JP3629164B2 - Face image processing device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a face image processing apparatus that discriminates an eye from another image by using a feature in which an eye has a thin line portion from an eye candidate image.
[0002]
[Prior art]
Conventional image processing apparatuses for processing a driver's face image are disclosed in, for example, Japanese Patent Application Laid-Open No. 8-175218 (the eye region is determined by matching with a template including an eyebrow region and an eye region), and Japanese Patent Application Laid-Open No. 8-300978. It is described in.
Hereinafter, JP-A-8-300978 will be described in detail.
FIG. 14 is a diagram showing an overall configuration of a conventional face image processing apparatus disclosed in Japanese Patent Application Laid-Open No. 8-300978.
In the figure, 1 is a camera for imaging a driver, 2 is near-infrared illumination used for imaging, 3 is illumination control means for controlling the near-infrared illumination 2, and 4 is a face image for processing an image captured by the camera 1. The processing apparatus comprises an image input means 5, a density level conversion means 6, a floating binarization means 7, an eye presence area setting means 8, a candidate area setting means 9, and an eye area determination means 10. 11 is a blink detection unit that detects blinking using the result of image processing performed by the face image processing device 4, 12 is a doze determination unit that determines a doze state from the image processing result, and 13 is an alarm unit.
[0003]
FIG. 15 is a flowchart showing the flow of the operation of the eye area determination unit of the conventional face image processing apparatus.
In the figure, 14 is an eye evaluation function calculating means, and 15 is an eye area selecting means.
In Example 1 of Japanese Patent Laid-Open No. 8-300978, the eye area determination means 10 reads the face lateral width EAW of the eye candidate area, and only when the EAW is within a predetermined range set as an eye area. After reading the X-axis histogram SUMX of the region and calculating the maximum value SUMXMAX of the SUMX, the evaluation function EFV1 = Σy (SUMXMAX−SUMX) / ΣySUMX and the evaluation function EFV2 = SUMXMAX−C (C: constant) are calculated, An evaluation function value EFV = EFV1 × EFV2 (or C1 × EFV1 + C2 × EFV2 (weight C1> weight C2 is better)) is determined as an eye region that is larger than a predetermined value EFVMIN. The one having the smallest distance between the representative point and the face center of gravity is finally determined as the eye area.
In Example 3, when the center of gravity position MC, the peak value PH, and the half-value width HW of the frequency distribution curve of the X-axis histogram are used, a single peak is shown in the eyebrow region and a twin peak is shown in the eye region. An eye candidate region having a maximum HW / PH and a peak on the left and right of the MC is determined as the eye region.
In the fourth embodiment, the eye area is determined using the temporal change in the eye evaluation function value in the first embodiment.
In the fifth embodiment, the deviation DX between the X coordinate XEC of the left and right eye candidate area representative point and the X coordinate XFC of the face center of gravity, and the deviation DY of the Y coordinate YEC of the left and right eye candidate area representative point and the Y coordinate YFC of the face center of gravity are both predetermined. A method is described in which a combination of a minimum value and a minimum value is determined as an eye area.
[0004]
[Problems to be solved by the invention]
However, it takes a long time to process all images with multi-valued data. Even in binarization, the state of the face (brows-eye relationship, etc.) depends on the presence of glasses and the attached information such as weather and hair. ) Changes and is unstable. However, in the above-described conventional device, the eye area is determined based on the eye evaluation function value using the X / Y-axis histogram created based on the binary data and the positional relationship between the eye candidate area and the face center of gravity or the left and right eye candidate areas. Therefore, it is easy to cause a false detection.
[0005]
The present invention has been made to solve the above-described problems. A face image processing apparatus configured to discriminate eyes by calculating a ratio of a thin line portion that hardly exists in an eyebrow or a spectacle frame. The purpose is to get.
