JP2003022441A - Method and program for area extraction, computer- readable recording medium with recorded program for area extraction, and device for area extraction. - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make a processing speed fast and to prevent erroneous detection. SOLUTION: An area extracting method includes a step (S1) for inputting an image, a step (S2) for setting a window area which is at least part of the inputted image, a step for computing image information in the window area, a step (S4) for judging by the computed image information whether the window area is an object to be processed, and a step (S8) for detecting an area which represents a face in the window judged as the object to be processed at the judgment step.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image extracting method, an image extracting program, a computer-readable recording medium recording the image extracting program, and an image extracting apparatus, and more particularly to a region in which a specific object is represented in an image. The present invention relates to a region extraction method for extracting a region, a region extraction program, a computer-readable recording medium recording the region extraction program, and a region extraction device.

[0002]

2. Description of the Related Art Conventionally, there has been developed a technique for detecting a region in which an image of a person is represented in an image having various backgrounds. Japanese Patent Application Laid-Open No. 8-339445 discloses a technique of defining a rectangular area in an image and analyzing whether or not a face is included in the rectangular area. To analyze whether or not a face is included in the rectangular area, pixel values in the rectangular area are vectorized and projected onto the principal component space. Then, the distance to the principal component space of the vector (DFFS: Distance From Feature
Space) and the distance in the principal component space (DIFS: Distance)
In Feature Space) is calculated, and it is determined based on these two processes whether or not the rectangular area includes the area in which the face is represented. By gradually moving the rectangular area in the entire image and analyzing it at each position, the rectangular area at the position where the face is included in the image is extracted.

In addition, Hori "Extraction of Face Area from Image by Statistical Method", "Journal of the Information Society of Japan", 1999, Vol. Four
0, NO. In Fig. 8, after the plane fitting normalization and the smoothing of the luminance histogram for each rectangular area, the pixel values in the rectangular area are vectorized, and FKL (Fisher-Fisher-Fisher- Karhunen-Loeve)
A method is described in which a pseudo face or a true face is discriminated by projecting onto a low-dimensional discriminant space using the method.

[0004]

However, in any of the conventional methods, in order to determine whether or not a face is represented in a rectangular area, the main area is scanned for all rectangular areas that scan the entire image. A process of projecting onto a feature space such as a component space or a discriminant space or a process of smoothing a luminance histogram is performed. For this reason, there has been a problem that a lot of time is required for the process of determining whether or not the face is represented in the rectangular area.

Further, when the rectangular area does not include the area where the face is represented but the area where a part of the background is represented, the luminance histogram is smoothed on the rectangular area. This enhances the image. As a result, there are cases where features similar to those of an image showing human eyes, nose, or mouth appear in the rectangular area, and there is a problem that the area is mistakenly detected as a face area.

The present invention has been made in order to solve the above-mentioned problems, and one of the objects of the present invention is to increase the processing speed and to prevent an erroneous detection. An object is to provide an area extraction program, a computer-readable recording medium recording the area extraction program, and an area extraction device.

[0007]

According to one aspect of the present invention for achieving the above object, an area extracting method is an area extracting method for extracting an area including a predetermined object from an image, A step of accepting input of an image,
Setting at least a partial area of the input image, calculating image information of the set area, and whether to set the set area as a processing target based on the calculated image information. And a step of detecting a predetermined object in the area determined as the processing target by the determination step.

According to the present invention, it is detected whether or not the image is an image including a predetermined object in the area determined to be processed. Therefore, it is possible to reduce the number of areas to be processed for detecting the predetermined object. You can Further, since the area obviously not including the predetermined object is not set as the processing target, the area including the predetermined object is not erroneously detected. As a result, the processing speed is increased,
In addition, it is possible to provide a region extraction method that prevents erroneous detection.

Preferably, the image information is a ratio of the number of pixels having a pixel value within a predetermined range to the number of pixels included in the set area.

Preferably, it is the ratio of pixels having a predetermined hue included in the set area.

Preferably, the image information is an average of luminance values of pixels included in the set area.

[0012] Preferably, the image information is a variance of the luminance values of the pixels included in the set area.

