JP2009140423A - Face centerline detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a face centerline detecting device capable of detecting a more accurate centerline. <P>SOLUTION: The face centerline detecting device 1 includes a face image extraction part 11 which receives input of a gray level image and extracts a face image containing a face from the gray level image; a concentration gradient image generation part 12 which substitutes the pixel value of each pixel of the face image extracted by the extraction part 11 by a numerical value according to the direction of concentration gradient (density gradient direction) in the position of the pixel to generate a concentration gradient image; and a centerline detection part 13 which detects the centerline of the face image using the concentration gradient image generated by the generation part 12. By using the concentration gradient direction larger in information quantity than edge, a further accurate centerline can be obtained, compared with the case using the edge. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a face centerline detection device that detects a face centerline in a captured image.

  2. Description of the Related Art Conventionally, a face centerline detection device that detects a centerline that is the center in the left-right direction of a face image that is a face area in a captured image has been provided. In this type of face centerline detection device, for example, the authentication target person registers the face image of the authentication target person extracted from the captured image and the reference face image indicating the face of the registrant registered in advance. In a face authentication device that determines whether or not a person is a person, it is used to match the position of the face image of the person to be authenticated with the position of the reference face image prior to comparison.

Conventionally, the edge of a face image is detected, and the center line of the face image is detected by detecting the symmetry of the edge (see, for example, Patent Document 1).
JP 2004-110543 A

  However, when another object such as a tree branch exists between the light source and the face, an edge due to the shadow of the object may be generated, and the symmetry of the edge is detected as described above. In the detection of the center line by, an error due to an edge caused by the illumination condition is likely to occur like the above-mentioned shadow.

  The present invention has been made in view of the above-described reasons, and an object thereof is to provide a face centerline detection device capable of detecting a more accurate centerline.

  According to the first aspect of the present invention, a face image extraction unit that receives a grayscale image and extracts a face image including a face from the grayscale image, and a pixel value of each pixel of the face image extracted by the face image extraction unit A density gradient image generation unit that generates a density gradient image replaced with a numerical value corresponding to the direction of the density gradient at the position, and a center that detects the center line of the face image using the density gradient image generated by the density gradient image generation unit And a line detection unit.

  According to this invention, for example, even if a shadow is generated on the face, the direction of the density gradient (density gradient direction) generated in the shadow can be used. In addition, since the density gradient direction with a large amount of information is used, a more accurate center line can be obtained as compared with the case where edges are used.

  According to a second aspect of the present invention, in the first aspect of the present invention, the face image extraction unit performs a conversion of enlargement or reduction on the extracted face image, whereby the size of the face image output to the density gradient image generation unit Is always constant.

  According to the present invention, it is possible to cope with a difference in face size in a grayscale image.

  According to a third aspect of the present invention, in the first aspect of the invention, the center line detection unit assumes a plurality of temporary center lines for the density gradient image generated by the density gradient image generation unit, and the density gradient for each temporary center line. Calculating the symmetry within a predetermined calculation area centered on the temporary center line in the image, and detecting the temporary center line having the highest symmetry among the temporary center lines as the center line of the face image; The center line detection unit is characterized in that the larger the number of pixels of the density gradient image, the wider the calculation area.

  According to the present invention, it is possible to cope with a difference in face size in a grayscale image.

  According to a fourth aspect of the present invention, in the invention according to any one of the first to third aspects, the center line detection unit assumes a plurality of temporary center lines for the density gradient image generated by the density gradient image generation unit, and each temporary center. Calculate the symmetry within a predetermined calculation area centered on the temporary center line in the density gradient image for each line, and detect the temporary center line with the highest symmetry among the temporary center lines as the center line of the face image Each temporary center line assumed by the center line detection unit has a width of 1 pixel or more, and the center line detection unit does not use the pixel value of the pixel on the temporary center line for the calculation of symmetry. It is characterized by.

  According to the present invention, a pixel that is highly likely to be a central part in the left-right direction of a face, which is a region that tends to be locally asymmetric, is set as a pixel on the temporary center line, and the pixel value is used for the calculation of symmetry. As a result, the center line can be detected more accurately than in the case where the pixel values of all the pixels in the density gradient image are used for the symmetry calculation.

  According to a fifth aspect of the present invention, in the first to fourth aspects of the invention, there is provided a face image rotation unit that generates a plurality of rotated face images that are rotated by different angles with respect to the face image extracted by the face image extraction unit. The density gradient image generation unit generates a density gradient image similar to the face image not rotated by the face image rotation unit for each rotated face image output by the face image rotation unit, and the center line detection unit For each density gradient image generated by the density gradient image generation unit, a plurality of temporary center lines parallel to the vertical direction are assumed, and each temporary center line has a predetermined center around the temporary center line in the density gradient image. A center line detection operation is performed in which the symmetry in the calculation area is calculated and the temporary center line having the highest symmetry among the temporary center lines is detected as a center line. The center line obtained from each density gradient image generated from the face image is rotated by the same angle in the opposite direction to the rotation in the face image rotation unit, and the center obtained from each density gradient image. A center line having the highest symmetry among the lines is detected as the center line of the face image.