[0006]
[Means for Solving the Problems]
In the face image processing apparatus according to the present invention, an image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, and a face stored in the image memory. Candidate extraction means for binarizing an image and extracting eye candidates, eye search area setting means for estimating an eye search area indicating a range in which an eye exists in a face image binarized by the candidate extraction means, Using the face image stored in the image memory for the eye candidates included in the eye search area estimated by the eye search area setting means Scans the face in the vertical direction sequentially in the horizontal direction of the face, Extract the thin line part, The ratio of the number of scanning lines from which the thin line portion is extracted to the number of scanning lines in the horizontal direction of the face of the image of the eye candidate is calculated, and eye candidates whose ratio is a predetermined value or more are determined as eyes. It is provided with an eye determination means.
The eye determination means extracts a thin line portion by comparing the luminance of a specific pixel in the face image with two pixels located at a predetermined length in the vertical direction of the face from the specific pixel.
[0007]
Also, An image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, binarizes the face image stored in this image memory, and extracts eye candidates Candidate extracting means for performing eye search area setting means for estimating an eye search area indicating a range where the eyes exist in the face image binarized by the candidate extracting means, and eye search estimated by the eye search area setting means For the eye candidates included in the region, an eye determination unit that extracts a thin line portion using a face image stored in the image memory and discriminates the eye based on a ratio of the thin line portion to the entire image of the eye candidate, The eye determination means compares the luminance of a specific pixel in the face image with two pixels located at a predetermined length in the vertical direction of the face from the specific pixel and a pixel located at a predetermined length in the horizontal direction of the face. To extract the thin line portion.
Further, the pixel at a position of a predetermined length in the horizontal direction is directed from the specific pixel toward the outside of the face.
[0008]
In addition, when there is a specific pixel outside the center of gravity of the eye candidate, the eye determination unit compares the luminance with a pixel at a predetermined length in the lateral direction toward the outside of the face, and When there is a specific pixel on the inner side from the center of gravity, the luminance is compared with a pixel located at a predetermined length in the horizontal direction toward the inner side of the face.
The eye determining means calculates a luminance difference value between a specific pixel and a pixel adjacent to the specific pixel in the vertical direction, and extracts a thin line portion based on a length from the peak to the peak of the calculated difference value. It is.
[0009]
Furthermore, the eye search region setting means sets a nostril region and estimates the eye search region based on the set nostril region.
The candidate extracting means extracts eye candidates from the face image through a MAX / MIN filter having a length corresponding to the width of the eyebrows and eyes.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Embodiment 1 FIG.
Embodiment 1 of the present invention will be described below with reference to the drawings.
FIG. 1 is a diagram showing a configuration of a face image processing apparatus according to Embodiment 1 of the present invention.
In the figure, 17 is a CCD camera that captures a face image, 18 is an image that is output from the CCD camera 17 and stores multi-value data of a face image that is configured to be displayed by a large number of pixels arranged in a matrix. A memory 19 is a CPU that performs image processing based on data in the image memory 18.
FIG. 2 is a flowchart showing an outline of processing in the CPU in FIG.
In the figure, 20 is a face image input means, 21 is a candidate extraction means, 22 is an eye search area setting means, and 23 is an eye determination means. The face image input means 20 inputs face image data from the image memory 18 to the CPU 19. In the input image, the candidate extraction means 21 performs binarization to extract eye candidates (binary data is stored in a separate memory so that input image data can be referred to later), and an eye search area The setting means 22 estimates an eye search area indicating a range where eyes are present. Further, the eyes are determined by the eye determination means 23 from the eye candidates in the set eye search area. Details of each means will be described later.
[0011]
FIG. 3 is a diagram showing an input image from the camera in the face image processing apparatus according to Embodiment 1 of the present invention, which is input to the CPU 19 by the face image input means 20.
In the figure, reference numeral 25 denotes a face image input from the CCD camera 17.
FIG. 4 is a diagram showing a binarized image binarized by the candidate extracting means 21 of the face image processing apparatus according to Embodiment 1 of the present invention.