Preferably, the image information is an average and a variance of the luminance values of the pixels included in the set area, and the determining step is set when the calculated average and the variance have a predetermined relationship. The step of determining the processed area as the processing target is included.

Preferably, the step of calculating the image information includes the step of calculating a plane that approximates the brightness value of the set area, and the image information is calculated based on a coefficient that defines the calculated approximate plane. It is a value.

According to another aspect of the present invention, the area extracting program is an area extracting program for causing a computer to execute a process of extracting an area including a predetermined object from an image, and a step of receiving an input of the image. And a step of setting at least a partial area of the input image, a step of calculating image information of the set area, and whether the set area is to be processed based on the calculated image information. The computer is caused to execute a step of determining whether or not there is a step and a step of detecting a predetermined object in the area determined as the processing target by the determination step.

According to the present invention, there is provided a computer-readable recording medium having an area extraction program or an area extraction program, which can be executed by a computer to increase the processing speed and prevent erroneous detection. Can be provided.

According to still another aspect of the present invention, an area extracting device is an area extracting device for extracting an area including a predetermined object from an image, and means for accepting input of the image and the input image. Means for setting at least a part of the area, means for calculating image information of the set area, and means for determining whether or not the set area is to be processed based on the calculated image information. And a means for detecting a predetermined object in the area determined as the processing target by the determination means.

According to the invention, the processing speed is increased,
Further, it is possible to provide a region extraction device capable of preventing erroneous detection.

[0019]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. In the drawings, the same reference numerals indicate the same or corresponding members, and the duplicated description will not be repeated.

FIG. 1 is a block diagram showing a schematic configuration of a region extraction device according to one of the embodiments of the present invention. The area extraction device 100 is composed of a personal computer, a workstation, or the like. Referring to FIG. 1, the area extraction device 100 includes a control unit 101 for controlling the entire area extraction device 100, an input unit 105 for inputting data, and a storage unit 103 for storing data. , An output unit 107 for outputting data, and an external storage device 109 for storing a program to be executed by the control unit 101.

The control unit 101 is a central processing unit (CPU)
And a read-only memory (ROM) for storing a program to be executed by this CPU, and a random access memory (RAM) used as a work area when the program is executed by the CPU.

The input unit 105 includes an image input device for photographing a person's face and outputting a two-dimensional image as digital data. The image input device includes a digital camera capable of directly photographing a person, a digital video camera, and an image scanner for reading a photograph showing a person's face. Further, it may be a reading device for reading the digital data of the image from a recording medium in which the digital data of the image including the area in which the person is represented is stored.

Further, the input section 105 includes a mouse or a keyboard. The user of the area extraction device 100 can input characters, numbers, or instruction commands from a mouse or a keyboard.

The output unit 107 includes a display such as a liquid crystal display device or a cathode ray tube and a printer. The output unit 107 also includes a network interface for connecting to an external computer via a network.
Therefore, it is possible to display certain data on the display,
It can be printed from a printer or sent to an external computer.

The storage unit 103 is a magnetic recording device or a magneto-optical recording device. The storage unit 103 stores a base vector database and an identification database (see FIG. 2) described later. The storage unit 103 also stores necessary information based on an instruction from the control unit 101.

The external storage device 109 reads the program and data recorded in the computer-readable recording medium 111 and sends them to the control unit 101. In addition, the control unit 10
According to the instruction from 1, the necessary programs and data are written in the computer-readable recording medium 111.

As the computer-readable recording medium 111, a tape system such as a magnetic tape or a cassette tape,
Magnetic disks (flexible disks, hard disk drives, etc.) and optical disks (CD-ROM / MO / MD / D)
Disk system such as VD), card system such as IC card (including memory card) and optical card, or mask R
The medium is a medium such as a semiconductor memory such as an OM, EPROM, or flash memory that carries a program in a fixed manner. Further, the recording medium 111 may be a medium that fluidly carries the program so that the program may be downloaded from a network.

The control unit 101 can execute the read program by reading the area extraction program recorded in the recording medium 111 with the external storage device 109.

