JPH0760462B2 - Image area dividing device - Google Patents

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 device, and more particularly to an image area dividing device for extracting a characteristic region from an image in the fields of image recognition, image understanding, image measurement and the like.

[0002]

2. Description of the Related Art In the fields of image recognition, image understanding, image measurement, etc., it is important to extract a characteristic region from an image, and the accuracy of the characteristic extraction affects recognition accuracy and measurement accuracy.

As a conventional method of analyzing the spatial frequency of an image and extracting a region corresponding to the spatial frequency based on the analysis result, for example, Fumiaki Tomit
a, "COMPUTER ANALYSIS OF VI" written by Saburo Tsuji
SUAL TEXTURE "(KLUWERACADE
The method of texture analysis introduced in MIC PUBLISHERS, 1990) is considered. In this document, as a method for extracting a high spatial frequency region of an image, the image is frequency-transformed, the distribution of its frequency components is analyzed, a specific frequency component is extracted and inverse transformation is performed, and then the region is extracted. A method is being considered.

[0004]

The above-mentioned method has a drawback in that the extraction accuracy is low even though rough extraction is possible because the area is extracted from the entire image.
Further, in the above-mentioned literature, there is a method of calculating a co-occurrence probability etc. in pixel units using a value of an image around a pixel to be calculated as a statistical analysis method and extracting a characteristic region related to the spatial frequency of the image. Although it can be extracted with high accuracy, it has a drawback that the processing speed is slow because it requires a huge amount of calculation. In particular, the above-mentioned drawback becomes more remarkable as an image having a very high resolution, such as high-definition, that is, a larger amount of data to be handled.

Therefore, a main object of the present invention is to accurately and accurately realize a region in which a spatial frequency component of an input image is concentrated and a region in which a spatial frequency component is not concentrated while realizing the shortening of the processing process. An object is to provide an image area dividing device that can be extracted.

[0006]

SUMMARY OF THE INVENTION The present invention is an image area dividing device, wherein an image pickup means for picking up an image, a dividing means for dividing the output of the image pickup means into images of a small image area, and an output of the dividing means are provided. Frequency component conversion means for converting each small region into a frequency component, integration means for integrating the output of the frequency component conversion means into an image of the original image size, and extraction of the spread of the spatial frequency component distribution from the output of the integration means. A distribution extraction unit, a region candidate separation unit that separates the output of the distribution extraction unit into a concentrated region candidate in which the spatial frequency component is concentrated and a distributed region candidate in which the spatial frequency component is not concentrated, and to each output of the region candidate separation unit. Accordingly, it is configured to include a region extraction unit that extracts a spatial frequency component concentration region and a spatial frequency component dispersion region.

[0007]

The image area dividing device according to the present invention divides the picked-up image into images of small areas, calculates the frequency component in each divided image area, and targets the small areas of the original image. By calculating the frequency component, integrating each divided small area to the original size, detecting the spread of the spatial frequency distribution in the small area in the integrated image, and extracting the area, high-speed and accurate spatial frequency It is possible to automatically extract the concentrated area and the dispersed area of the component.

[0008]

1 is a schematic block diagram of an embodiment of the present invention. First, the configuration of the present invention will be described with reference to FIG. The image area dividing device includes a video camera 1 as an image pickup means for picking up an image of an object. The video signal picked up by the video camera 1 is given to the A / D converter 2 and converted into a digital signal image S1. The digital signal image S1 converted by the A / D converter 2 is supplied to the color system conversion unit 3 and converted into an image S2 representing the perceptual color system. This image S2 is given to the small area dividing unit 4 and divided into small area images S3. The divided image S3 of each small region is given to the frequency component conversion unit 5 and converted into an image S4 of frequency components. The image S4 converted into the frequency component for each small area is given to the area integration unit 6 and integrated into the image S5 having the size of the original image.

