JP2007139535A - Face end detection device and method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To detect both end positions of a face more accurately based on a face image having a possibility of wrongly detecting both end positions of a face, such as a face image photographed in the nighttime or a female face image, to the furthermore inside than actual. <P>SOLUTION: A face image acquired by photographing a driver's face in the nighttime by a camera 10 or a face image acquired by photographing a female driver's face or the like is stored in an image memory 22. A CPU 24 detects a temporary contour position in the right and left directions of the face based on pixel data of the face image. The CPU 24 determines a plurality of contour candidate positions which are candidates for the contour position based on pixel data positioned on the periphery of the detected temporary contour position, and uses a position extracted based on statistics from the plurality of determined contour candidate positions as the contour position. Hereby, both end positions of the face can be detected more accurately based on the face image photographed in the nighttime, the female face image or the like. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a face edge detection apparatus, a face edge detection method, and a program capable of detecting both ends of a face more accurately from a face image.

  Accurate detection of both ends or contours of a face based on the face image is an important issue in the image processing of the face image.

  Patent Document 1 discloses a technique for processing an image obtained by photographing a human face and detecting the position of the face with high accuracy without being affected by the movement or background of the person.

Patent Document 2 captures an image obtained by photographing an area where a driver can exist, detects a vertical edge image (both side edges of the face) from the captured image, and based on the detected vertical edge image. In addition to setting a plurality of face center candidate lines, for each face center candidate line, the accuracy of determining the presence or absence of a face is improved based on a value representing the likelihood that the face center candidate line is the face center line. Techniques to do this are disclosed.
JP 2004-310396 A JP 2005-011097 A

The driving environment of the vehicle changes variously, and the driver's image captured by the camera also changes variously. For example, at night, the difference between the brightness values near the both ends of the face is not clear, and it is often impossible to detect the exact positions of both ends of the face. For example, when the driver is a woman, due to the influence of the hairstyle and makeup, the difference between the brightness values in the vicinity of both ends of the face is not clear, and it is often impossible to detect the exact positions of both ends of the face.
If an error occurs in the detection of both end positions of the face, the subsequent processing cannot be performed accurately.

  The present invention has been made in view of the above problems, and even in a face image that may be erroneously detected on the inner side than the actual position of both ends of the face, such as an image photographed at night or a female face image, It is an object of the present invention to provide a face edge detection device, a face edge detection method, and a program capable of detecting both end positions of a face.

  In order to achieve the above object, a face edge detection device according to a first aspect of the present invention stores a face image storage means for storing a face image obtained by photographing a face and the face image storage means. Process the pixel data of the face image to detect a temporary contour position in the left-right direction of the face, and based on the pixel data located around the detected temporary contour position, a plurality of contour candidates that are candidates for the contour position A contour candidate position determining unit that determines a position; and a contour position correcting unit that uses a position extracted from the plurality of contour candidate positions determined by the contour candidate position determining unit as a contour position. .

  Further, in the face position detection device, the contour candidate position determination unit processes the pixel data of the face image stored in the face image storage unit, and detects a temporary contour position in the horizontal direction of the face. Based on position detection means, pixel data located adjacent to the temporary contour position in the left-right direction detected by the temporary contour position detection means, and two pixel data located above and below the pixels located adjacent to each other, Determining means for determining whether or not to change the temporary contour position to the position of the pixel located next to the temporary contour position, and when the determination means determines to change the temporary contour position, The changing means for changing the position, the determining means, and the changing means are repeatedly operated so that the temporary contour position when the determining means determines not to change the temporary contour position is the contour candidate position. Hand When, may be characterized in that it comprises a.

  Further, in the face position detecting device, the contour position correcting means creates a histogram of the plurality of contour candidate positions determined by the contour candidate position determining means, and a histogram created by the histogram creating means. A standard deviation calculating means for calculating a standard deviation of the contour candidate position, and a contour candidate position located on the outermost side of the face image among the contour candidate positions within the standard deviation calculated by the standard deviation calculating means. And means for setting the contour position.

  The face edge detection method according to the second aspect of the present invention includes a face image storage step for storing a face image obtained by photographing a face, and pixel data of the face image stored in the face image storage step. To detect a temporary contour position in the left-right direction of the face, and determine a plurality of contour candidate positions that are candidates for the contour position based on pixel data located around the detected temporary contour position A candidate position determining step; and a contour position correcting step using a position extracted from the plurality of contour candidate positions determined in the contour candidate position determining step as a contour position.

  According to a third aspect of the present invention, there is provided a program for storing a face image obtained by photographing a face in a computer, and pixel data of the face image stored in the face image storing procedure. To detect a temporary contour position in the left-right direction of the face, and determine a plurality of contour candidate positions that are candidates for the contour position based on pixel data located around the detected temporary contour position A candidate position determination procedure and a contour position correction procedure in which positions extracted from the plurality of contour candidate positions determined in the contour candidate position determination procedure are used as a contour position are executed.

  According to the present invention, both end positions of a face can be detected more accurately even in a face image such as a face image photographed at night or a face image of a woman that may misdetect both ends of the face inside. be able to.