The program is a concept including not only a program directly executable by the central processing unit, but also a program in a source program format, a compressed program, an encrypted program and the like.

FIG. 2 is a functional block diagram showing an outline of the function of the control unit 101 of the area extraction device 100 in this embodiment. Referring to FIG. 2, the control unit 101 includes a window cutout unit 120 for setting a rectangular region (hereinafter, referred to as “window”) of a part of an image received from the input unit 105, masking for the set window, and the like. The preprocessing unit 121 for performing the preprocessing of the above, the projection unit 122 for vectorizing the pixel values in the preprocessed window and projecting the pixel values into the feature space defined by the base vector database 131, and the projection unit The comparison unit 123 that determines whether or not the face is an area where the face is represented by using the reference value stored in the identification database 132 as the generated parameter, and the comparison unit 1
The post-processing unit 12 that integrates a plurality of windows determined to represent a face by 23 and determines one window
Including 4 and.

The storage unit 103 stores a base vector database 131 and an identification database 132. The base vector database 131 stores a space representing the characteristics of an image showing a human face. The feature space is defined by a set of orthonormalized basis vectors in the principal component analysis space. The feature space is obtained as follows. First, a plurality of images including a region showing a human face are prepared, and pixel values of each image are vectorized. Then, a principal component analysis is performed on a plurality of vectors obtained from a plurality of images,
A set of orthonormalized basis vectors in the principal component analysis space is calculated. The set of calculated basis vectors defines the feature space. In addition to the principal component analysis space, a linear discriminant space or the like can be used as the feature space.

The identification database 132 stores a threshold value serving as a criterion. When a discriminant function is used for determination, information on the discriminant function is stored.

The window cutout unit 120 is provided in the input unit 10.
The position of the window in the image received from 5 is set, and the image included in the window at the set position (hereinafter referred to as "window area") is transmitted to the preprocessing unit. The window has two vertical and horizontal pixels.
It is a rectangular area consisting of 0 pixels.

The window cutout unit 120 scans the image by moving the position of the window image by one pixel in the vertical and horizontal directions. The window area specified by the moved window at each position is transmitted to the preprocessing unit 121. Although the window movement amount is one pixel here, the window movement amount is not limited to this and may be a plurality of pixels.

Further, the window cutout unit 120 enlarges or reduces the image so that the whole face contains the human face. Scan the window for a magnified or reduced image. The size of the window may be changed instead of enlarging or reducing the image.

The preprocessing section 121 is a window cutout section 1
The window area specified by the window set in 20 is masked to remove the influence of the background unrelated to facial features. The mask used for masking is
It is a predetermined shape that imitates the contour of a human face. The masking masks the area at the edge of the window area.

Further, the preprocessing unit 121 makes a rough judgment as to whether or not the image of the unmasked portion in the window area includes the area in which the human face is represented. If it is determined by this determination that the window area does not include an area representing a human face, an instruction for cutting out the next window without transmitting the window area to the projection unit 122. Is transmitted to the window cutout unit 120. On the contrary, when it is not determined that the window area does not include the area representing the human face,
The unmasked area of the window area is transmitted to the projection unit 122.

The projecting unit 122 vectorizes the pixel values of the image of the window area received from the preprocessing unit 121, and projects the pixel values into the feature space defined by the basis vectors stored in the basis vector database 131. Thereby, the parameter in the feature space corresponding to the image in the window area is obtained. The obtained parameter is transmitted to the comparison unit 123.

The comparing section 123 calculates DFFS or DIFS using the parameters received from the projecting section 122 and compares it with the threshold value stored in the identification database 132. If DFFS or DIFS is within the threshold range, it is determined that the window includes an area in which a human face is represented, and the storage section 103 outputs the window area image to the post-processing section 124. The position and size of the window area in the image are stored in. The position of the window area is the center of gravity of the window area. Conversely, DFF
If S or DIFS is not within the threshold range, an instruction to cut out the next window is transmitted to the window cutout unit 120.