In the figure, 26 indicates an image obtained by binarizing the face image 25, and 27 indicates a processing direction.
FIG. 5 is an explanatory diagram of the MAX / MIN filter used in the candidate extraction means 21 of the face image processing apparatus according to Embodiment 1 of the present invention, and FIG. 5 (a) is an image before and after applying the MAX filter and the MIN filter. FIG. 5B is a diagram showing a signal, and FIG. 5B schematically shows luminance values corresponding to FIG. FIG. 5C shows the image signal after the MAX / MIN filter processing, and FIG. 5D schematically shows the luminance value corresponding to FIG. 5C.
In the figure, 28 is an image signal of a part of the input image, 29 is an image signal after applying the MAX filter, 30 is an image signal after applying the MIN filter, 31 is a luminance value corresponding to the image signal 28, and 32 is A luminance value corresponding to the image signal 29 and 33 is a luminance value corresponding to the image signal 30. Here, one frame surrounding each luminance value represents one pixel, and the luminance value is a value from 0 to 20 for simplicity.
[0012]
FIG. 6 is an explanatory diagram of the candidate extraction means 21 of the face image processing apparatus according to Embodiment 1 of the present invention, and FIG. 6 (a) is a diagram showing an image signal after the MAX / MIN filter is applied to the input image. FIG. 6B is a diagram showing a signal after binarization.
In the figure, 35 is an image signal after applying the MAX / MIN filter, and 36 is a threshold value for binarization.
FIG. 7 is a flowchart showing the flow of the eye search area setting means of the face image processing apparatus according to Embodiment 1 of the present invention.
In the figure, 37 is a nostril area setting means, 38 is a nostril search means, 39 is a nostril determination means, and 40 is an eye area estimation means. Details will be described later.
FIG. 8 is a diagram showing part of the nostril region setting means, nostril search means, and nostril determination means of the face image processing apparatus according to Embodiment 1 of the present invention, and FIG. FIG. 8B is a diagram showing a search area, and FIG. 8B is a diagram showing a binarized image and histograms in the X-axis direction and the Y-axis direction.
[0013]
FIG. 9 is a diagram showing the eye search area set by the eye search area setting means 22 of the face image processing apparatus according to Embodiment 1 of the present invention.
FIG. 10 is a diagram showing a fine line portion of the face image for explaining the eye determination means 23 of the face image processing apparatus according to Embodiment 1 of the present invention. FIG. 10 (a) shows candidate candidates based on the input image. FIG. 10B is a diagram for explaining the detection of the thin line portion, and FIG.
[0014]
The details of each means shown in FIG. 2 will be described below with reference to FIG.
Face image data is input to the CPU 19 from the image memory 18 by the face image input means 20.
Next, binarization is performed by the candidate extraction means 21 to extract eye candidates.
The candidate extraction means 21 uses a MAX / MIN filter to extract eye candidates. First, a MAX filter is applied to the input image. The MAX filter has a predetermined length (the number of pixels is a. Hereinafter referred to as filter size), and converts the luminance value of the pixel at the center of the filter into the maximum luminance value in the filter. . For example, if a is 5 and the MAX filter is applied to the thick line portion of the luminance model 31 in FIG. 5B, the maximum luminance is 9 here, so the luminance value of the pixel 31a in the center of the filter is converted from 7 to 9. The When the pixel is shifted to the right by one pixel and the MAX filter is applied in the same manner, the luminance model 31 in FIG. 5B is converted into a luminance model 32.