  Hereinafter, a face edge detection device according to an embodiment of the present invention will be described.

  As shown in FIG. 1, the face edge detection device of this embodiment includes a camera 10 that captures a driver's face and generates a face image, an illumination light source 12 that illuminates the driver's face, and both end positions of the driver's face. And a display device 16 connected to the computer 14.

  The camera 10 is composed of a CCD camera, for example, and acquires a gradation image of the driver's face. The face image generated by the camera 10 includes not only the driver's face but also its background.

  The display device 16 is composed of an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube), or the like, and displays a binarized image generated from a face image taken by the camera 10.

  The computer 14 is a device that processes the face image acquired by the camera 10 and detects the left and right end positions of the face. However, the positions of both ends of the face detected first in the face image taken at night or the face image of a woman are often erroneously detected from the center of the face. Therefore, the computer 14 corrects the both end positions of the face with respect to a portion that can be detected erroneously based on the left and right end positions detected first, and then obtains the final left and right end positions of the face. As shown in FIG. 2, the computer 14 includes an image capture device 21, an image memory 22, a ROM 23, a CPU 24, a RAM 25, a display control device 27, and a light source control device 28.

The image capture device 21 converts an analog image signal captured by the camera 10 into a digital signal.
The image memory 22 stores image data generated by the camera 10 and digitized by the image capture device 21.

  The ROM 23 stores a program for controlling the operation of the CPU 24. Further, the ROM 23 stores various fixed data for executing image processing to be described later.

  The CPU 24 executes a program stored in the ROM 23 to process the face image acquired by the camera 10 and detect both end positions of the face. Then, the CPU 24 corrects the both end positions based on the detected both end positions, and obtains the left and right end positions of the face again.

The RAM 25 functions as a work area for the CPU 24.
The display control device 27 controls the display device 16 under the control of the CPU 24.
The light source control device 28 controls turning on / off of the illumination light source 12.

  Next, an example of fixed data stored in the ROM 23 will be described with reference to FIG. First, the ROM 23 stores operators of forward and reverse vertical edge detection and horizontal edge detection Sobel filters as shown in FIGS. 3 (a), 3 (b), and 3 (c). The forward / reverse vertical edge detection Sobel filter and horizontal edge detection Sobel filter shown in FIGS. 3A, 3B, and 3C are respectively shown in FIGS. ), An operator for emphasizing the vertical gray level difference and the horizontal gray level difference as shown in (f).

The ROM 23 stores various programs such as a program for detecting both end positions of the face, a program for correcting both end positions of the face, and a program for performing statistical processing. Further, as shown in FIG. 3G, it is assumed that, among the face images stored in the RAM 25, data for specifying positions where the images of the eyes and eyebrows are present, and both end positions of the face are present. Data for specifying the position to be stored may be stored. Specifically, the coordinate value i of the right end position of the face, the coordinate value j of the left end position of the face, the coordinate value m of the position above the eyes and eyebrows, and the coordinate value n of the position below the mouth are stored.
Furthermore, the ROM 23 stores parameters that are fixed values as shown in FIG. Specifically, the maximum number of moving pixels for correcting both end positions of the face in the horizontal direction is stored as 5. Further, a vertical Sobel threshold value for determining whether or not the pixel is a vertical edge is stored as 8 from the luminance value of each pixel processed using the Sobel filter for detecting the vertical edge. Further, the boundary value data number ratio for determining whether or not to correct is stored as 40% according to the ratio of the correction candidate pixel number to the predetermined number of pixels arranged in the vertical direction.

Next, the operation of the face edge detection apparatus having the above configuration will be described.
When the power is turned on, the CPU 24 in the computer 14 repeatedly executes the process shown in FIG.

  That is, the CPU 24 repeatedly executes a process composed of a preprocess (step S01), a face position detection process (step S02), and a correction process (step S03).

  The pre-processing (step S01) and face position detection processing (step S02) constituting the processing shown in FIG. 4 are processing for obtaining the position (coordinate values) of the driver in the horizontal direction and vertical direction.

  The correction process (step S03) is a process for correcting the left and right face end positions of the driver obtained in step S02 to obtain more accurate left and right face end positions.

  As shown in FIG. 5, the preprocessing (step S01) includes a capture process (step S011), a coordinate conversion process (step S012), and a Sobel filter process (step S013).

  The capture process (step S011) is a process of capturing a face image for one frame of the driver photographed by the camera 10 through the image capture device 21 and storing it in the image memory 22.

  The coordinate conversion process (step S012) is a process of thinning out pixels to the extent that they can be processed.

  The Sobel filter process (step S013) emphasizes the vertical edges in the face image by processing the face image after coordinate conversion using the vertical edge detection operator (FIG. 3A) stored in the ROM 23. In addition, a process for performing a process for emphasizing the horizontal edge in the face image by processing the face image after the coordinate conversion using the horizontal edge detection operator (FIG. 3B).

  The face position detection process (step S02) in FIG. 4 is a process for detecting the left and right end positions and the vertical position of the face using the preprocessed face image. As shown in FIG. S021) and face vertical position detection processing (step S022).