  According to the present invention, when the face is tilted with respect to the image, the center line obtained from the rotated face image rotated by the angle closest to the tilt is detected. Compared with the case where no part is used, the center line can be detected more accurately.

  The invention according to claim 6 is the invention according to any one of claims 1 to 5, wherein the center line detection unit is a plurality of temporary centers parallel to each other in the vertical direction with respect to the density gradient image generated by the density gradient image generation unit. Assuming a line and calculating the symmetry within a predetermined calculation area around the temporary center line in the density gradient image for each temporary center line, the temporary center line having the highest symmetry among the temporary center lines is calculated. The center line detection unit detects the center line of the face image, and the center line detection unit compares the pixel values of pixels located on opposite sides of the temporary center line when calculating the symmetry, The operation of increasing the symmetric value when the direction of the density gradient indicated by the other pixel value is symmetric with respect to the temporary center line with respect to the direction of the density gradient indicated by the value, Predetermined calculation area Of all the pixels on both sides of the temporary center line, the higher the symmetry value finally obtained, the higher the symmetry, and the calculation area is divided into two divided calculation areas in the vertical direction. One of the divided calculation areas on the lower side has a larger increase width when the symmetry value increases than the other divided calculation area on the upper side.

  According to the present invention, since the lower part of the face that tends to be bilaterally symmetrical is more important than the upper part of the face that tends to be bilaterally asymmetric due to factors such as one-eye glasses and hairstyle, the value added to the symmetrical value As compared with the case where the density gradient image is common to all the density gradient images, the center line can be detected more accurately.

  According to a seventh aspect of the present invention, in any one of the first to sixth aspects, the pixel value of each pixel in the density gradient image depends only on the direction of the density gradient, and the pixel value of each pixel in the density gradient image takes a value. There are a total of four possible values that correspond one to one vertically and horizontally.

  According to the present invention, the data amount of the density gradient image is reduced and the load applied to the center line detection unit is reduced as compared with the case where the pixel value in the density gradient image can take five or more values.

  The invention according to claim 8 is the invention according to claim 7, wherein the center line detection unit assumes a plurality of temporary center lines parallel to each other in the vertical direction for the density gradient image generated by the density gradient image generation unit. Calculates the symmetry within the predetermined calculation area around the temporary center line in the density gradient image with respect to the center line, and detects the temporary center line with the highest symmetry among the temporary center lines as the center line of the face image When calculating symmetry, the pixel values are compared between pixels located on opposite sides of the temporary center line, and the pixel values coincide with each other in numerical values corresponding to the density gradient in the vertical direction. In the first case and in the second case where the pixel values are values corresponding to density gradients in the left-right direction and opposite to each other, a value corresponding to each case is added to the symmetric value. Operation, concentration It is performed for all pixels on both sides of the temporary center line in a predetermined calculation area centered on the temporary center line in the layout image, and the higher the symmetry value finally obtained, the higher the symmetry, In the calculation of symmetry, the value added to the symmetry value is larger in the first case than in the second case.

  According to the present invention, elements such as eyebrows, eyes, and mouth that cause a concentration gradient in the y direction, that is, the vertical direction, are compared with the case where the numerical value added to the symmetric value is shared between the first case and the second case. Can be reflected in the symmetry value more strongly and can be linked to the selection of a more appropriate center line.

  The invention according to claim 9 is the invention according to claim 7 or claim 8, wherein the center line detecting unit receives the density gradient image from the density gradient image generating unit and then detects the density gradient image before detecting the center line. For each numerical value that can be taken by each pixel value, the minimum number of pixels set in advance for each numerical value is compared with the number of pixels whose pixel value is the numerical value in the density gradient image. When the number of pixels is less than the minimum number of pixels, the center line is not detected using the density gradient image.

  According to the present invention, for example, the face image extraction unit clearly uses a range that does not include a face as a face image, such as when the number of pixels that have a pixel value indicating a vertical density gradient in the density gradient image is abnormally small. In the case of erroneous extraction, it is possible to avoid unnecessary detection of the center line for the erroneous face image.

  According to a tenth aspect of the present invention, in the first or second aspect of the present invention, the center line detection unit has a width corresponding to a plurality of pixels in the horizontal direction and a width corresponding to one pixel in the vertical direction in the density gradient image. It is obtained by detecting the center point that maximizes the symmetry in the left-right direction for each of the plurality of linear regions, and applying the least square method to the group of center points detected from the plurality of linear regions. It is characterized by detecting a straight line as a center line.

  According to the present invention, it is possible to detect the center line relatively accurately even when the face is inclined with respect to the image.

  According to the first aspect of the present invention, for example, even if a shadow is generated on the face, the direction of the density gradient (density gradient direction) generated in the shadow can be used. By using the density gradient direction having a larger amount of information than the edge, a more accurate center line can be obtained as compared with the case where the edge is used.