[0015]
Next, the MIN filter is applied to the image after the MAX filter processing. The MIN filter also has the same filter size as the MAX filter, and converts the luminance value of the pixel at the center of the filter to the minimum luminance value in the filter. When the luminance pattern 32 in FIG. 5B after the MAX filter processing is shifted to the right by one pixel and the MIN filter is applied, the luminance pattern 33 in FIG. 5B is converted. Then, when the input image is drawn from the image after the MAN / MIX filter processing, a low luminance area having a width equal to or smaller than the filter size is extracted. In the figure, FIG. 5D obtained by subtracting the luminance model 31 from the luminance model 33 of FIG. 5B is obtained, and when converted to an image signal, FIG. 5C is obtained. However, since both the MAX filter and the MIN filter start conversion from the center of the first filter, and the conversion ends at the center of the last filter, the data after the MAX filter processing is shown in FIGS. The data after the MIN filter processing is shown in FIG. 5B 33a to 33b, and the image after the MAX / MIN filter processing is eventually smaller by a-1 pixels than the processing start position and the processing end position.
[0016]
In the first embodiment, processing is performed line by line along the vertical direction of the face (processing direction: 27 in FIG. 4). The image signal 35 in FIG. 6A is an image signal after the input image is subjected to a MAX / MIN filter having a length corresponding to the width of the eyebrows and the eyes. Here, the nose, mouth, etc., which are narrower than the eyebrows and eyes, are also extracted. Since the feature points of the face image such as the eyebrows and the eyes have a large luminance difference with the surroundings, the luminance value after the difference is large (35a), but the shadow and clothes have a small luminance difference with the surroundings (35b). Therefore, a threshold value 36 is set as shown in FIG. 6A, and binarization is performed based on this. FIG. 6B shows a signal after binarization. FIG. 4 shows an image obtained by the above processing.
[0017]
In the binarized image extracted by the candidate extraction means 21, the eye search area setting means 22 for estimating the eye search area indicating the range where the eye candidates exist is, for example, as shown in the flowchart of FIG. 8 includes a means 37, a nostril search means 38, a nostril determination means 39, and an eye area estimation means 40. The nostril search area 5n in the nostril search mode and the nostril tracking area 5nn in the nostril tracking mode. Set. The nostril search mode refers to a state in which nostril position learning has not yet been performed (a state in which eyes have not yet been found), and the nostril tracking mode refers to a state in which nostril learning has been completed. The search area 5n is set to be limited to a range where the face enters with a normal driving posture on the screen. The tracking region 5nn sets a region having a predetermined width from the nostril learning position (or the previous nostril position) 5gn. The nostril search means 38 searches an area where two equal circles are arranged in the lateral direction of the face as shown in FIG. The search is performed using, for example, an X-axis direction histogram and a Y-axis direction histogram as shown in FIG. The nostril determination means 39 selects the most likely nostril from the retrieved candidates and calculates its position. After the nostril is determined, the eye area estimation means 40 sets 5r5l in FIG. 9 which is an area where the left and right eyes are supposed to exist based on the nostril position. The size of 5r5l is sufficient for the eyes to enter, and the eye determination means 23 in FIG.
The nostril position determined by the nostril determination means 39 is stored in the memory and used for setting the nostril region in the nostril tracking mode.
[0018]
Next, the eye determination means 23 determines the eye candidates and eye positions in the region extracted above. For the eye candidates, element blocks existing in the eye search area estimated by the eye search area setting unit 22 are extracted using, for example, an X-axis histogram and a Y-axis histogram. Next, in order to discriminate the eyebrows and eyes having similar shapes with respect to the extracted element block, as shown in FIG. 10A, the fine line of each candidate based on the input image data (multi-value) Detect part. The detection is performed by sequentially scanning each vertical line of the face, and a line having a point (41) passing through filter points (41a, 41b) composed of at least two points as shown in FIG. Count as. Passing is a condition that the luminance value of the pixel of interest (41) is smaller than the luminance value of the filter points (41a, 41b) by a predetermined value (FIG. 10 (b)).
[0019]
The fine line ratio is represented by the number of fine line lines with respect to the number of lines in the face horizontal direction of each candidate. Since it is estimated that the eyes have the most thin line portions (the wrinkle lines), those having a fine line ratio equal to or greater than a predetermined value and taking the maximum value are determined as eyes. The filter point 41a or 41b has a fixed OFST from the point of interest, or the brightest point of a pixel existing within a predetermined range from the point of interest, an average value of pixel luminance values existing within a predetermined range from the point of interest, or the like Can be considered.