  The face both ends detection process (step S021) is a process for identifying lines constituting both ends of the face of the face image operated by the operator for detecting the vertical edge, and any known method can be adopted. it can.

  In order to identify the lines constituting the both ends of the face, for example, as shown in FIG. 7, a histogram creation process for detecting both ends of the face is performed (step S0211). Subsequently, a predetermined number of high histogram peak values are extracted and sorted (step S0212), and end points are extracted based on the histogram values (step S0213). For example, if the top one or two of the histogram values are extremely large compared to the other, that point is set as the end point.

  Next, it is determined whether or not two end points (both ends) have been extracted (step S0214). If two are extracted (step S0214; Yes), the process proceeds to step S0216. If two end points have not been extracted (step S0214; No), the end point is determined by extracting a combination in which the distance between the two points has a plausible interval as a human face width (step S0215). Then, a process for determining both ends (x coordinate values) of the face is performed (step S0216).

  Further, as disclosed in Patent Document 1, the time derivative of the captured image is calculated, and the pixel value of the pixel value time derivative image is projected in the vertical direction to create a histogram. The histogram and the histogram of the pixel value time-differential image may be summed to extract the one having a high peak value of the histogram, determine a plausible point as a human face width, and detect the face edge position.

  Next, the face vertical position detection process in step S022 of FIG. 6 is a process for detecting the position of the eyebrow (upper end) and the position of the mouth (lower end) by performing the same process as described above for the horizontal edge. For example, as shown in FIG. 8, the process includes a histogram creation process (step S 0221), a current candidate detection process (step S 0222), and a face vertical position calculation process (step S 0233).

  The histogram creation process (step S0221) is a process of creating a histogram by projecting the value of each pixel after the Sobel filter process using the horizontal edge detection Sobel filter in the horizontal direction.

The current candidate detection process (step S0222) is a process of selecting a histogram value candidate corresponding to the eyes, eyebrows, mouth, and the like based on the histogram value.
The face vertical position calculation process (step S0223) is a process for detecting the upper and lower end positions (y coordinate values) (for example, the positions of eyes and eyebrows) from the selected candidates. The face upper end position (y coordinate value) is, for example, a position above a detected eyebrow by a predetermined number of pixels (for example, 3 pixels), and the lower end position is a predetermined number of pixels from the detected mouth (for example, for example, It is set to a lower position (between mouth and chin) or the like.

The CPU 24 stores the both end (left and right end) positions (x coordinate values) of the face and the vertical position (y coordinate values) of the face obtained in steps S021 and S022 in this manner in the RAM 25. Specifically, as shown in FIG. 9, the x coordinate value of the right end of the face obtained in step S021 is stored as the right end initial value a, and the x coordinate value of the left end of the face obtained in step S022 is stored as the left end initial value b. To do. Further, the y coordinate value of the upper end of the upper and lower position of the face obtained in step S022 is stored as the lower limit value c, and the y coordinate value of the lower end of the face obtained in step S022 is stored as the upper limit value d. When the positions of the right end initial value a, the left end initial value b, the lower limit value c, and the upper limit value d are shown as an example in a female face image, the positions shown in FIG. 9 are obtained.
FIG. 9 is a diagram after a female face image is processed using a vertical edge detection operator and binarized. The case where the processing is performed using the forward direction vertical detection operator shown in FIG. 3A and the case where the processing is performed using the reverse direction vertical detection operator shown in FIG. And binarized images that are distinguished from light gray and superimposed.

  Next, the correction process (step S03) in FIG. 4 has a configuration shown in, for example, the correction process 1 in FIG. 10 and the correction process 2 in FIG. The correction process is a process for correcting the position of the vertical edge of the face image taken at night or the face image of a woman because the vertical edges at both ends of the face are often detected inside the face. The correction process 1 is a correction process related to the right end position of the face of the face image, and the correction process 2 is a correction process related to the left end position of the face of the face image. The correction process 1 and the correction process 2 may be performed in parallel, or the correction process 2 may be started sequentially after the correction process 1 is completed.

  The operation of the correction process 1 related to the right edge position of the face will be described with reference to the flowchart shown in FIG. First, the CPU 24 sets the y coordinate value (hereinafter referred to as the search y coordinate value) of the pixel to be processed in the face image to 0 and starts processing (step S031).

  Subsequently, the CPU 24 reads, from the RAM 25, the right end initial value a of the face end position (x coordinate value) obtained in step S021, and the x coordinate value (hereinafter referred to as the search x coordinate value) of the pixel to be processed in the face image. ) Is set to a, the search y-coordinate value is incremented, and the search coordinate value is set to (a, 1) (step S032).

  Further, the CPU 24 reads the lower limit value c of the vertical position (y coordinate value) of the face obtained in step S02 from the RAM 25, and determines whether or not the search y coordinate value is equal to or greater than the lower limit value c (step S033). If the search y-coordinate value is not greater than or equal to the lower limit value c (step S033; No), the process returns to step S032, and the search y-coordinate value is incremented. Thereby, in the face image, useless processing in pixels outside the vertical search range where the y coordinate value shown in FIG. 3F is in the range of the lower limit value c or more and the upper limit value d or less is omitted.

