JP2004206665A - Human face detecting method and human face detecting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To improve rapidity and accuracy of human face detection executed by a computer. <P>SOLUTION: A computer 12 acquires an image from a video camera 11. The computer 12 determines whether a part of a photographed image area satisfies a set flesh-color area standard. When it is determined that a part of the image area satisfies the flesh-color area standard, the computer 12 sets a search area which includes the part of the image area. When it is determined that a part of the image area does not satisfy the flesh-color area standard, or that a face area does not exist within the search area, the computer 12 determines whether a dynamic area is included in the image. When it is determined that the dynamic area is included in the image, the computer 12 sets an auxiliary search area which includes the dynamic area. The computer 12 uses pattern matching to determine whether the face area exists within the set search area or the auxiliary search area. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a face detection method and a face detection device.
[0002]
[Prior art]
In surveillance work and marketing research using video information, it is important to detect a person in an image. For human detection, it is conceivable to use a face detection method using a color (skin color) as disclosed in Non-Patent Document 1. In the face detection method using the skin color, an area detected by the skin color is set as a face area. Therefore, if the range of the reference skin color can be set appropriately, high-speed face detection is possible.
[0003]
It is also conceivable to use a face detection method using pattern matching as disclosed in Patent Document 1 for human detection. In pattern matching, face detection is performed by comparing a previously registered face pattern with an acquired image.
[0004]
[Non-patent document 1]
Hitoshi Hongo, Kazuhiko Yamamoto, "Extraction of Facial Region and Facial Parts by Estimating Skin Color in Moving Region", Journal of the Institute of Image Information and Television Engineers, Institute of Image Information and Television Engineers, December 1998, Vol. 52, No. 12, p. 86-93
[Patent Document 1]
JP-A-7-141506
[0005]
[Problems to be solved by the invention]
In color-based face detection, a range of reference skin color must be determined in order to distinguish skin from other parts. However, the skin color in an image greatly changes according to camera parameters such as camera exposure and aperture, and the environment (brightness and darkness). Therefore, it is difficult to fixedly set the reference skin color range, and it is necessary to set an appropriate face detection parameter in accordance with a change in environment. In addition, even when an appropriate face detection parameter is to be reset, there is a problem that the detection accuracy is rather lowered if resetting is performed based on the erroneously detected skin color.
[0006]
In addition, in order to detect a face, it is necessary to exclude an object having a color similar to the skin color other than the face or to distinguish between a hand and a face. However, there is a problem that it is difficult to remove an object or to distinguish between a hand and a face.
[0007]
On the other hand, in face detection using pattern matching, judgment is made based on the shape, so it is not easily affected by changes in the environment or changes in camera parameters, so that things similar to skin color other than the face can be excluded or distinguished from hands etc. It is relatively easy to do. However, in consideration of the possibility that face images in an image are present at various sizes and at various positions, regions of various sizes are prepared for scanning the image, and the prepared various sizes are prepared. It is necessary to scan the entire image using each of the image areas. Therefore, in face detection using pattern matching, there is a problem that it takes a long time to calculate and it is difficult to speed up face detection.
[0008]
An object of the present invention is to provide a face detection method and apparatus that improve both the speed and accuracy of face detection in order to solve the above problems.
[0009]
[Means for Solving the Problems]
In order to achieve the above object, the invention according to claim 1 is a face detection method executed by a computer, wherein the computer sets an imaging step for imaging a face and a part of an area of the imaged image. A first determination step of performing one of a satisfaction determination that the determined skin color region criterion is satisfied, and a non-satisfaction determination that a part of the region does not satisfy the skin color region criterion; When a satisfaction determination is made, a search area setting step of setting a search area including a part of the area of the image satisfying the skin color area criterion, and determining whether a face area exists in the set search area. A second determination step of determining using pattern matching, and when the unsatisfied determination is made, or when it is determined that no face area exists in the search area, a dynamic area is included in the image. An auxiliary determination step of determining whether or not the image is included, and an auxiliary search area setting for setting an auxiliary search area including the dynamic area when it is determined that the image includes a dynamic area. And a third determining step of determining whether or not a face area exists in the set auxiliary search area by using pattern matching.