(In FIG.
Eyebrow: Fine line ratio = (Number of fine line lines 19 x 100) / face side Number of direction lines 41 = 46%
Eye: Fine line ratio = (Number of fine line lines 26 × 100) / face side Number of direction lines 36 = 72%)
At this time, thin eyeglass frames can be cut at the time of eye candidate extraction, so there is no problem, and thick objects and eyebrows etc. have many parts that are wide in the vertical direction of the face and it is difficult to pass through the filter, so it is difficult to enter candidates Seem. Therefore, eyes can be extracted with high accuracy.
Further, if there is a fine line condition passing point, the processing of the line is completed there, so that the processing can be performed at a relatively high speed.
[0020]
Embodiment 2. FIG.
FIG. 11 is a diagram showing a fine line portion of a face image of the face image processing apparatus according to Embodiment 2 of the present invention.
In the second embodiment, the eye determination means 23 of the first embodiment is changed as follows.
For the eye candidates, element blocks existing in the eye search area estimated by the eye search area setting unit 22 are extracted using, for example, an X-axis direction histogram and a Y-axis direction histogram. Next, in order to discriminate eyebrows and eyes having similar shapes with respect to the extracted element block, as shown in FIG. 11A, each candidate is based on the input image data (multivalue). The thin line part of is detected. The detection is performed by sequentially scanning the vertical lines of each face, and a line having a point (42) passing through filter points (42a, 42b, 42c) composed of at least three points as shown in FIG. 11 (a). The line is counted as a fine line (however, the left eye is shown in FIG. 11A, and the filter point 42c needs to be contrasted with the vertical line of the face for the right eye). The passage condition is that the luminance value of the pixel of interest (42) is smaller than the luminance values of the filter points (42a, 42b, 42c) by a predetermined value (FIG. 11 (b)). The fine line ratio is represented by the number of fine line lines with respect to the number of lines in the face horizontal direction of each candidate. Since the eye has the most thin line portions (the wrinkle line) and the curvature of the eye corner line is high, it is estimated that the condition of the filter point 42c is easily satisfied. Judge as.
[0021]
The filter points 42a, 42b, and 42c are points having a fixed OFST from the point of interest, or the brightest point of a pixel existing within a predetermined range from the point of interest, or a pixel luminance value existing within a predetermined range from the point of interest. Average values are possible.
(In FIG.
Eyebrow: Fine line ratio = (Number of fine line lines 15 x 100) / face side Number of direction lines 41 = 36%
Eye: Fine line ratio = (Number of fine line lines 21 x 100) / face side Number of direction lines 36 = 58%)
Here, thin eyeglass frames can be cut when extracting eye candidates, so there is not much problem, and thick objects and eyebrows have many parts that are wide in the vertical direction of the face, and it is difficult to pass through the filter. Seem. In addition, since the face horizontal direction line (42c) is also seen, it is considered that an eyebrow having a smaller curvature than the eyes is difficult to pass through the filter and to enter a candidate. Therefore, eyes can be extracted with high accuracy. Further, if there is a thin line condition passing point, the processing of the line is completed at that point, so that the processing can be performed at a relatively high speed.
[0022]
Embodiment 3 FIG.
FIG. 12 is a diagram showing a fine line portion of a face image of the face image processing apparatus according to Embodiment 3 of the present invention.
In the third embodiment, the eye determination means 23 of the first embodiment is changed as follows.