Further, instead of calculating DFFS or DIFS, it may be determined whether or not the image in the window area includes the area in which the face is represented, by the discrimination function in the discrimination space. Information on the discriminant function used at this time is stored in advance in the identification database 132.

1 represented in the image by the comparison unit 123
It is determined that the plurality of window areas for one face include the area in which the face is represented. This is one for different parts of the face
This is because it is determined that the window areas included in the copy are areas in which the face is represented. Further, when the image is enlarged or reduced, in the window determined to be the area in which the face is represented, there are windows in which the same face is represented in different sizes.

The post-processing unit 124 determines one detection frame from a plurality of window areas in which the same face is displayed. In the process of determining the detection frame, it is assumed that the window regions with close positions are the regions in which the same face is represented, and the average position is calculated by averaging the positions of the centers of gravity of the window regions with close positions. The average size is obtained by averaging the sizes of a plurality of window areas close to each other. In this way, the detection frame is determined by the calculated average position and average size.

Although the size of the window area is the same for 20 × 20 pixels, the size of the window area is obtained by using the magnification of the image enlarged or reduced. And
The position and size of the window area overlapping the detection frame determined by the determined position and size are deleted from the storage unit 103. The image of the area surrounded by the detection frame defined by the position and the size is output to the output unit 107.

The output unit 107 displays the image received from the input unit 105 on the display, and also displays the detection frame determined by the post-processing unit 124 on the displayed image. As a result, in the image input from the input unit 105, the area in which the human face is represented is specified by the outer frame and displayed.

FIG. 3 is a flowchart showing the flow of the area extraction processing executed by the area extraction device 100 according to this embodiment. Referring to FIG. 3, in the area extraction processing,
First, the image input from the input unit 105 is received (step S1).

In the next step S2, a window is set in the window cutout unit 120 in order to select a part of the input image. The window area specified by the set window is transmitted to the preprocessing unit 121. In the area extraction processing in the present embodiment,
The processes of steps S3 to S10 are performed on the window area specified by the window set in step S2. To set the window across the image, we now scan the entire image. Therefore, the window is set by moving it by one pixel in the vertical direction or the horizontal direction. Then, for the window area specified by each set window, step S
The processing from 3 to step S10 is performed respectively.

In the present embodiment, an example in which the window area is set by moving it by one pixel for one input image is shown. However, in reality, the window area is an area in which a face is represented. In order to include it, it is necessary to sample the input image and reduce it, or to generate some enlarged images by linear interpolation, and perform area extraction processing for each reduced or enlarged image. Become.

In the next step S3, the edge portion of the window area is masked in order to exclude the background area unrelated to the area in which the face is displayed. The shape of the mask used for this masking is a predetermined shape. In particular, it is preferable to use a mask that resembles the contour shape of a human face. Since the mask used here is intended to remove the background area other than the area where the face is represented, it is not necessary to completely mask the background area. The background area may be included in part by masking.

In the next step S4, non-face area discrimination processing is performed using the image of the unmasked portion of the window area. This non-face area discrimination processing will be described in detail later. The non-face area determination processing is processing for determining that the window area does not include an area in which a face is represented. In other words, it is a process of determining a window area in which it is clear that a face is not represented, and a process of performing a rough determination of whether or not a face is represented. The non-face area determination process roughly determines whether the window area is a window area in which no face is displayed (non-face area) or a window area in which a face is displayed (face area).

In step S5, it is determined whether or not the window area is determined to be a non-face area by the non-face area determination processing in step S4. If it is determined to be a non-face area, the process proceeds to step S11, and if not, the process proceeds to step S6.

In step S6, the luminance histogram is flattened for the image of the unmasked portion in the window area. The brightness histogram flattening process is a process of converting pixel values so that the frequencies are the same at all brightness levels of the brightness histogram.

The processing of steps S2 to S6 is executed by the preprocessing unit 121. In the next step S7, the pixel values subjected to the luminance histogram flattening process in step S5 are vectorized and projected onto the feature space (step S7). By this projection, the parameter corresponding to the window area is calculated.