  According to the invention of claim 2, the face image extraction unit constantly enlarges or reduces the size of the extracted face image so that the size of the face image output to the density gradient image generation unit is always constant. According to the invention of claim 3, the center line detecting unit assumes a plurality of temporary center lines for the density gradient image generated by the density gradient image generating unit, and each of the temporary center lines includes a temporary line in the density gradient image. Calculating symmetry within a predetermined calculation area centered on the center line, and detecting the temporary center line with the highest symmetry among the temporary center lines as the center line of the face image. As the number of pixels of the density gradient image is larger, the unit widens the calculation area, so that each can cope with a difference in face size in the grayscale image.

  According to the invention of claim 4, each temporary center line assumed by the center line detection unit has a width of one pixel or more, and the center line detection unit calculates the pixel value of the pixel on the temporary center line for symmetry calculation. Is not used, the pixel that is likely to be the central part in the left-right direction of the face, which is a region that tends to be locally asymmetric, is regarded as the pixel on the temporary center line, and the pixel value is used for the calculation of symmetry. The absence of this makes it possible to detect the center line more accurately than when the pixel values of all the pixels of the density gradient image are used for symmetry calculation.

  According to the invention of claim 5, the centerline detection unit includes a density gradient image generated from the rotated face image output from the face image rotation unit, and the symmetry of the centerlines obtained from the respective density gradient images. Since the highest center line is detected as the center line of the face image, if the face is tilted with respect to the image, the center line obtained from the rotated face image rotated by the angle closest to the tilt is Therefore, the center line can be detected more accurately than when the face image rotation unit is not used.

  According to the invention of claim 6, the calculation area is divided into two divided calculation areas in the vertical direction, and one divided calculation area on the lower side is more symmetric than the other divided calculation area on the upper side. Since the increase range when the value increases is large, the lower part of the face that tends to be symmetric is emphasized rather than the upper part of the face that tends to be asymmetrical due to factors such as eyeglasses for one eye and hairstyle. Compared to the case where the value added to the value is common to the entire density gradient image, the center line can be detected more accurately.

  According to the seventh aspect of the present invention, the pixel value of each pixel in the density gradient image depends only on the direction of the density gradient, and the values that can be taken by the pixel value of each pixel in the density gradient image are one-to-one on the top, bottom, left, and right. Since there are 4 types in total, the data amount of the density gradient image is reduced and the load on the center line detection unit is reduced as compared with the case where the value that the pixel value in the density gradient image can take is 5 or more. The

  According to the eighth aspect of the invention, in the first case where the pixel values coincide with each other in the numerical values corresponding to the density gradient in the vertical direction, the pixel values are the density gradients in the horizontal direction and the density gradients opposite to each other. Since the value added to the symmetric value is larger than the second case, which is a numerical value corresponding to, the numerical value added to the symmetric value in the first case and the second case is compared with the case where the numerical value added to the symmetric value is common. The symmetry of the elements such as the eyebrows, eyes, and mouth that cause the density gradient in the vertical direction can be more strongly reflected in the symmetry value, and can be linked to the selection of a more appropriate center line.

  According to the ninth aspect of the present invention, the center line detection unit determines each numerical value that the pixel value of the density gradient image can take after detecting the center line after the density gradient image is input from the density gradient image generation unit. Each of the numerical values is compared with the minimum number of pixels predetermined for each numerical value and the number of pixels whose pixel value is the numerical value in the density gradient image, and for any numerical value, the number of pixels is less than the minimum number of pixels. If there is, the center line is not detected using the density gradient image. For example, as in the case where the pixel value indicating the density gradient in the vertical direction in the density gradient image is abnormally small, the face image When the extraction unit has mistakenly extracted a range that does not include a face as a face image, it is possible to avoid unnecessary detection of the center line for the erroneous face image.

  According to the tenth aspect of the present invention, the center line detection unit is configured so that each of the plurality of linear regions having a width corresponding to a plurality of pixels in the left-right direction and a width corresponding to one pixel in the vertical direction in the density gradient image. Detect the center point with the highest symmetry in the direction, and detect the straight line obtained by applying the least square method to the group of center points detected from multiple linear regions as the center line It is characterized by.

  According to the present invention, it is possible to detect the center line relatively accurately even when the face is inclined with respect to the image.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

(Embodiment 1)
As shown in FIG. 1, a face centerline detection device 1 according to the present embodiment receives a grayscale image from an imaging device 2, detects a face centerline in the inputted grayscale image, and detects a face authentication device ( A face image extracting unit 11 that extracts a face image from a grayscale image input from the imaging device 2, and each pixel of the face image extracted by the face image extracting unit 11 A density image generated by replacing the value with a numerical value corresponding to the direction of the density gradient at the position of the pixel, a density gradient image generating unit 12, and a face image using the density gradient image generated by the density gradient image generating unit 12 And a center line detection unit 13 for detecting the center line of. The imaging device 2 is a known so-called digital camera having an imaging element using, for example, a CCD or a CMOS. Furthermore, the image storage unit 14 in which the grayscale image input from the imaging device 2 is stored is provided, and the face image extraction unit 11 is stored in the image storage unit 14 as well as the grayscale image directly input from the imaging device 2. It is also possible to extract a face image from the gray image. For example, when there is an unprocessed grayscale image in the image storage unit 14, the input source of the grayscale image to the face image extraction unit 11 is set as the input source, and the image processing unit 14 is unprocessed. If there is no grayscale image, the face image extraction unit 11 automatically switches to the imaging device 2 as an input source.