  On the other hand, if the CPU 24 determines in step S033 that the search y-coordinate value is equal to or greater than the lower limit value c (step S033; Yes), then the upper limit of the vertical position (y-coordinate value) of the face obtained in step S02. The value d is read from the RAM 25. Then, the CPU 24 determines whether or not the search y coordinate value is greater than or equal to the upper limit value d (step S034).

  Processing in a case where it is determined in step S034 that the search y coordinate value is greater than or equal to d will be described later. If it is determined in step S034 that the search y-coordinate value is not greater than or equal to d (step S034; Yes), the CPU 24 determines whether or not the search x-coordinate value is greater than or equal to a value obtained by subtracting 5 from the right end initial value a read from the RAM 25. It discriminate | determines (step S035). This is because the range (x coordinate direction) to be corrected from the right end initial value a of the both end positions of the face obtained in step S021 is limited to 5 pixels. This limited range is stored in advance in the ROM 25 or the like as a parameter for the horizontal search range, as shown in FIG. Note that this limited range does not have to be 5 pixels and may be any value.

  If the CPU 24 determines in step S035 that the search x coordinate value is equal to or greater than the value (a-5) obtained by subtracting 5 from the right end initial value a (step S035; Yes), the search coordinate value is set to (x, y). Luminance values after processing by the vertical edge detection operator at the pixels at the coordinates of (x-1, y-1), (x-1, y), (x-1, y + 1) Refer to the value). The positional relationship of these coordinates is shown in FIG. Then, the CPU 24 determines whether or not all the vertical sobel values of these three coordinates are larger than 8 which is a vertical sobel threshold value stored in advance in the ROM 23 as shown in FIG. 3 (h) (step S036). . This is because the position of the (x-1, y) coordinate is more likely to be the vertical edge than the current search coordinate position if the vertical Sobel value in the pixels of these three coordinates is larger than the vertical Sobel threshold value. It is.

  If the CPU 24 determines in step S036 that the vertical sobel values of the three coordinates referred to are larger than the vertical sobel threshold value (step S036; Yes), the search x coordinate value is set to x-1 (step S037). Return to processing.

By repeating the processes in steps S035, S036, and S037, the CPU 24 scans the image toward the outside of the face while determining whether or not the both end positions (x coordinate values) of the face should be corrected. ing.
The criteria for determining whether or not to correct both end positions (x coordinate values) of the face are three coordinates (x-1, y-1), (x-1, y), and (x-1, y + 1). It is not necessary to be based on the vertical Sobel value of the pixel located, and is arbitrary.

  On the other hand, if the CPU 24 determines in step S035 that the search x-coordinate value is equal to or greater than the value (a-5) obtained by subtracting 5 from the right end initial value a (step S035; No), the CPU 24 sets the current search x-coordinate value. The boundary value data (right end candidate value) of the right end position (x coordinate value) of the face is stored in the RAM 25 (step S038). Here, the boundary value data (right end candidate value) is a coordinate value that becomes a candidate for the right end position of the face, and the right end position of the face is finally determined from the boundary value data (right end candidate value). To do.

  Even when it is determined in step S036 that the vertical sobel values of the three coordinates referred to are not larger than the vertical sobel threshold value (step S036; No), the current search x-coordinate value is used as the right edge position of the face (x-coordinate value). ) Is stored in the RAM 25 (step S038). The retrieved x-coordinate values in step S038 are stored in a boundary value recording array A (hereinafter referred to as array A) as shown in FIG. An array is prepared for each search y coordinate, and the search x coordinate value is stored as boundary value data (right end candidate value).

  The CPU 24 determines whether or not the boundary value data (right end candidate value) stored in the RAM 25 in step S038 is the right end initial value a of both face positions (x coordinate values) detected in step S021 (step S039). ). Here, if the CPU 24 determines that the boundary value data (right end candidate value) is the right end initial value a (step S039; Yes), the right end initial value a of the face end positions (x coordinate values) is changed. Since there was not, the process proceeds to step S032.

  On the other hand, when the CPU 24 determines in step S039 that the boundary value data (right end candidate value) is not the right end initial value a (step S039; No), both end positions (x coordinate values) of the face detected in step S021. The number N of boundary value data for counting the total number of boundary value data (right end candidate values) changed from the right end initial value a is incremented (step S0310).

  The CPU 24 repeats the above-described processing from step S032 to step S0310, and obtains boundary value data (right end candidate value) of the right end position (x coordinate value) of the face in the range where the search y coordinate value is not less than the lower limit value c and not more than d. The data is stored in the RAM 25.

  Then, when the CPU 24 determines in step S034 that the search y coordinate value is greater than or equal to d (step S034; Yes), the CPU 24 shifts the processing to the processing described in the flowchart of FIG.

  The processing described in the flowchart of FIG. 14 performs statistical processing based on the boundary value data (right end candidate value) of the right end position (x coordinate value) of the face stored in the RAM 25 by the above processing, and the right end of the face. This is a process for determining the position.