The feature space is a space defined by the basis vectors stored in the basis vector database 131 of the storage unit 103. As the feature space, a principal component analysis space or a linear discriminant space is used. For example, when the principal component analysis space is used as the feature space, a vector of pixel values in the window area is set as an input vector x, and a set of orthonormalized basis vectors in the principal component analysis space is {e ₁ , e ₂ , ..., E _n }, the projection vector y is
It is expressed by equation (1). Here, (,) represents the inner product of the vectors, and x ₀ represents the average vector.

In the next step S8, the DFFS is calculated using the equation (2) in the comparison unit 123 for the projection vector y projected onto the feature space, and the threshold ε
Compared to. If the calculated DFFS is larger than the threshold value ε, it is determined as a non-face area and step S11 is performed.
If it is smaller than the threshold value ε, it is determined to be a face area and the process proceeds to step S10 (step S9).

[0055]

[Equation 1]

Instead of the comparison using the DFFS, or in addition to the DFFS, the distance DIFS in the principal component space
May be used for comparison. Furthermore, using a discriminant space created by using a plurality of images of a region in which a face is represented and images including a pseudoface region in which a face is not represented, the distribution status of the true face class and the pseudo face class Whether or not the face area is included may be determined by a discriminant function calculated in advance based on the above. The discriminant space and discriminant function or threshold value may be stored in the discrimination database 132.

In step S10, the position and size in the image of the window area determined to include the area in which the face is represented are stored in the storage unit 103. Then, the process proceeds to step S11. The size of the window area here is
It is represented by the magnification of the image enlarged or reduced.

In step S11, it is determined whether the window has scanned the entire image. If the entire image has been scanned, the process proceeds to step S12, and if not, the process returns to step S2.

In step S2, the window is moved by one pixel in the vertical direction or the horizontal direction to set a new window. For the window area specified by the set new window, steps S3 to S10 are performed. The process is executed.

When it is determined in step S11 that the scanning of the entire image is completed, the storage unit 103 stores the positions and sizes of the plurality of window areas.
In step S12, the position and size of the window area, which is said to include the area in which the face is represented, stored in the storage unit 103 is read, and the detection frame is determined from the read window area (step S12). To determine the detection frame, specifically, the average value of the position and the size of the window regions close to each other is calculated, and the detection frame determined by the calculated average position and the average size is determined. Then, the position and size of the window area overlapping the determined detection frame are deleted from the storage unit 103. Note that a plurality of detection frames may be determined in the entire image by the detection frame determination processing in step S12.

In step S13, the image obtained by adding the detection frame determined in step S12 to the image input in step S1 is transmitted to the output unit 107. This allows
On the display of the output unit 107, the image input by the input unit 105 is displayed, and the detection frame determined in step S12 is displayed by, for example, a white outline on the displayed image. Therefore, the region extraction device 100
The user of can check the detection frame including the area where the extracted face is displayed on the display.

FIG. 4 is a flow chart showing the flow of the non-face area discrimination processing executed in step S4 of FIG.
Referring to FIG. 4, a plane fitting normalization process is performed on the image of the unmasked portion in the window region in order to reduce the influence of light and shadow due to illumination (step S21). The plane fitting normalization process is a process in which a plane approximate to the brightness value in the window area is obtained by using the least square method, and the pixel value of the obtained approximate plane is subtracted from the pixel value in the window area. By performing this plane fitting normalization process, the illumination gradient is corrected. Note that the process is not limited to the plane fitting normalization process as long as it is a process for reducing the influence of light and shadow due to illumination.

In the plane fitting normalization process, if the coordinates (x, y) in the window area and the luminance value of the pixel corresponding to the coordinates are I (x, y), the coefficient a, which satisfies the equation (3),
The luminance distribution is approximated by the plane defined by b and c. In the equation (3), M indicates an unmasked portion in the window area.

Equation (3) can be solved by singular value decomposition or LU decomposition method. Subtracting the plane calculated based on equation (3) from the unmasked portion (original image) in the window area causes pixels with negative luminance values to appear, so add an appropriate offset value to the pixel value. It is possible to visualize by doing. Here, the pixel value is 0
˜255, and the offset value is 127.