  Hereinafter, the operation of each unit will be described in detail. The face image output from the face image extraction unit 11 to the density gradient image generation unit 12 is, for example, a rectangular region cut out from the grayscale image so as to include the face portion. Since a method for detecting a face portion from a grayscale image and a method for determining an area to be output as a face image can be realized by a well-known technique, detailed description thereof is omitted. In addition, the face image extraction unit 11 constantly enlarges or reduces the area cut out from the grayscale image, so that the size of the face image output to the density gradient image generation unit 12 is always constant. Accordingly, it is possible to cope with a case where the size of the area that the face image extraction unit 11 determines as a face is different in the grayscale image.

  In generating the density gradient image, the density gradient image generation unit 12 firstly, for each pixel of the face image, the density gradient dx in the left-right direction (x direction) at the position of the pixel and the vertical direction (y direction). ) With respect to the concentration gradient dy. For the calculation of the density gradients dx, dy, for example, a known Sobel filter is used. That is, as shown in FIG. 2A, the pixel value of a pixel that is the target of density gradient calculation (hereinafter referred to as “target pixel”) is P5, and the upper left, upper, upper right, The pixel values of the left, right, lower left, lower, and lower right pixels are respectively P1, P2, P3, P4, P6, P7, P8, and P9, x is positive in the right direction, and y is positive in the downward direction. As shown in FIGS. 2B and 2C, when the Sobel filter is used, the density gradient dx in the x direction at the position of the target pixel is expressed as dx = (P1 + 2 × P4 + P7) − (P3 + 2 × P6 + P9). The density gradient dy in the y direction at the position of the target pixel is expressed as dy = (P1 + 2 × P2 + P3) − (P7 + 2 × P8 + P9). The density gradients dx and dy take positive values when the pixel value decreases in the positive direction.

  If dx> 0 and dy ≧ 0, as shown in FIG. 4, the position of the target pixel is expressed using an angle expression in which the numerical value increases clockwise with the right direction (positive direction of x) being 0 °. The angle θ (hereinafter referred to as “density gradient direction”) θ indicating the direction of the density gradient at (that is, the direction in which the pixel value in the face image decreases) θ is expressed as θ = arctan (dy / dx). .

  Except for the case where the concentration gradients dx and dy are both 0, the concentration defined as described above for all the concentration gradients dx and dy by appropriate classification according to the signs of the concentration gradients dx and dy. The value of the gradient direction θ can be determined within a range of 0 ° ≦ θ <360 °. Specifically, when dx = 0, if dy> 0, θ = 90 ° and if dy <0, θ = 270 °, and if dx> 0 and dy <0, θ = arctan (dy / dx ) + 360 °, and if dx <0, θ = arctan (dy / dx) + 180 °. Then, the density gradient image generation unit 12 determines the density gradient direction θ within the above range for the target pixel from which the density gradients dx and dy are obtained.

  Next, the density gradient image generation unit 12 determines a pixel value in the density gradient image for the target pixel for which the density gradient direction θ is determined. The values that can be taken by the pixel values in the density gradient image are four values that correspond one-to-one in the four directions, up, down, left, and right. Specifically, the pixel value of the target pixel indicates a density gradient in the right direction if the density gradient direction θ is 0 ° ≦ θ <45 ° or 315 ° ≦ θ <360 ° as indicated by DR1 in FIG. If the density gradient direction θ is 45 ° ≦ θ <135 ° as indicated by DR2 in FIG. 4, the pixel value indicates a downward density gradient, and the density gradient is indicated by DR3 in FIG. If the direction θ is 135 ° ≦ θ <225 °, the pixel value indicates the density gradient in the left direction. As indicated by DR4 in FIG. 4, the direction is upward if the density gradient direction θ is 225 ° ≦ θ <315 °. The pixel value indicates the density gradient. By performing the above operation for all the pixels of the face image, a density gradient image based on the face image is generated. For example, a density gradient image DP shown in FIG. 3B is generated from the face image FP shown in FIG.