The image S5 having the original image size integrated by the above-mentioned area integration section 6 is given to the distribution extraction section 7, and the spread of the distribution of the frequency component of the frequency component for each small area is extracted. The image S6 representing the spread of the distribution with respect to the extracted frequency is given to the region candidate separation unit 8 and separated into a concentrated region candidate in which the spatial frequency component is concentrated and a dispersed region candidate in which the spatial frequency component candidate is not concentrated. To be done.
The separated image S7 is supplied to the concentrated area extraction unit 9, and the image S9 in which the concentrated area of the spatial frequency component is extracted is output. The image S8 separated by the region candidate separation unit 8 is given to the dispersion region extraction unit 10, and the image S10 in which the dispersion region of the spatial frequency component is extracted is output.

FIG. 2 shows output images S1 and S in FIG. 1 in an embodiment in which a human face is used as an object.
2 is a diagram showing the second area, and FIG.
FIG. 16 is a diagram for explaining the frequency component conversion unit 5 and the region integration unit 6, which is used when a human face is used as an object.
FIG. 6 is a diagram showing output images S3, S4, and S5 of FIG. 1 in one embodiment in which × 16 pixels are divided into small regions, and FIGS. 4 and 5 are for explaining extraction of the spread of distribution with respect to frequency of the frequency component image. FIG. 9 is a diagram showing an example in the case of being divided into small regions of 16 × 16 pixels. FIG. 6 shows 4 × 4 when a human face is used as an object.
FIG. 8 is a diagram showing the output images S6, S7, and S8 of FIG. 1 in one embodiment divided into small regions of pixels, and FIG. 7 is a small region of 4 × 4 pixels when a human face is used as an object. FIG. 2 is a diagram showing output images S9 and S10 of FIG. 1 in an example divided into two.

The specific operation of the embodiment of the present invention will be described below with reference to FIGS. A face image as an object is captured by the video camera 1. The image signal captured by the video camera 1 is converted by the A / D converter 2 into an RGB digital signal S1 of 512 × 512 pixels. As shown in FIG. 2A showing the image of the G component and the coordinate axes of the RGB digital signal S1, the coordinate axes are set as an image obtained by projecting the object on the xy plane. The RGB digital signal S1 is supplied to the color system conversion unit 3, and the CIE1976L ^* a ^* b ^* shown in FIG.
It is converted into an image signal S2 representing the hue of the perceptual color system using the conversion formula of the color system.

In one embodiment in which a person's face is used as an object, it is effective to perform color system conversion and use a hue image that is not easily affected by shading. One embodiment of the present invention includes a color system conversion unit 3. The image S2 converted by the color system conversion unit 3 is given to the small region dividing unit 4, and the image S3 of each small region shown in FIG.
Is divided into FIG. 3A shows a state in which the image S2 is divided into 32 × 32 blocks in a small region of 16 × 16 pixels, where the x-axis and the y-axis represent the coordinate axes of the image and the z-axis represents the hue value. There is. The image S3 in each of the small regions divided by the small region dividing unit 4 is given to the frequency component converting unit 5, and the square of the power spectrum is calculated using the optical Fourier transform shown in FIG. Is converted into an image S4 representing the frequency component in the small region.

FIG. 3B shows an image converted into frequency components in one small area of 16 × 16 pixels divided by the small area dividing unit 4, and Fx and Fy are x in the small area.
The z-axis represents the frequency component strength, and the z-axis represents the frequency axis in the y-axis direction. The image S4 of the frequency component converter 5 converted into the frequency component for each small area is given to the area integration unit 6 and integrated as an image S5 having the original image size shown in FIG. The image S5 having the frequency components in each small area integrated by the area integration unit 6 is given to the distribution extraction unit 7, and the spread of the distribution with respect to the frequency is extracted by performing threshold processing and binarization processing. It