[0010]
According to a second aspect of the present invention, in the first aspect, when it is determined that a face area exists in the set search area or in the set auxiliary search area, based on the color data in the face area. And resetting the skin color region reference.
[0011]
3. The face detection method according to claim 2, wherein in the resetting step, the resetting of the skin color area reference differs according to a predetermined condition.
[0012]
According to a fourth aspect of the present invention, in addition to the configuration of the face detecting method according to the third aspect, in the resetting step, it is determined whether it is the search area or the auxiliary search area for which it is determined that a face area exists. Based on the determination in the detection area determination step, the resetting of the skin color area reference is different depending on whether the face area is determined to be present in the search area or when the face area is determined to be present in the auxiliary search area. It is characterized by.
[0013]
According to a fifth aspect of the present invention, in addition to the configuration of the face detecting method according to the fourth aspect, the resetting step includes the step of determining whether a face area exists in the auxiliary search area. Based on the determination of the dynamic region ratio determining step of determining whether the ratio is equal to or greater than the predetermined threshold or less than the predetermined threshold, the detected dynamic region ratio is equal to or greater than the predetermined threshold, and It is characterized in that the resetting of the flesh color area reference is different in each case.
[0014]
According to a sixth aspect of the present invention, in the resetting step, in addition to the configuration of the face detecting method according to the fifth aspect, when it is determined that a face area exists in the auxiliary search area, When the ratio is less than the predetermined threshold, the past history region is determined by the past history determination step of determining the ratio of the past face region detected in the search region. , The resetting of the skin color region reference is different between the case where the skin color region reference is less than the predetermined threshold value.
[0015]
8. The face detecting apparatus according to claim 7, wherein said face detecting device includes a computer, wherein said image capturing means captures a face of said computer, and a skin color area in which a part of an area of the image captured by said image capturing means is set. A first determination unit that performs one of a satisfaction determination that a criterion is satisfied and a non-satisfaction determination that a part of the area does not satisfy the skin color area criterion, and the first determination A search area setting unit configured to set a search area including a part of the image area that satisfies the skin color area standard when the satisfaction determination is performed; and whether a face area exists in the set search area. A second determining unit that determines whether or not a face region is present in the set search region when the first determining unit performs the non-satisfaction determination, or The judgment is the second judgment. Means for determining whether or not a dynamic area is included in the image, and determination of whether or not the image includes a dynamic area by the auxiliary determination means. An auxiliary search area setting means for setting an auxiliary search area including the dynamic area; and a third determining whether or not a face area exists in the set auxiliary search area by using pattern matching. It is a face detection device that functions as a determination unit.
[0016]
The invention according to claim 8 is the method according to claim 3, wherein a determination is made by the second determination means that a face area exists in the search area set by the search area setting means, or the auxiliary search is performed. Resetting means for resetting the skin color area reference based on color data in the face area when the auxiliary determining means determines that a face area exists in the auxiliary search area set by the area setting means; It is characterized by including.
[0017]
According to a ninth aspect of the present invention, in the ninth aspect, the resetting means is a face detecting apparatus that changes the resetting of the skin color region reference according to a predetermined condition.
[0018]
According to a tenth aspect of the present invention, in the ninth aspect, the resetting means further comprises a detection area determining means for determining whether the face area is determined to be the search area or the auxiliary search area. A face detection apparatus that sets the skin color area reference differently when the face area is determined to be present in the search area and when the face area is determined to be present in the auxiliary search area It is characterized by being.
[0019]
According to an eleventh aspect of the present invention, in the tenth aspect, when it is determined that the face area exists in the auxiliary search area, the resetting unit determines whether a ratio of the detected dynamic area is equal to or greater than a predetermined threshold value. Based on the determination of the dynamic region ratio determining means that determines whether the ratio is less than the threshold value, the case where the ratio of the detected dynamic region is equal to or greater than a predetermined threshold value, It is a face detection device that performs different resetting.