  The center line detection unit 13 receives each numerical value (that is, the above four values) that the pixel value of the density gradient image DP can take after detecting the center line after the density gradient image DP is input from the density gradient image generation unit 12. Pixel value), the number of pixels whose pixel value is the numerical value in the density gradient image DP is counted, the minimum number of pixels set in advance for each numerical value that the pixel value can take, and the density gradient The pixel number of the image DP is compared with the number of pixels, and if the number of pixels is less than the minimum number of pixels for any value, the center line is detected using the density gradient image DP. Absent. As a result, for example, when the number of pixels having a pixel value indicating a vertical density gradient in the density gradient image DP is abnormally small, the face image extraction unit 11 clearly sets a range that does not include a face as the face image FP. If it is extracted by mistake, it can be avoided that the center line is detected unnecessarily for the erroneous face image FP. In addition, what is necessary is just to determine about the specific numerical value of the minimum pixel number with reference to the data in an actual face image.

  On the other hand, if the number of pixels in the input density gradient image DP is greater than or equal to the minimum number of pixels for all the numerical values that can be taken by the pixel values of the density gradient image DP, the center line detection unit 13 inputs the density gradient image. A center line detection operation for detecting the center line in the DP is started. In the center line detection operation, the center line detection unit 13 first, as shown in FIGS. 3C to 3E, a plurality of temporary center lines in the density gradient image DP input from the density gradient image generation unit 12. Assuming CL1 to CL3, calculation areas CZ1 to CZ3 centering on the temporary center lines CL1 to CL3 are set for the temporary center lines CL1 to CL3, respectively. In the present embodiment, each of the temporary center lines CL1 to CL3 is a straight line parallel to the vertical direction (y direction), has a width of at least one pixel in the x direction, and a length of a plurality of pixels in the y direction. It is an area | region which has. The calculation areas CZ1 to CZ3 are rectangular areas set symmetrically about the temporary center lines CL1 to CL3 as the centers in the left-right direction, and the shapes and sizes of all the calculation areas CZ1 to CZ3 are the same. It is common. Further, the temporary center lines CL1 to CL3 are respectively arranged in the center of the density gradient image DP in the y direction (up and down direction), and the calculation areas CZ1 to CZ3 fit in the density gradient image DP in the x direction (left and right direction). Within the range, the temporary center lines CL1 to CL3 are slightly shifted in position. Then, the center line detection unit 13 calculates the symmetry values SP1 to SP3, which are numerical values indicating the symmetry with respect to the temporary center line CL1, for each of the calculation regions CZ1 to CZ3. Although the number of temporary center lines CL1 to CL3 is three in the figure, the number of temporary center lines is not limited to this. For example, the position in the x direction is shifted by one pixel within a range where a calculation area is secured. As many temporary centerlines as possible may be assumed.

  Hereinafter, the calculation method of the symmetric value SP1 with respect to one temporary center line CL1 will be described with reference to FIGS. When the center line detector 13 starts to calculate a symmetric value (S1), first, the symmetric value SP1 is returned to the initial value 0 (reset) (S2). Next, in the calculation area CZ1 shown in FIG. 6, the pixel value of one pixel (hereinafter referred to as “target pixel”) P on one side (for example, the left side) of the temporary center line CL1 and the temporary center line CL1 are sandwiched. The pixel value of one pixel (hereinafter referred to as “corresponding pixel”) P ′ located on the opposite side (symmetrical position) is compared. First, it is determined whether or not the pixel values corresponding to the density gradient in the vertical direction (y direction) match (S3). When the pixel values match, that is, when the pixel values corresponding to the upward direction match or the pixel values corresponding to the downward direction match, 2 is added to the symmetry value SP1 (S4), and the next step Proceed to S7. If they do not match in step S3, then whether or not the pixel values are pixel values corresponding to the density gradient in the left-right direction (x direction) and corresponding to opposite directions. That is, it is determined whether one of the pixel values is a pixel value corresponding to the left direction and the other of the pixel values is a pixel value corresponding to the right direction (S5). If the pixel values are as described above in step S5, 1 is added to the symmetry value SP1 (S6), and the process proceeds to the next step S7. In step S5, if the pixel values do not have the above relationship, the symmetry value SP1 is not changed and the process proceeds to the next step S7. That is, the symmetry value SP1 is added according to the direction of the concentration gradient only when the direction of the concentration gradient is symmetric with respect to the temporary center line CL1. In the next step S7, it is determined whether or not there is a pixel that has not yet become the target pixel P on the same side (left side in FIG. 6) of the temporary center line CL1 in the calculation region CZ1. For example, any of the pixels that are not yet the target pixel P is newly set as the target pixel P (S8), and the process returns to step S3. As a method for selecting the target pixel P, specifically, for example, the upper left pixel of the calculation area CZ1 is first set as the target pixel P, and the position of the target pixel P is shifted to the right by one pixel each time step S8 is reached. Then, in the next step S8 when the target pixel P reaches the left side of the temporary center line CL1, the position of the target pixel P is shifted downward by one pixel, and the position of the target pixel P is moved to the right again in order from the left end of the calculation area CZ1. The operation of changing is repeated until the pixel on the left side of the temporary center line CL1 and the lower end of the calculation area CZ1 becomes the target pixel P. Then, when all the pixels on one side (left side in FIG. 6) of the temporary center line CL1 in the calculation area CZ1 become the target pixel P one time, that is, on both sides of the temporary center line CL1 (that is, the temporary center line CL1 in the calculation area CZ1). When all the pixels other than the pixel on the center line CL1 are to be compared (that is, either the target pixel P or the corresponding pixel P ′), the symmetric value SP1 for the temporary center line CL1 and the calculation region CZ1. This calculation is finished (S9). That is, the symmetry value SP1 is added when the direction of the density gradient shown in the pixel values of the pixels P and P ′ located on the opposite sides with respect to the temporary center line CL1 is symmetric with respect to the temporary center line CL1. The higher the symmetry of the concentration gradient in the calculation area CZ1 with respect to the temporary center line CL1, the greater the finally obtained symmetry value SP1.