A luminance histogram is generated for the image subjected to the plane fitting and normalizing process in step S21 (step S22). In the brightness histogram, the horizontal axis represents the brightness value and the vertical axis represents the frequency.

Next, using the shape of the generated brightness histogram, it is determined whether or not the unmasked portion in the window area includes the area where the face is represented (step S23). Specifically, the total number of pixels included in the range ([α, β]) in which the brightness level is α or more and β or less is calculated, and whether or not the calculated total number of pixels is within a predetermined range. Is determined (step S23). If it is within the predetermined range, the process proceeds to step S24, and if not, the process proceeds to step S25.

In step S23, the number of pixels whose pixel value is within a predetermined range may be determined with respect to the number of pixels in the unmasked portion in the window area.

When the background is uniform, the brightness value is 1
Since the frequency of the pixels in the vicinity of 27 becomes high, it is determined that when 1/3 or more of the number of pixels in the unmasked portion in the window area is concentrated in the vicinity of 127, it is determined not to be within the predetermined range, and step S25 To proceed to. In this case, α = 117, β = 137, ε ₁ = M / 3. Here, M represents the number of pixels in the unmasked portion in the window area.

In step S23, a predetermined range ([α,
It is also determined whether the number of pixels in β]) exceeds the threshold ε ₂ . If the threshold ε ₂ is exceeded, step S24
Otherwise, to step S25.

In step S24, the window area is determined to be an area in which a face is represented (face candidate area), and step S2
In 5, the window area is determined to be an area in which no face is displayed (non-face area).

In the area discriminating apparatus according to the present embodiment, whether or not the window area includes an area in which a face is represented based on the shape of the brightness histogram is determined by the brightness histogram flattening process or whether the area is a face. It is determined before the determination process (step S8). Therefore, whether or not the area in which the face is represented is included in the next window area without performing complicated calculations in the luminance histogram flattening processing (step S6) and the face determination processing (step S8). It is possible to proceed to the process of whether or not. Therefore, the amount of calculation can be reduced and the processing speed can be increased. Further, if the brightness histogram is flattened in a window area that does not include the area in which the face is displayed, the image may be emphasized and misjudgment may occur. Since the window area that does not include is not subjected to the process of determining whether or not it is a face (step S8), it is possible to prevent erroneous determination.

<Modification of Non-Face Area Discrimination Processing> Next, referring to FIG.
Another non-face area determination process executed in step S4 will be described. FIG. 5 is a flowchart showing the flow of another non-face area determination process executed in step S4 of FIG.

Referring to FIG. 5, in step S31, the number of skin-colored pixels is counted in the image of the unmasked portion in the window area, and the ratio to the number of pixels in the unmasked portion is obtained. . As a method for obtaining the number of pixels of skin color, for example, the image input in step S1 is converted pixel by pixel from the RGB space to the uniform perceptual color space Luv, and the pixels of skin color are extracted from the value of the (u, v) coordinate. Any method can be used. In addition, it is preferable to perform morphological processing in order to remove isolated points and connect the skin color areas.

Next, it is judged whether or not the ratio of the number of pixels having the pixel value of the skin color counted in step S31 is a predetermined value or more (step S32). If the ratio is greater than or equal to the predetermined value, the process proceeds to step S33, and if not, the process proceeds to step S41. This is because when the ratio of the number of pixels of skin color is smaller than a predetermined value, the probability that the region in which the face is expressed is included is low.

Next, the average of the luminance values of the pixels included in the image of the unmasked portion in the window area is calculated (step S33). Then, it is determined whether the average brightness value is within a predetermined range (step S34). If the average brightness value is within the predetermined range, step S35.
Otherwise, to step S41. An image having a low average luminance value is an extremely dark image, and an image having a high average luminance value is an extremely bright image. Since such an image has a low probability of being an area representing a face, it is determined to be a non-face area. The predetermined range is preferably 60 or more and 200 or less ([60,200]).

In the next step S35, the plane fitting normalization process is executed on the image of the unmasked portion of the window area (step S35).