  When the number N of boundary value data added in step S0310 in FIG. 10 is small, the reliability of the processing result is low even if statistical processing is performed based on the small data. Therefore, the CPU 24 first reads the boundary value data number ratio (for example, 40%) stored in advance in the ROM 23 as shown in FIG. The CPU 24 multiplies the boundary value data number ratio by the number of pixels (d−c + 1) in which the y coordinate value in the face image is in the range of c to d. Subsequently, the CPU 24 determines whether or not the number N of boundary value data is equal to or greater than the value resulting from the multiplication (step S051). Therefore, when the CPU 24 determines that the number N of boundary value data is not equal to or greater than the result of multiplication (step S051; No), the CPU 24 sets the right end initial value a of the face end positions (x coordinate values). Without changing (step S057), the correction process 1 is terminated.

  On the other hand, if the CPU 24 determines that the number N of boundary value data is equal to or greater than the value resulting from the multiplication (step S051; Yes), the CPU 24 performs statistical processing. Specifically, the CPU 24 creates a histogram of all boundary value data (right end candidate values) stored in the RAM 25, obtains an average value and a standard deviation (step S052), and stores both values in the RAM 25. .

Subsequently, the CPU 24 refers to the average value and standard deviation of the boundary value data (right end candidate value) stored in the RAM 25 in the subsequent processing, and determines the final right end position (x coordinate value) of the face shown in FIG. ) Find the minimum boundary value (minimum boundary value data (right end candidate value)) for determining α.
An end condition for the process of obtaining the minimum boundary value is that the boundary value counter, which is an integer value, is equal to the number N of boundary value data stored in the RAM 25.
Therefore, first, the CPU 24 sets the initial value of the boundary value counter to 0, and determines whether or not the boundary value counter is smaller than the boundary value data number N (step S053).

  When the CPU 24 determines that the boundary value counter is smaller than the boundary value data number N (step S053; Yes), the boundary value data of the array A having an argument that matches the boundary value counter as shown in FIG. (Right end candidate value) is read from the RAM 25. The CPU 24 determines whether or not the read boundary value data (right end candidate value) is within i) the standard deviation and ii) smaller than the minimum boundary value (the initial value is the right end initial value a) ( Step S054).

  If the CPU 24 determines in step S054 that both conditions i) and ii) are satisfied (step S054; Yes), the boundary value data (right end candidate value) of the array A is set as the minimum boundary value, and the value Is updated and stored in the RAM 25 (step S055). As a result, the CPU 24 excludes outlier values that exceed the standard deviation from the boundary value data (right end candidate values), and obtains boundary value data (right end candidate values) having a value smaller than the minimum boundary value up to that point. Select a new minimum boundary value.

  On the other hand, if the CPU 24 determines in step S054 that either or both of the conditions i) and ii) are not satisfied (step S054; No), the CPU 24 does not update the minimum boundary value stored in the RAM 25. And move to the next process.

  Subsequently, the CPU 24 increments the boundary value counter (step S056), and returns the process to the process of step S053.

  The CPU 24 repeats the processes in steps S053 to S056 until the boundary value counter becomes equal to the number N of boundary value data stored in the RAM 25. Thereby, the CPU 24 extracts the minimum minimum boundary value from all the boundary value data (right end candidate values) of the right end position (x coordinate value) of the face stored in the RAM 25.

  If the CPU 24 determines in step S053 that the boundary value counter is equal to the boundary value data number N (step S053; No), the CPU 24 ends the process of obtaining the minimum boundary value and refers to the minimum boundary value stored in the RAM 25. . Then, the CPU 24 stores the minimum boundary value in the RAM 25 as the right end position (x coordinate value) α of the face (step S058), and ends the correction process 1. The right end position (x coordinate value) α of this face is shown as an example in the face image of FIG.

  Through the above processing, correction relating to the right end position of the face is performed, and a more accurate right end position of the face can be detected even in a face image photographed at night, a female face image, or the like.

  Next, the operation of the correction process 2 related to the left end position of the face will be described with reference to the flowchart shown in FIG. First, the CPU 24 sets the search y coordinate value to 0 and starts processing (step S041).

  Subsequently, the CPU 24 reads the left end initial value b of the both end positions (x coordinate values) of the face obtained in step S02 from the RAM 25, sets the search x coordinate value in the face image to b, and increments the search y coordinate value. The search coordinate value is set to (b, 1) (step S042).

  Further, the CPU 24 reads the lower limit value c of the vertical position (y coordinate value) of the face obtained in step S02 from the RAM 25, and determines whether or not the search y coordinate value is equal to or greater than the lower limit value c (step S043). If the search y-coordinate value is not greater than or equal to the lower limit value c (step S043; No), the process returns to step S042, and the search y-coordinate value is incremented. Thereby, in the face image, useless processing in pixels outside the vertical search range where the y coordinate value shown in FIG. 3F is in the range of the lower limit value c or more and the upper limit value d or less is omitted.