Here, FIGS. 4A, 4B and 5
Extraction of the spread of the distribution of frequency components in a small area will be described with reference to (a) and (b). 4 and 5, Fx and Fy represent frequency axes in the x-axis direction and the y-axis direction in the small region. FIGS. 4A and 5A show 16 × 1 divided by the small area dividing unit 4.
An image obtained by converting frequency components in a small region of 6 pixels is shown, and the x-axis represents frequency component intensity. It is generally known that most of the signal power of the image signal is concentrated on the low frequency side, and FIG.
Since (a) is an example in which a person's face is used as an object, it is an example in which it is concentrated on the low frequency side as a whole, that is, near the center of FIGS. 4 (a) and 5 (a). . 4 (b) and 5 (b) are shown in FIG. 4 (a) and FIG.
An example of the result after performing threshold processing and binarization processing on each image shown in (a) is shown, the z axis represents the presence or absence of a frequency component, zero indicates no frequency component, and zero. Other than the above, the result detected when there is a frequency component is expressed.

FIG. 4A shows that in the divided small area,
For example, FIG. 4 shows an example of image distribution when low frequency components are concentrated as a specific frequency component.
FIG. 4B shows an example of the result after performing thresholding and binarization processing on the image having the distribution of FIG. 4A, and outputs an image concentrated in a place with a small area. To do. Figure 5
FIG. 5A shows an example of image distribution in a case where the image is not concentrated on a certain specific frequency component in the divided small regions, and FIG. 5B shows an image having the distribution shown in FIG. On the other hand, an example of the result after performing threshold processing and binarization processing is shown, and an image spread in a small area is output. As shown in FIG. 4 (b) and FIG. 5 (b),
By performing the threshold processing and the binarization processing, the spread of the distribution of frequency components can be detected.

In the distribution extraction unit 7, the threshold value processing and the binarization processing described in FIGS. 4 and 5 are performed on the image S5 having the frequency components in each small area integrated by the area integration unit 6. By doing so, the spread of the distribution of frequency components in the small area is detected in the entire image. In FIG. 6A showing the output image S6 of FIG. 1 in an example in which the face of a person is used as an object and divided into small regions of 4 × 4 pixels, only low-frequency components such as hair regions are shown in FIG. 6A. Instead, in a region including high frequency components, small regions indicating a state in which the distribution of the frequency components of the distribution extracting unit 7 has spread are gathered, and in a region where low frequency components are concentrated such as a face region, the frequency components are It shows a state of concentration in a small area.

The image S6 showing the spread of the distribution with respect to the frequency extracted by the distribution extracting unit 7 is given to the region candidate separating unit 8 and is a binary value such as 2 × 2 or less isolated pixel removal, bridge connection and majority decision processing. 3 × for digitized image
Spatial image processing is performed for 3 pixels or less.
For example, as an example for a person's face, as shown in FIG.
The concentrated area candidate image S7 in which the spatial frequency components are concentrated as shown in (b) and the dispersed area candidate image S8 in which the spatial frequency components are not concentrated as shown in FIG. 6C can be easily separated.

The images S7 and S8 separated by the area candidate separating unit 8 are concentrated area extracting unit 9 and dispersed area extracting unit 10.
, 2x2 or less isolated pixel removal, bridge connection, and majority image processing for binarized image spatial image processing for 3x3 pixels or less, contour extraction and closed region The spatial frequency component concentrated region and the spatial frequency component dispersed region are extracted by performing image processing related to region extraction such as painting. As an example, the image S9 showing the spatial frequency component concentrated region output shown in FIG. 7A and the image S10 showing the spatial frequency component dispersion region output shown in FIG. 7B are output.

The color system conversion unit 3 for converting to the perceptual color system shown in FIG. 1 uses, as an example, a face image as a target and performs processing that is less susceptible to the shadow of the light source. ing. As the conversion formula of this color system, CIE1
In addition to the 976L ^* a ^* b ^* color system conversion formula, for example, CIE197 described in “New Handbook of Color Science” (The University of Tokyo Press, 1985) edited by the Japan Color Society.
6L ^* u ^* v ^* color system conversion formulas and the like can be used. However, depending on the target of the area division, the color system conversion unit 3
, And the RGB digital signal S1 output by the A / D converter 2 shown in FIG. 1 can be directly used. In addition to the Fourier transform, the conversion formula for the frequency component in the frequency component conversion unit 5 shown in FIG. 1 is described in, for example, “Television Image Information Engineering Handbook” (Ohm Co., 1990) edited by the Television Society. The Walsh-Hadamard conversion, etc., can be used.