As the eye candidates, element blocks existing in the eye search area determined by the eye search area setting unit 22 are extracted using, for example, an X-axis direction histogram and a Y-axis direction histogram. Next, in order to discriminate eyebrows and eyes having similar shapes with respect to the extracted element blocks, as shown in FIG. 12A, the thin line portions of each candidate based on the input image data (multivalue) Is detected. The detection is performed by sequentially scanning each face longitudinal direction line and passing through filter points 43, 43a, 43b, 43c or 44, 44a, 44b, 44c composed of at least three points as shown in the figure (43, 44) is counted as a thin line. However, the filter point 43 is used for the left lateral half of the face from the center of gravity position of the eye candidate, 44 is used for the right half of the face, and the luminance value of the pixel of interest (43 or 44) is passed through the filter point (43a, 43b, 43c or 44a, 44b, 44c) is a condition that the predetermined value is smaller than the luminance value (FIG. 12B). The fine line ratio is represented by the number of fine line lines with respect to the number of lines in the face horizontal direction of each candidate. Since the eye has the most thin line parts (the wrinkle line) and the curvature of the eye corner line is high, it is presumed that the filter point condition is easily satisfied. judge.
[0023]
Filter points 43a, 43b, 43c or 44a, 44b, 44c are points having a fixed OFST from the point of interest, or the brightest point of a pixel existing within a predetermined range from the point of interest, or within a predetermined range from the point of interest. An average value of the existing pixel luminance values can be considered.
(In FIG.
Eyebrow: Fine line ratio = (Number of fine line lines 19 x 100) / face side Direction line 41 = 46%
Eye: Fine line ratio = (Number of fine line lines 26 × 100) / face side Number of direction lines 36 = 72%)
Here, thin eyeglass frames can be cut when eye candidates are extracted, so there is no problem. Thick objects, eyebrows, etc. have many parts that are wide in the vertical direction of the face, and do not easily pass through filters, making it difficult to enter candidates. I think that the.
Further, since the face horizontal direction line (43c or 44c) is also seen, it is considered that an eyebrow having a smaller curvature than the eyes is difficult to pass through the filter and to enter a candidate. Therefore, eyes can be extracted with high accuracy.
Further, if there is a fine line condition passing point, the processing of the line is completed there, so that the processing can be performed at a relatively high speed.
[0024]
Embodiment 4 FIG.
FIG. 13 is a diagram showing a fine line portion of a face image of the face image processing apparatus according to Embodiment 4 of the present invention.
In the fourth embodiment, the eye determination means 23 of the first embodiment is changed as follows.
For the eye candidates, element blocks existing in the eye search area estimated by the eye search area setting unit 22 are extracted using, for example, an X-axis direction histogram and a Y-axis direction histogram. Next, in order to discriminate eyebrows and eyes having similar shapes with respect to the extracted element block, as shown in FIG. 13A, the fine line of each candidate is based on the input image data (multivalue). Detect part. The detection is performed by sequentially scanning the vertical lines of each face, calculating the difference in luminance between the target pixel and the previous pixel (= differential value, FIG. 13B), and the width from the peak 45a to the peak 45b of the differential value is equal to or less than a predetermined value. Are counted as thin line.
[0025]
The fine line ratio is represented by the number of fine line lines with respect to the number of lines in the face horizontal direction of each candidate. Since it is estimated that the eyes have the most thin line portions (the wrinkle lines), those having a fine line ratio equal to or greater than a predetermined value and taking the maximum value are determined as eyes.
(In FIG.
Eyebrow: fine line ratio = (number of fine line lines 16 x 100) / face side Number of direction lines 41 = 39%
Eye: fine line ratio = (number of fine line lines 24 × 100) / face side Number of direction lines 36 = 66%)
Here, a thin eyeglass frame can be cut at the time of eye candidate extraction, so there is not much problem, and a thick object, eyebrows, etc. have many parts with a wide width in the vertical direction of the face, and it seems difficult to enter candidates. Therefore, eyes can be extracted with high accuracy.
Further, if there is a fine line condition passing point, the processing of the line is completed there, so that the processing can be performed at a relatively high speed.
[0026]
【The invention's effect】
Since the present invention is configured as described above, the following effects can be obtained.