  On the other hand, if the CPU 24 determines in step S043 that the search y-coordinate value is equal to or greater than the lower limit value c (step S043; Yes), then the upper limit of the vertical position (y-coordinate value) of the face obtained in step S02. The value d is read from the RAM 25. Then, the CPU 24 determines whether or not the search y coordinate value is greater than or equal to the upper limit value d (step S044).

  Processing in the case where it is determined in step S044 that the search y coordinate value is equal to or greater than d (step S044; Yes) will be described later. If it is determined in step S044 that the search y-coordinate value is not greater than or equal to d (step S044; No), the CPU 24 determines whether or not the search x-coordinate value is equal to or less than a value obtained by adding 5 to the left end initial value b read from the RAM 25. A determination is made (step S045). This is because the range (x coordinate direction) to be corrected from the left end initial value b of the both end positions of the face obtained in step S021 is limited to 5 pixels. This limited range is stored in advance in the ROM 25 or the like as a parameter for the horizontal search range, as shown in FIG. Note that this limited range does not have to be 5 pixels and may be any value.

  In step S045, if the CPU 24 determines that the search x-coordinate value is equal to or less than the value (b + 5) obtained by adding 5 to the left end initial value b (step S045; Yes), the search coordinate value is set to (x, y). The vertical Sobel value in the pixel at the coordinates of (x + 1, y−1), (x + 1, y), (x + 1, y + 1) is referred to. The positional relationship of these coordinates is shown in FIG. Then, the CPU 24 determines whether or not all the vertical sobel values of these three coordinates are larger than 8 which is a vertical sobel threshold value stored in advance in the ROM 23 as shown in FIG. 3 (h) (step S046). . This is because, if the vertical Sobel value in the pixel of these three coordinates is larger than the vertical Sobel threshold, the position of the (x + 1, y) coordinate is more likely to be a vertical edge than the current search coordinate position. It is.

  In step S046, if the CPU 24 determines that the vertical sobel value of the referenced three coordinates is larger than the vertical sobel threshold value (step S046; Yes), the search x coordinate value is set to x + 1 (step S047), and the process of step S045 is performed. Return.

By repeating the processes in steps S045, S046, and S047 described above, the CPU 24 scans the image toward the outside of the face while determining whether or not both end positions (x coordinate values) of the face should be corrected. ing.
The determination as to whether or not the positions of both ends of the face (x coordinate value) should be corrected is based on the vertical Sobel values of the pixels located at the three coordinates (x + 1, y−1), (x + 1, y), and (x + 1, y + 1). It is not necessary to be based on and is arbitrary.

  On the other hand, if the CPU 24 determines in step S045 that the search x-coordinate value is greater than the left end initial value b plus 5 (b + 5) (step S045; No), the current search x-coordinate value is set to the left end of the face. The boundary value data (left end candidate value) of the position (x coordinate value) is stored in the RAM 25 (step S048). Here, the boundary value data (left end candidate value) is a coordinate value that is a candidate for the left end position of the face, and finally the left end position of the face is determined from these boundary value data (left end candidate value). To do.

  Even when it is determined in step S046 that the vertical sobel values of the three coordinates referred to are not larger than the vertical sobel threshold value (step S046; No), the current search x-coordinate value is set to the left edge position (x-coordinate value) of the face. ) Is stored in the RAM 25 (step S048). The search x-coordinate value in step S048 is stored in a boundary value recording array B (hereinafter referred to as array B) as shown in FIG. An array is prepared for each search y-coordinate, and the search x-coordinate value is stored as boundary value data (left end candidate value).

  The CPU 24 determines whether or not the boundary value data (left end candidate value) stored in the RAM 25 in step S048 is the left end initial value b of both face positions (x coordinate values) detected in step S021 (step S049). ). Here, when the CPU 24 determines that the boundary value data (left end candidate value) is the left end initial value b (step S049; Yes), the left end initial value b of the both end positions (x coordinate value) of the face is changed. Since there was not, the process proceeds to step S042 as it is.

  On the other hand, if the CPU 24 determines in step S049 that the left end candidate value is not the left end initial value b (step S049; No), the left end initial value b of the both end positions (x coordinate values) of the face detected in step S021. The number M of boundary value data for counting the total number of boundary value data (left end candidate values) changed from 1 is incremented (step S0410).

  The CPU 24 repeats the above-described processing from step S042 to step S0410, and in the range where the search y coordinate value is not less than the lower limit value c and not more than the upper limit value d, the boundary value data (left candidate) of the left edge position (x coordinate value) of the face Value) is stored in the RAM 25.

  When the CPU 24 determines in step S044 that the search y-coordinate value is greater than or equal to d (step S044; Yes), the process proceeds to the process described in the flowchart of FIG.

  The processing described in the flowchart of FIG. 14 performs statistical processing based on the boundary value data (left end candidate value) of the left end position (x coordinate value) of the face that has been stored in the RAM 25 by the above processing. This is a process for determining the left end position.