  The center line detector 13 calculates the symmetry values SP1 to SP3 as described above for all the temporary center lines CL1 to CL3. Then, when the symmetry values SP1 to SP3 are obtained for all the temporary center lines CL1 to CL3, the center line detection unit 13 determines which of the temporary center lines CL1 to CL3 has the highest symmetry value SP1 to SP3. The temporary center lines CL1 to CL3 (for example, the central temporary center line CL2) having the highest symmetry values SP1 to SP3 are output as the center lines of the density gradient image DP (that is, the original face image FP), End the centerline detection operation.

  According to the above configuration, for example, even if a shadow is generated on the face, the density gradient direction generated in the shadow can be used, so that the density has a larger amount of information than the edge used in the conventional example. By using the gradient direction, a more accurate center line can be obtained as compared with the case of using the edge.

  Also, since the four possible pixel values in the density gradient image DP are four, the data amount of the density gradient image DP is larger than in the case where there are five or more possible pixel values in the density gradient image. Is reduced, and the load on the center line detection unit 13 is reduced.

  Further, when calculating the symmetry, the case where the concentration gradients are in the same direction in the vertical direction as in step S4 is more than the case where the concentration gradients are in the opposite direction in the left-right direction as in step S6. Since the numerical value added to the symmetric values SP1 to SP3 is increased, for example, the eyebrows that cause the density gradient in the vertical direction compared to the case where the numerical values added to the symmetric values SP1 to SP3 are common in the above cases, for example. As a result, the symmetry of elements such as the eyes and the mouth can be more strongly reflected in the symmetry values SP1 to SP3, and can be linked to the selection of a more appropriate center line.

  Further, although the central part in the left-right direction of the face is likely to be locally asymmetric, in this embodiment, a pixel that is highly likely to be the central part in the left-right direction of the face is set as a pixel on the temporary center line CZ1, Since the pixel values are not used for the calculation of the symmetric value SP1, it is possible to detect the center line more accurately than when the pixel values of all the pixels of the density gradient image DP are used for the calculation of objectivity. .

  Note that the face image extraction unit 11 outputs the size of the face image FP output to the density gradient image generation unit 12 at a size that is cut out from the original grayscale image instead of always keeping the size of the face image FP constant. The center line detection unit 13 increases the lengths of the temporary center lines CL1 to CL3 and increases the calculation areas CZ1 to CZ3 as the size of the density gradient image DP increases (that is, as the number of pixels of the density gradient image DP increases). And the number of temporary center lines CL1 to CL3 to be assumed may be increased. In this case, the minimum number of pixels used in the determination performed prior to the detection of the center line is also changed according to the size of the density gradient image. For example, the minimum number of pixels for each pixel value is calculated by multiplying a predetermined coefficient for each pixel value by the total number of pixels of the density gradient image DP.

  Further, in the calculation of symmetry, each of the calculation areas CZ1 to CZ3 is divided into upper and lower divided calculation areas, and a numerical value (that is, an increase width) added when the symmetry values SP1 to SP3 are added, The divided calculation area (for example, the lower half of the calculation areas CZ1 to CZ3) may be larger (for example, doubled) than the upper divided calculation area. By adopting this configuration, the lower center part of the face, which tends to be bilaterally symmetrical, is more important than the upper part of the face, which tends to be bilaterally asymmetric due to factors such as eyeglasses for one eye and hairstyle. Can be detected.

(Embodiment 2)
Since the basic configuration of the present embodiment is the same as that of the first embodiment, common portions are denoted by the same reference numerals, and illustration and description thereof are omitted.

  In the present embodiment, as shown in FIGS. 7 and 8, a face image rotation unit that generates a plurality of rotated face images RP rotated by different angles with respect to the face image FP extracted by the face image extraction unit 11. 15. The maximum value of the angle by which the face image rotation unit 15 rotates the face image FP is, for example, the maximum value as long as the angle at which the center line of the face can tilt with respect to the vertical direction (y direction) of the image is assumed. Further, only two rotated face images RP are shown in FIG. 8, but rotated face images RP with different rotation angles may be further added.