An image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, binarizes the face image stored in this image memory, and extracts eye candidates Candidate extracting means for performing eye search area setting means for estimating an eye search area indicating a range where the eyes exist in the face image binarized by the candidate extracting means, and eye search estimated by the eye search area setting means Using eye images stored in the image memory for eye candidates included in the area Scans the face in the vertical direction sequentially in the horizontal direction of the face, Extract the thin line part, The ratio of the number of scanning lines from which the thin line portion is extracted to the number of scanning lines in the horizontal direction of the face of the image of the eye candidate is calculated, and eye candidates whose ratio is a predetermined value or more are determined as eyes. Since the eye determination means is provided, the extraction of the thin line portion is performed with the multi-valued face image input to the image memory. And determine the eye by the ratio of the number of scanning lines extracted in the thin line portion As a result, the eyes can be accurately discriminated and the eye search area is estimated using the binarized image, so that high-speed processing can be performed.
Further, the eye determination means extracts the fine line portion by comparing the luminance between the specific pixel in the face image and two pixels located at a predetermined length in the vertical direction of the face from the specific pixel. A thin line portion can be extracted.
[0027]
Also, An image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, binarizes the face image stored in this image memory, and extracts eye candidates Candidate extracting means for performing eye search area setting means for estimating an eye search area indicating a range where the eyes exist in the face image binarized by the candidate extracting means, and eye search estimated by the eye search area setting means For the eye candidates included in the region, an eye determination unit that extracts a thin line portion using a face image stored in the image memory and discriminates the eye based on a ratio of the thin line portion to the entire image of the eye candidate, The eye determination means compares the luminance of a specific pixel in the face image with two pixels located at a predetermined length in the vertical direction of the face from the specific pixel and a pixel located at a predetermined length in the horizontal direction of the face. To extract the thin line part, The eye determination means can accurately determine the eye by performing the extraction of the thin line portion with the multi-valued face image input to the image memory, and also using the pixels in the lateral direction of the face, The thin line portion can be extracted more accurately.
Furthermore, since the pixel at the position of the predetermined length in the horizontal direction is outside the face from the specific pixel, the certainty of the thin line portion extraction is increased.
[0028]
In addition, when there is a specific pixel outside the center of gravity of the eye candidate, the eye determination unit compares the luminance with a pixel at a predetermined length in the lateral direction toward the outside of the face, and When there is a specific pixel on the inner side from the center of gravity, since the luminance is compared with a pixel at a position of a predetermined length in the horizontal direction toward the inner side of the face, the certainty of thin line portion extraction is increased.
Further, the eye determination means calculates the difference value of the luminance between the specific pixel and the pixel adjacent to the specific pixel in the vertical direction, and extracts the thin line portion by the length from the peak to the peak of the calculated difference value. The thin line portion can be extracted with high accuracy.
[0029]
Furthermore, the eye search area setting means sets the nostril area and estimates the eye search area based on the set nostril area, so that the eye search area can be reliably estimated.
In addition, since the candidate extraction unit extracts the eye candidates through the MAX / MIN filter having a length corresponding to the width of the eyebrows and the eyes, the candidate extraction unit can efficiently extract the eye candidates.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a face image processing apparatus according to Embodiment 1 of the present invention.
FIG. 2 is a flowchart showing a process flow of the face image processing apparatus according to the first embodiment of the present invention.
FIG. 3 is a diagram showing an input image from a camera in the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 4 is a diagram showing a binarized image of the face image processing apparatus according to Embodiment 1 of the present invention.
FIG. 5 is an explanatory diagram of a MAX / MIN filter of the face image processing apparatus according to Embodiment 1 of the present invention.