  When the number M of boundary value data added in step S0410 in FIG. 11 is small, the reliability of the processing result is low even if statistical processing is performed based on the small amount of data. Therefore, the CPU 24 first reads the boundary value data number ratio (for example, 40%) stored in advance in the ROM 23 as shown in FIG. The CPU 24 multiplies the boundary value data number ratio by the number of pixels (d−c + 1) in which the y coordinate value in the face image is in the range of c to d. Subsequently, the CPU 24 determines whether or not the number M of boundary value data is equal to or greater than the value resulting from the multiplication (step S051). Therefore, when the CPU 24 determines that the number M of boundary value data is not equal to or greater than the value obtained by the multiplication (step S051; No), the CPU 24 sets the left end initial value b of both end positions (x coordinate values) of the face. Without modification (step S057), the correction process 2 is terminated.

  On the other hand, when the CPU 24 determines that the number M of boundary value data is equal to or greater than the value resulting from the multiplication (step S051; Yes), the CPU 24 performs statistical processing. Specifically, the CPU 24 creates a histogram of all boundary value data (leftmost candidate values) stored in the RAM 25, obtains an average value and a standard deviation (step S052), and stores both values in the RAM 25. .

Subsequently, in the subsequent processing, the CPU 24 refers to the average value and standard deviation of the boundary value data (left end candidate value) stored in the RAM 25, and determines the final face left end position (x coordinate value) shown in FIG. ) Find the maximum boundary value (maximum boundary value data (leftmost candidate value)) for determining β.
The condition for ending the process of obtaining the maximum boundary value is that the boundary value counter, which is an integer value, is equal to the number M of boundary value data stored in the RAM 25.
Therefore, first, the CPU 24 sets the initial value of the boundary value counter to 0, and determines whether or not the boundary value counter is smaller than the boundary value data number M (step S053).

  When the CPU 24 determines that the boundary value counter is smaller than the boundary value data number M (step S053; Yes), the boundary value data of the array B having an argument that matches the boundary value counter as shown in FIG. (Left end candidate value) is read from the RAM 25. The CPU 24 determines whether or not the read boundary value data (left end candidate value) is within i) the standard deviation and ii) greater than the maximum boundary value (the initial value is the left end initial value b) ( Step S054).

  When the CPU 24 determines in step S054 that both the conditions i) and ii) are satisfied (step S054; Yes), the boundary value data (left end candidate value) of the array B is set as the maximum boundary value, and the value Is updated and stored in the RAM 25 (step S055). As a result, the CPU 24 excludes outlier values that exceed the standard deviation from the boundary value data (left end candidate values), and obtains boundary value data (left end candidate values) having a value larger than the maximum boundary value up to that point. Select a new maximum boundary value.

  On the other hand, if the CPU 24 determines in step S054 that either or both of the conditions i) and ii) are not satisfied (step S054; No), the CPU 24 does not update the maximum boundary value stored in the RAM 25. And move to the next process.

  Subsequently, the CPU 24 increments the boundary value counter (step S056), and returns the process to the process of step S053.

  The CPU 24 repeats the processes in steps S053 to S056 until the boundary value counter becomes equal to the boundary value data count M stored in the RAM 25. Thereby, the CPU 24 extracts the maximum maximum boundary value from all the boundary value data (upper point candidate values) of the left end position (x coordinate value) of the face stored in the RAM 25.

  If the CPU 24 determines in step S053 that the boundary value counter is equal to the boundary value data count M (step S053; No), the CPU 24 ends the process of obtaining the maximum boundary value and refers to the maximum boundary value stored in the RAM 25. . Then, the CPU 24 stores the maximum boundary value in the RAM 25 as the left end position (x coordinate value) β of the face (step S058), and ends the correction process 2. The left end position (x coordinate value) β of this face is shown as an example in the face image of FIG.

  Through the above processing, correction related to the left end position of the face can be performed, and a more accurate left end position of the face can be detected at night and in a female face image.

Returning to FIG. 4, after completing the correction processing step S <b> 03, the CPU 24 repeatedly executes the processing of FIG. 4 and continues to obtain both end positions (x coordinate values) of the driver's face. Further, the obtained both end positions (x coordinate values) of the face are stored in the RAM 25 and can be used for desired processing thereafter.
Through the above processing, the face edge detection device according to the present embodiment can more accurately detect the positions of both ends of a face even in a face image photographed at night, a female face image, or the like.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation and application are possible.

  For example, in the above embodiment, the Sobel filter is applied to detect the horizontal edge and the vertical edge of the face image. However, the present invention is not limited to this. For example, a pre-witt filter or a Roberts filter is applied to Edges and vertical edges may be detected.

  In the above embodiment, the vertical position (lower limit c, upper limit d) of the face is obtained from the eyebrows (or eyes) and the tone image of the mouth, but the part of the face (parts) used to determine the vertical position ) Is optional. For example, the vertical position of the face may be obtained by adding ears, cheeks, hair, or the like.

  In addition, in order to obtain the vertical position (y coordinate value) of the face, there is an eye or eyebrow image stored in advance in a ROM or the like without using the eyebrow (or eyes) and the tone image of the mouth. Then, data for specifying the assumed vertical position (y coordinate value) of the face may be used.