As described above, according to this embodiment, FIG.
Is converted into an image having a local frequency component relating to a small region of the image by the small region division unit 4, the frequency component conversion unit 5, and the region integration unit 6, and is converted into the entire image for the conventional extraction of the frequency component. On the other hand, since the conversion is performed on the small area of the image, the disadvantage of low extraction accuracy can be alleviated. Further, the feature is that the converted image of the frequency component is used as it is without performing the inverse conversion after the frequency analysis. In addition, as a statistical analysis method, the value of the image around the pixel to be calculated is used to calculate the co-occurrence probability etc. in pixel units, and in comparison with the method of extracting the characteristic region related to the spatial frequency of the image, in small region units. The conversion amount can be reduced and the processing speed can be increased. In particular,
In one embodiment of the present invention, since the area extraction is performed using the output image converted into the perceptual color system which is not easily influenced by the shadow of the light source by the color system conversion unit 3 shown in FIG. Can be extracted.

[0021]

As described above, according to the present invention, the captured image is divided into images of small areas from the original size, and then the frequency component is calculated for each divided image area to obtain the original image. Find the frequency component for the small area of the image of
Each divided small area is integrated into the original size, the spread of the spatial frequency distribution in the small area is detected for the integrated image, and the area extraction is performed, so that the spatial frequency component of the spatial frequency component is accurately measured at high speed. The concentrated area and the dispersion area of the spatial frequency component can be automatically extracted.

[Brief description of drawings]

FIG. 1 is a schematic block diagram showing an embodiment of the present invention.

FIG. 2 is a diagram showing output images S1 and S2 of FIG. 1 in an example in which a human face is used as an object.

FIG. 3 is a diagram for explaining a small region dividing unit 4, a frequency component converting unit 5 and a region integrating unit 6 in FIG. 1, and is a 16 × 16 pixel small region when a human face is used as an object. Output images S3 and S4 of FIG. 1 when divided into
It is a figure which shows and S5.

FIG. 4 is a diagram illustrating an example of division into small regions of 16 × 16 pixels for explaining extraction of spread of distribution of frequency components with respect to frequency.

FIG. 5 is a diagram showing an example in the case of being divided into small regions of 16 × 16 pixels for explaining extraction of the spread of distribution of frequency components with respect to frequencies.

6 is an output unit S of FIG. 1 in an example in which a person's face is used as an object and is divided into small regions of 4 × 4 pixels.
It is a figure which shows 6, S7, and S8.

7 is an output image S of FIG. 1 in an example divided into small regions of 4 × 4 pixels when a human face is used as an object.
It is a figure which shows 9 and S10.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Video camera 2 A / D converter 3 Color system conversion unit 4 Small region division unit 5 Frequency component conversion unit 6 Region integration unit 7 Distribution extraction unit 8 Region candidate separation unit 9 Concentrated region extraction unit 10 Distributed region extraction unit

Claims (1)

[Claims] 1. An image pickup unit for picking up an image, a dividing unit for dividing an output image output from the image pickup unit into small image regions, and an image for each small image region divided by the dividing unit as a spatial frequency component. A frequency component converting means for converting; an integrating means for integrating an image of each small region representing the frequency component converted by the frequency component converting means into a size of an image captured by the image capturing means; and an output from the integrating means. Distribution extraction means for extracting the spread of the distribution of spatial frequency components from the output image; concentrated area candidates in which the spatial frequency components are concentrated and distributed areas in which the spatial frequency components are not concentrated in the output image output from the distribution extraction means A region candidate separating means for separating into candidates, and a spatial frequency component concentrated region according to each output of the region candidate separating means, With a region extracting means for extracting between frequency component dispersion region, the image area dividing device.