Then, it is judged whether or not the value obtained from the inclination of the approximate plane is within a predetermined range (step S3).
6). If it is determined to be within the predetermined range, step S
If not, the process proceeds to step S41. Here, an example is shown in which when the inclination of the approximate plane is large, it is not within the predetermined range, and when the inclination of the approximate plane is not large, it is within the predetermined range.

Whether or not the inclination of the approximate plane is large is determined by
A value (a ² + determined by using the coefficients a and b of the equation (3)
b ² ) is used. If this value is larger than the predetermined threshold value ε _3, it is determined that the inclination of the approximate plane is large, and if not, it is not determined that the inclination of the approximate plane is large. This is because when the inclination of the approximate plane is large, the probability that it is a region in which a face is represented is low, and such a window region is determined as a non-face region without performing subsequent processing.

Next, the variance of the brightness value is calculated for the image of the unmasked part of the window area which is not determined to be the non-face area in step S36 (step S37). Then, it is determined whether the calculated variance is within a predetermined range (step S38). If it is determined to be within the predetermined range, the process proceeds to step S39,
If not, the process proceeds to step S41. If the variance is not within the predetermined range, it is an image with a large change in luminance value, and such an image is unlikely to be a region in which a face is represented. Therefore, in such an image, the window area to be processed is determined to be a non-face area that does not include an area including a face at this stage. Distributed V
The predetermined range for ar is a variance of 90 or more and 1700
The following range (90 ≦ Var ≦ 1700) is preferable.

In the next step S39, it is determined whether or not the average and the variance of the brightness values have a predetermined relationship. If it is determined that there is a predetermined relationship, the process proceeds to step S40,
If not, the process proceeds to step S41. This is because when the average and the variance are not in a predetermined relationship, it is highly likely that the window region does not include the region in which the face is represented, and such a window region is determined as the non-face region.
The predetermined relationship between the average and the variance is Ave for the average and V for the variance.
It is represented by equation (4) as ar.

As a result, without performing the subsequent processing,
The window area can be excluded from the image to be processed, and the processing speed can be increased.

In step S40, the window area is determined to be a face candidate area including the area in which the face is expressed, and the process proceeds to step S5. On the other hand, in step S41, the window area is determined to be a non-face area that does not include the area in which the face is represented, and the process proceeds to step S5.

[0083]

[Equation 2]

FIG. 6 is a diagram showing an example of the relationship between the average and variance of the pixel values of the window area determined as the face area. Referring to FIG. 6, circles and crosses indicate the pixel values of the window area determined as the face area as a result of the area extraction processing without performing the non-face area determination processing in step S4 of FIG. The mean and variance of are shown. The circles indicate the variance and average of the window area in which the face is actually displayed. The crosses indicate the variance and average of the window areas where no face is actually represented.

The circles are almost included in the region where the dispersion Var is 90 or more and 1700 or less ([90,1700]). Also, the circles are almost included in the region where the average Ave is 200 or less. In addition, ○ indicates the average Av
Most of e and the dispersion Var are included in the region of the range determined by the relationship shown in the equation (4). The x mark indicates that the dispersion Var is 9
It is mostly included outside the range of 0 to 1700 ([90,1700]). Also, the average Ave is 20
It is mostly contained outside the region of 0 or less. Further, the average Ave and the variance Var are almost included outside the range of the range defined by the relationship shown in the equation (4).

As described above, in the area extraction processing of this embodiment, the window area is determined to be the non-face area based on the shape of the brightness histogram. Further, when the average luminance value and the variance of the window area are not within the predetermined range, the window area is determined to be a non-face area. Furthermore, when the average and the variance do not have a predetermined relationship, it is determined as a non-face area. Furthermore, when the coefficient that determines the approximate plane obtained by plane fitting normalization is extremely large, it is determined as a non-face area. These determinations are performed before performing the brightness histogram flattening process (step S6), the feature space projection process (S7), and the face determination process (S8) shown in FIG. This is a simpler process than these processes. For this reason, it is determined whether the window area is a non-face area or a face candidate area before performing a process that requires a large amount of calculation. You don't have to. As a result, the processing speed can be increased.