  The system configuration described with reference to FIGS. 1 and 2 is also an example, and can be arbitrarily changed. For example, if an infrared camera that captures an image with the near-infrared ray or the like is used for the camera 10, it is possible to acquire images of each part of the face relatively accurately without being influenced by race, skin, or hair color.

The above-described flowchart can be arbitrarily changed as long as the same function can be realized.
For example, the boundary value data number ratio may be changed to 50%, or the vertical Sobel threshold value may be changed to 10.

  In each of the above embodiments, the present invention is applied when detecting the positions of both ends of a face by photographing a driver. However, the present invention is not limited to this, and humans, animals, dolls, robots, etc. can be used in any scene. The present invention can be widely applied to processing for detecting a contour.

  Further, in the above embodiment, the average value and the standard deviation are used for the statistical processing, but the present invention is not limited to this, as long as reliable data can be selected as data used for correction of both end positions of the face. Any method may be used.

  The present invention is not limited to processing while acquiring an image with a camera, and can be used, for example, to determine the positions of both ends of a face for each of one or a plurality of face images photographed elsewhere.

  Further, a computer program for causing a computer to execute the above-described processing may be stored in the ROM via an arbitrary recording medium or a network.

It is a block diagram of a face edge detection device concerning an embodiment of the present invention. It is a block diagram which shows the structure of the computer shown in FIG. It is a figure for demonstrating the various data stored in ROM. 3 is a flowchart for explaining the operation of the face edge detection device shown in FIG. 1. It is a flowchart for demonstrating the specific example of the pre-processing of FIG. 5 is a flowchart for explaining a specific example of the face position detection process of FIG. 4. It is a flowchart for demonstrating the specific example of the face both ends detection process of FIG. It is a flowchart for demonstrating the specific example of the face vertical position detection process of FIG. This is an image obtained by filtering a female face image using a Sobel filter for detecting a vertical edge and binarizing the image. It is a flowchart for demonstrating the specific example of the correction process 1 of FIG. It is a flowchart for demonstrating the specific example of the correction process 2 of FIG. It is a figure which shows the positional relationship of the pixel of a search coordinate, and the pixel of the coordinate referred to in order to discriminate | determine whether an image is scanned toward the outer side of a face. It is a figure which shows an example of the boundary value data stored in the boundary value recording array A and the boundary value recording array B, respectively. 12 is a flowchart for explaining a specific example of processing following correction processing 1 and 2 in FIGS. 10 and 11.

Explanation of symbols

10 Camera 22 Image memory (Face image storage means)
23 ROM (contour candidate position determining means, contour position correcting means)
24 CPU (contour candidate position determining means, contour position correcting means)

Claims (5)

Face image storage means for storing a face image obtained by photographing a face;
The face image pixel data stored in the face image storage means is processed to detect a temporary contour position in the horizontal direction of the face, and based on the pixel data located around the detected temporary contour position, Contour candidate position determining means for determining a plurality of candidate contour positions that are position candidates;
A contour position correcting unit that uses a position extracted from the plurality of contour candidate positions determined by the contour candidate position determining unit as a contour position;
A face edge detection device comprising: The contour candidate position determining means processes the pixel data of the face image stored in the face image storage means, and detects a temporary contour position in the horizontal direction of the face;
The temporary contour position based on the pixel data located adjacent to the temporary contour position in the left-right direction detected by the temporary contour position detection means and the two pixel data located above and below the neighboring pixel. Determining means for determining whether or not to change to the position of the pixel located next to,
A changing means for changing the temporary contour position when the determining means determines to change the temporary contour position;
Means for repeatedly operating the determining means and the changing means, and setting the temporary contour position when the determining means determines not to change the temporary contour position as the contour candidate position;
The face edge detection device according to claim 1, further comprising: The contour position correcting means creates a histogram of the plurality of contour candidate positions determined by the contour candidate position determining means,
A standard deviation calculating means for calculating a standard deviation of the candidate contour position based on the histogram created by the histogram creating means;
Means for setting the contour candidate position located on the outermost side of the face image out of the contour candidate positions within the standard deviation calculated by the standard deviation calculating means;
The face edge detection device according to claim 1, further comprising: A face image storing step for storing a face image obtained by photographing a face;
The face image pixel data stored in the face image storage step is processed to detect a temporary contour position in the horizontal direction of the face, and based on the pixel data located around the detected temporary contour position, A contour candidate position determining step for determining a plurality of contour candidate positions that are position candidates;
A contour position correcting step in which a position extracted from the plurality of contour candidate positions determined in the contour candidate position determining step is a contour position;
A method for detecting a face edge. On the computer,
A face image storing procedure for storing a face image obtained by photographing a face;
The face image pixel data stored in the face image storage procedure is processed to detect a temporary contour position in the left-right direction of the face, and based on the pixel data located around the detected temporary contour position, A contour candidate position determination procedure for determining a plurality of contour candidate positions to be position candidates;
A contour position correction procedure in which a position extracted from the plurality of contour candidate positions determined in the contour candidate position determination procedure is a contour position;
A program that executes

Images