  The density gradient image generation unit 12 generates a density gradient image similar to the face image FP that has not been rotated by the face image rotation unit 15 for each rotated face image RP output by the face image rotation unit 15.

  Further, the center line detection unit 13 assumes a plurality of temporary center lines for each density gradient image generated by the density gradient image generation unit 12, and determines the temporary center line in the density gradient image for each temporary center line. A centerline detection operation similar to that of the first embodiment is performed in which symmetry within a predetermined calculation area as a center is calculated and a centerline having the highest symmetry among the temporary centerlines is detected as a centerline. The center line obtained from the density gradient image generated from the rotated face image RP is rotated in the opposite direction to the rotation in the face image rotation unit 15 by the same angle (that is, to cancel the rotation in the face image rotation unit 15). Then, the center line having the largest symmetry value (ie, the highest symmetry) among the center lines obtained from the density gradient images generated from the common face image FP is selected. It is detected as the center line of the final facial image FP.

  According to the above configuration, when the exact center line of the face is tilted with respect to the vertical direction of the image, the center line obtained from the rotated face image RP rotated by an angle closest to the tilt is detected. Therefore, the center line can be detected more accurately than in the first embodiment.

(Embodiment 3)
Since the basic configuration of the present embodiment is the same as that of the first embodiment, common portions are denoted by the same reference numerals, and illustration and description thereof are omitted.

  In the present embodiment, the content of the center line detection operation of the center line detection unit 13 is different from that of the first embodiment.

  More specifically, when the density gradient image DP is input from the density gradient image generation unit 12, the center line detection unit 13 first has a plurality of values in the x direction in the input density gradient image DP as shown in FIG. A plurality of linear regions LZ having a width for pixels and a width for one pixel in the y direction are set. For example, the linear regions LZ are set at equal intervals (every several pixels) in the y direction.

  Next, the center line detection unit 13 detects the center point CP that maximizes the symmetry in the x direction for each linear region LZ. The detection of the center point CP in each linear region LZ assumes a plurality of temporary center points (corresponding to the temporary center lines CL1 to CL3 in the first embodiment) in the same manner as the detection of the center line in the first embodiment. For each temporary center point, a symmetric value within a predetermined width (corresponding to the calculation regions CZ1 to CZ3 in the first embodiment) is calculated, and the temporary center point having the highest symmetry value is determined as the center point CP of the linear region LZ. It can be done by.

  When the center point CP is detected in all the linear regions LZ, the center line detection unit 13 uses the straight line obtained by applying the least square method to the group of detected center points CP as the center line. Detect as CL.

  According to the above configuration, in addition to the same effects as those of the first embodiment with respect to the parts common to the first embodiment, even when the accurate center line of the face is inclined with respect to the vertical direction (y direction) of the image, A relatively accurate center line CL can be detected.

It is a block diagram which shows Embodiment 1 of this invention. (A)-(c) is explanatory drawing used for description of operation | movement of a density gradient image generation part same as the above, respectively, (a) shows the correspondence between the code | symbol in a text, and the position of a pixel, (b) Indicates a Sobel filter for the x direction, and (c) indicates a Sobel filter for the y direction. (A)-(e) is explanatory drawing which respectively shows operation | movement same as the above, (a) shows the example of a face image, (b) shows the density | concentration gradient image produced | generated from the face image of (a), (C)-(e) shows the example of the temporary center line and calculation area | region assumed with respect to the density | concentration gradient image of (b), respectively. It is explanatory drawing which shows the correspondence of the pixel value of the density gradient image in the same as the above, and the direction of a density gradient. It is a flowchart which shows the operation | movement of the symmetrical value calculation of a centerline detection part same as the above. It is explanatory drawing which shows the positional relationship of the pixels by which a pixel value is compared in the operation | movement of the symmetrical value calculation of a centerline detection part same as the above. It is a block diagram which shows Embodiment 2 of this invention. It is explanatory drawing which shows operation | movement of a face image rotation part same as the above. It is explanatory drawing which shows operation | movement of the centerline detection part of Embodiment 3 of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Face centerline detection apparatus 11 Face image extraction part 12 Density gradient image generation part 13 Centerline detection part 15 Face image rotation part CL Centerline CL1-CL3 Temporary centerline CP Center point CZ1-CZ3 Calculation area DP Density gradient image FP Face Image LZ Linear area

Claims (10)

