divided into regions I to VI based on the integrated label. The result of the re-region division as described above is output to the region integration unit 20.
[0030]
The region integration unit 20 performs the same processing as the processing in the minute region integration processing unit 16 described above on the input result of the re-region division, but the threshold value of the number of pixels when determining the processing target region is For example, it may be different from the above threshold value. The area integration processing means 20 performs the area integration process only based on the luminance color information. As a result of the re-region division as shown in FIG. 3C, when it is determined that the region III and the region V among the regions I to VI are the processing target regions, the region III is the region C together with the region IV. Therefore, the distance with the luminance color information as a representative value is 0, and as shown in FIG. 3D, the region III and the region IV are integrated into the region 4, and both the region V and the region II are the region B. Therefore, the distance between the representative values of both is 0, and the region V and the region II are integrated into the region 2 as shown in FIG. In addition, since the area IV is large and the number of pixels is larger than the threshold value, the area IV becomes an independent new area 3 as shown in FIG. By performing the region integration processing as described above, if the region becomes a minute region in the stage of FIG. 3C, the region is restored to the region divided based on the luminance color information, and if the region does not become the minute region Thus, the area divided based on the luminance color information can be further divided. That is, it is possible to generate a region divided image having the advantages of the region divided image based on luminance color information and the region divided image based on texture information.
[0031]
In addition, when a human sees an image from a distance, it is considered that the object region is divided mainly by luminance change and color change. Therefore, the region integration processing is performed based on the luminance color information as described above. By doing so, it is possible to perform region division closer to a human sense.
[0032]
In FIG. 3D, the region integration process is performed based on the YCC component that is the luminance color information. However, the region integration process may be performed using only the luminance information (Y component).
[0033]
In the above embodiment, the clustering process and the labeling process are performed on each of the luminance color information and the texture information, and then the micro area integration process is performed on the result of each labeling process. The region dividing unit 18 may perform the re-region division process based on the results of the respective labeling processes.
[0034]
In the above embodiment, the clustering process is performed on each of the luminance color information and the texture information, and then the labeling process is performed on each clustering process result. However, the labeling process may not be performed. After performing clustering processing for each of the luminance color information and texture information, an integrated label is attached to each pixel using the numerical value of the cluster attached to each pixel according to the result of each clustering processing, and based on the integrated label Re-region division may be performed. In the above case, for example, when the number of clusters in the clustering processing result based on the luminance color information is 16 and the number of clusters in the clustering processing result based on the texture information is 4, 16 × 4 = 64 integrated labels Will be attached.
[0035]
The image dividing method according to the present invention may be provided as a program for causing a computer to execute the image dividing method.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an embodiment of an image dividing apparatus according to the present invention. FIG. 2 is a diagram for explaining a micro area integration process in the image dividing apparatus shown in FIG. FIG. 4 is a diagram for explaining processing in the region dividing unit and region integrating unit of the apparatus. FIG. 4 is a diagram for explaining processing in the region dividing unit of the image dividing apparatus shown in FIG.
DESCRIPTION OF SYMBOLS 10 Feature-value extraction means 11 RGB-YCC conversion process part 12 Clustering process means 13 Texture information extraction process part 14 Labeling process means 16 Minute area | region integration process means 18 Area division means 20 Area integration means

Claims (4)

Obtaining a plurality of types of feature amounts for each pixel of the image, integrating pixels having the same combination of the plurality of types of feature amounts, and dividing the image into a plurality of regions formed by the integrated pixel group; At least one region of the region is set as a region of interest, and the region of interest is integrated with another region adjacent to the region of interest having at least one type of feature amount equivalent among the plurality of types of feature amounts. An image dividing method, wherein the image is divided into a plurality of regions. When there are a plurality of the other areas,
Of the plurality of other regions, the other region having the smallest distance in the feature amount space of any one of the plurality of types of feature amounts with the attention region is integrated with the attention region, and the image is obtained. The image dividing method according to claim 1, wherein the image is divided into a plurality of regions. Feature amount extraction means for obtaining a plurality of types of feature amounts for each pixel of an image, and a plurality of regions formed by integrating the pixels in which the combination of the plurality of types of feature amounts is equivalent to constitute the image An area dividing means for dividing the area into at least one of the areas as an attention area, and the attention area is adjacent to the attention area having at least one type of feature quantity equivalent among the plurality of types of feature quantities. An image dividing apparatus comprising: an area integration unit that integrates an area and divides the image into a plurality of areas. The region integration means is
When there are a plurality of the other areas,
Of the plurality of other regions, the other region having the smallest distance in the feature amount space of any one of the plurality of types of feature amounts with the attention region is integrated with the attention region, and the image is obtained. 4. The image dividing apparatus according to claim 3, wherein the image dividing apparatus is divided into a plurality of regions.

Images