Further, by performing the brightness histogram flattening process (S6) on the area where the face is not represented, it is determined whether the image is emphasized and the face is a face (S6).
8) There is a possibility of making an erroneous determination. Before performing the brightness histogram flattening process (S6), it is determined that a window region that does not include a region in which a face is clearly displayed is a non-face region, and the processes after step S6 are not performed. Misjudgment can be prevented.

Although it is determined in step S37 and step S38 in FIG. 5 whether or not the area is a non-face area by using variance, step S22 and step S in FIG.
It can be replaced with a process of determining whether or not it is a non-face area by using the shape of the luminance histogram shown in FIG.

Further, although the area extracting apparatus has been described in the present embodiment, the area discriminating method for executing the processing shown in FIG. 3 and FIG. 4 or FIG.
It goes without saying that the invention can be understood as an area determination program for causing a computer to execute the processing of FIG. 4 or 5.

The embodiments disclosed this time are to be considered as illustrative in all points and not restrictive. The scope of the present invention is shown not by the above description but by the claims, and is intended to include meanings equivalent to the claims and all modifications within the scope.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a region extraction device according to an embodiment of the present invention.

FIG. 2 is a functional block diagram showing an outline of functions of a control unit of the area extraction device according to the present embodiment.

FIG. 3 is a flowchart showing a flow of region extraction processing executed by the region extraction device according to the present embodiment.

FIG. 4 is a flowchart showing a flow of non-face area determination processing executed in step S4 of FIG.

FIG. 5 is a flowchart showing the flow of another non-face area determination process executed in step S4 of FIG.

FIG. 6 is a diagram showing an example of a relationship between averages and variances of pixel values in a window area determined as a face area.

[Explanation of symbols]

100 area extraction device, 101 control unit, 103 storage unit, 105 input unit, 107 output unit, 109 external storage device, 111 recording medium, 120 window cutout unit, 121 preprocessing unit, 122 projection unit, 123 comparison unit, 124 after Processing unit, 131 basis vector database, 132 identification database.

Claims (10)

[Claims] 1. A region extraction method for extracting a region including a predetermined object from an image, the step of receiving an input of an image, the step of setting at least a partial region of the input image, A step of calculating the image information of the area, a step of determining whether or not the set area is a processing target based on the calculated image information, and a step of determining the processing target by the determination step. And a step of detecting the predetermined object in the area. 2. The area extraction method according to claim 1, wherein the image information is a ratio of the number of pixels having a pixel value within a predetermined range to the number of pixels included in the set area. 3. The area extraction method according to claim 1, wherein the image information is a ratio of pixels having a predetermined hue included in the set area. 4. The area extraction method according to claim 1, wherein the image information is an average of luminance values of pixels included in the set area. 5. The area extraction method according to claim 1, wherein the image information is a variance of luminance values of pixels included in the set area. 6. The image information is an average and a variance of luminance values of pixels included in the set area, and the determining step includes a case where the calculated average and the variance have a predetermined relationship. The method according to claim 1, further comprising: determining the set area as a processing target. 7. The step of calculating the image information includes the step of calculating a plane that approximates the brightness value of the set region, and the image information is based on a coefficient that defines the calculated approximate plane. The area extraction method according to claim 1, wherein the area extraction method is a calculated value. 8. A region extraction program for causing a computer to execute a process of extracting a region including a predetermined object from an image, the step of receiving an input of the image, and the region of at least a part of the input image. Setting a step, calculating image information of the set area, determining whether to set the set area on the basis of the calculated image information, An area extraction program for causing a computer to execute the step of detecting the predetermined object in the area determined to be processed by the determination step. 9. A computer-readable recording medium in which the area extraction program according to claim 8 is recorded. 10. An area extraction device for extracting an area including a predetermined object from an image, comprising means for receiving an input of the image, means for setting at least a partial area of the input image, and setting. Means for calculating image information of the area, a means for judging whether or not the set area is to be processed based on the calculated image information, and a judgment means for judging the processing object A region extraction device, comprising means for detecting the predetermined object in the generated region.