FIG. 6 is an explanatory diagram of candidate extraction means of the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 7 is a flowchart showing a flow of eye search area setting means of the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 8 is a diagram showing a part of nostril region setting means, nostril search means, and nostril determination means of the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 9 is a diagram showing an eye search area of the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 10 is a diagram showing a fine line portion of a face image of the face image processing apparatus according to Embodiment 1 of the present invention;
FIG. 11 is a diagram showing a fine line portion of a face image of a face image processing device according to Embodiment 2 of the present invention;
FIG. 12 is a diagram showing a fine line portion of a face image of a face image processing apparatus according to Embodiment 3 of the present invention.
FIG. 13 is a diagram showing a fine line portion of a face image of a face image processing apparatus according to Embodiment 4 of the present invention.
FIG. 14 is a diagram showing the overall configuration of a conventional face image processing apparatus.
FIG. 15 is a flowchart showing a flow of operation of an eye area determination unit of a conventional face image processing apparatus.
[Explanation of symbols]
17 CCD camera, 18 image memory, 19 CPU,
20 face image input means, 21 candidate extraction means,
22 eye search area setting means, 23 eye determination means,
37 nostril region setting means, 38 nostril search means,
39 Nostril determination means, 40 eye area estimation means.

Claims (8)

An image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, binarizes the face image stored in the image memory, and extracts eye candidates Candidate extraction means for performing eye search, an eye search area setting means for estimating an eye search area indicating a range in which an eye exists in the face image binarized by the candidate extraction means, and eye search estimated by the eye search area setting means for eye candidates included in the region, using the face image stored in the image memory is performed sequentially in the vertical direction of the scan of the face in the lateral direction of the face, to extract the fine line portion for each of the scan, the fine line portion calculate the ratio but the lateral direction of the scanning line number of the face images of the first candidate for the scan line number that is extracted, comprise an eye determination means that the ratio is determined to eye eye candidates equal to or more than a predetermined value Face image processing apparatus characterized by. The eye determination means extracts a thin line portion by comparing the luminance of a specific pixel in the face image with two pixels located at a predetermined length in the vertical direction of the face from the specific pixel. Item 2. The face image processing apparatus according to Item 1. An image memory that stores a face image that is captured by a camera and displayed by a large number of pixels arranged in a matrix, binarizes the face image stored in the image memory, and extracts eye candidates Candidate extraction means for performing eye search, an eye search area setting means for estimating an eye search area indicating a range in which an eye exists in the face image binarized by the candidate extraction means, and eye search estimated by the eye search area setting means For the eye candidates included in the region, an eye determination unit that extracts a thin line portion using a face image stored in the image memory and discriminates an eye based on a ratio of the thin line portion to the entire image of the eye candidate, It said second determining means includes a specific pixel in the face image, and the pixels from the specific pixel to 2 pixels and lateral position of the predetermined length of the face in the longitudinal direction of the predetermined length of the position of the face Face image processing apparatus you and extracting the thin line portion by performing a comparison of brightness. 4. The face image processing apparatus according to claim 3, wherein the pixel located at a predetermined length in the horizontal direction is directed from the specific pixel toward the outside of the face. When there is a specific pixel outside the center of gravity of the eye candidate, the eye determination means compares the luminance with the pixel located at a predetermined length in the lateral direction toward the outside of the face, and the center of gravity of the eye candidate. 4. The face image processing apparatus according to claim 3, wherein when there is a specific pixel on the inner side of the face, luminance comparison is performed with a pixel at a position of a predetermined length in the horizontal direction toward the inner side of the face. The eye determination means calculates a luminance difference value between the specific pixel and a pixel adjacent to the specific pixel in the vertical direction, and extracts a thin line portion based on a length from the peak to the peak of the calculated difference value. The face image processing apparatus according to claim 1. The face image processing device according to any one of claims 1 to 6, wherein the eye search region setting means sets a nostril region and estimates the eye search region based on the set nostril region. . 8. The face according to claim 1, wherein the candidate extraction unit extracts the eye candidate from the face image through a MAX / MIN filter having a length corresponding to the width of the eyebrows and the eyes. Image processing device.

Images