  A face image extraction unit that inputs a grayscale image and extracts a face image including a face from the grayscale image, and a pixel gradient value of each pixel of the face image extracted by the face image extraction unit is a direction of a density gradient at the position of the pixel A density gradient image generation unit that generates a density gradient image replaced with a numerical value corresponding to the density gradient image generation unit, and a center line detection unit that detects the center line of the face image using the density gradient image generated by the density gradient image generation unit. A face centerline detection device characterized by the above.   2. The face image extraction unit is characterized in that the size of the face image output to the density gradient image generation unit is always constant by performing enlargement or reduction conversion on the extracted face image. Face centerline detection device. The center line detection unit assumes a plurality of temporary center lines for the density gradient image generated by the density gradient image generation unit, and within a predetermined calculation area centered on the temporary center line in the density gradient image for each temporary center line. Is calculated, and the temporary center line having the highest symmetry among the temporary center lines is detected as the center line of the face image,
The face centerline detection device according to claim 1, wherein the centerline detection unit widens the calculation area as the number of pixels of the density gradient image increases. The center line detection unit assumes a plurality of temporary center lines for the density gradient image generated by the density gradient image generation unit, and within a predetermined calculation area centered on the temporary center line in the density gradient image for each temporary center line. Is calculated, and the temporary center line having the highest symmetry among the temporary center lines is detected as the center line of the face image,
Each temporary center line assumed by the center line detection unit has a width of one pixel or more, and the center line detection unit does not use a pixel value of a pixel on the temporary center line for calculation of symmetry. Item 4. The face centerline detection device according to any one of Items 1 to 3. A face image rotation unit that generates a plurality of rotated face images rotated by different angles with respect to the face image extracted by the face image extraction unit;
The density gradient image generation unit generates a density gradient image similar to a face image that has not been rotated by the face image rotation unit for each rotated face image output by the face image rotation unit,
The center line detection unit assumes a plurality of temporary center lines parallel to the vertical direction for each density gradient image generated by the density gradient image generation unit, and the temporary center line in the density gradient image with respect to each temporary center line. Performing a center line detection operation of calculating a symmetry in a predetermined calculation area centered on the center line and detecting a temporary center line having the highest symmetry among the temporary center lines as a center line,
The center line detection unit rotates each density gradient after rotating the center line obtained from each density gradient image generated from the rotated face image by the same angle in the opposite direction to the rotation in the face image rotation unit. 5. The face centerline detection device according to claim 1, wherein a centerline having the highest symmetry among centerlines obtained from an image is detected as a centerline of a face image. . The center line detection unit assumes a plurality of temporary center lines parallel to the vertical direction of the density gradient image generated by the density gradient image generation unit, and each temporary center line is centered on the temporary center line in the density gradient image. Calculating the symmetry within the predetermined calculation area, and detecting the temporary center line having the highest symmetry among the temporary center lines as the center line of the face image,
When calculating the symmetry, the center line detection unit compares pixel values between pixels located on opposite sides of the temporary center line, and the other pixel with respect to the direction of the density gradient indicated by one pixel value. When the direction of the density gradient indicated by the value is symmetric with respect to the temporary center line, an operation of increasing the symmetry value is performed on both sides of the temporary center line in a predetermined calculation area centered on the temporary center line in the density gradient image It is performed for all pixels, and the higher the symmetry value finally obtained, the higher the symmetry,
The calculation area is divided into two divided calculation areas in the vertical direction, and one of the divided calculation areas on the lower side has a larger increase width when the symmetrical value increases than the other divided calculation area on the upper side. The face centerline detection apparatus according to claim 1, wherein the face centerline detection apparatus is a face centerline detection apparatus.   The pixel value of each pixel in the density gradient image depends only on the direction of the density gradient, and the pixel value of each pixel in the density gradient image can be a total of four values that correspond one-to-one vertically and horizontally. The face centerline detection device according to claim 1, wherein the face centerline detection device is a feature of the face centerline detection device. The center line detection unit assumes a plurality of temporary center lines parallel to the vertical direction of the density gradient image generated by the density gradient image generation unit, and each temporary center line is centered on the temporary center line in the density gradient image. Is calculated as the center line of the face image, and the temporary center line is detected as the center line of the face image. A pixel value is compared between pixels located on the opposite sides of the line, and the pixel value is a value corresponding to a density gradient in the vertical direction. An operation of adding a value corresponding to each case to a symmetric value in the second case, which is a numerical value corresponding to a concentration gradient that is a gradient and opposite to each other, is centered on the temporary center line in the concentration gradient image. Predetermined Performed for all the pixels on both sides of the temporary center line of the calculation region, the finally obtained symmetric values been made to a high higher symmetry,
The face centerline detection device according to claim 7, wherein in the calculation of symmetry, the value added to the symmetry value is larger in the first case than in the second case.   The center line detection unit is predetermined for each numerical value for each numerical value that can be taken by the pixel value of the density gradient image before the center line is detected after the density gradient image is input from the density gradient image generation unit. The minimum number of pixels is compared with the number of pixels whose pixel value is the numerical value in the density gradient image. If the number of pixels is less than the minimum pixel number for any numerical value, the density gradient image is The face centerline detection device according to claim 7 or 8, wherein the centerline used is not detected.   The center line detection unit maximizes the symmetry in the left-right direction for each of the plurality of linear regions each having a width of a plurality of pixels in the vertical direction and a width of one pixel in the horizontal direction in the density gradient image. The center point is detected, and a straight line obtained by applying the least square method to a group of center points detected from a plurality of linear regions is detected as the center line. Item 3. The face centerline detection device according to Item 2.