[0020]
According to a twelfth aspect of the present invention, in the eleventh aspect, the resetting means determines that a ratio of the detected dynamic area is less than a predetermined threshold when it is determined that a face area exists in the auxiliary search area. In the case where the rate at which the past face area is detected in the search area by the past history determining means for determining the rate at which the past face area is detected in the search area is equal to or greater than a predetermined threshold value, and when the rate is less than the predetermined threshold value, And a face detection device that sets the skin color region reference again differently.
[0021]
(Function) In the first and seventh aspects of the present invention, the skin color region reference includes, for example, a skin color effective range set as a reference skin color and a skin color included in the skin color effective range within an area of a certain size. This is combination data with the existence ratio that the region exists in a region equal to or greater than a predetermined ratio. Alternatively, only the skin color effective range set as the reference skin color may be used as the skin color area reference. When it is determined that a part of the acquired image area satisfies the skin color area standard, a search area including the skin color area specified by the skin color area standard is set. When it is determined that a part of the acquired image area does not satisfy the skin color area standard (non-satisfaction determination), or when it is determined that no face area exists in the search area, A determination is made as to whether a dynamic area is included in the image. When it is determined that the image includes a dynamic area, an auxiliary search area including the dynamic area is set. The set search area or the set auxiliary search area is a target of the face area search by pattern matching.
[0022]
The specification of the flesh-colored area using the flesh-colored area criterion indicates that there is a possibility that a face area is present in the set search area. Even when a part of the image area does not satisfy the skin color area standard, the auxiliary search area set using the dynamic area suggests a possibility that a face area exists inside the auxiliary search area. The detection of a face area by pattern matching is performed by comparing a face pattern registered in advance with an acquired image. A method using pattern matching that is hardly affected by a change in environment or a change in camera parameter is a suitable method for determining whether or not the specified skin color region actually includes a face region. It takes time to search the entire image by a method using pattern matching, and it is difficult to speed up face detection. A face detection method that combines a face detection method using color and a face detection method using pattern matching improves the speed and accuracy of face detection. A face detection method in which a face detection method using a dynamic area is further combined with the above method further improves face detection accuracy.
[0023]
In the second and eighth aspects of the present invention, the resetting of the flesh color area reference is desirable in coping with a change in environment and a change in camera parameters.
According to the third and ninth aspects of the present invention, the re-setting of the flesh-color area reference is based on the reliability of the newly detected flesh-color, when rewriting is increased in the case of reliable data, or when it is considered that false detection is performed. Is preferable for improving the accuracy of more appropriate skin color detection.
[0024]
In the invention of claims 4 and 10, when the face detection is successful in the search area in which the search range is set based on the skin color, it can be determined that the reliability of the data is high. On the other hand, when the face is detected in the auxiliary search area in which the search range is set by motion, it can be determined that the reliability of the data is low. Therefore, it is desirable that the resetting of the skin color area reference also be different depending on the face detection.
[0025]
In the invention of claims 5 and 11, when face detection is performed in the auxiliary search area in which a search range is set by motion, if a dynamic area exists at a rate equal to or higher than a specific threshold, the illumination is increased. It is also predicted that it has changed, the camera has moved, or the equipment has failed. Therefore, in such a case, since it is necessary to adapt to a new environment with a new flesh color region reference, it is desirable to make a reset so that the flesh color region standard is quickly replaced.
[0026]
In the invention of claims 6 and 12, even if a face area is detected in an auxiliary search area in which a search range is set by motion, a past face area is determined by a predetermined threshold in the search area set by skin color. If there is a history of detection as described above, it is desirable to reset the current flesh color area reference so that the current data is less reliable and to reduce the influence of the detected data.
[0027]
BEST MODE FOR CARRYING OUT THE INVENTION
(First Embodiment) Hereinafter, a first embodiment of the present invention will be described with reference to FIGS.
[0028]
As shown in FIG. 1A, the face detection device 10 includes a video camera 11 as an imaging unit and a computer 12. The video camera 11 is a CCD camera. Individual frames (image data) captured by the video camera 11 are input to the computer 12 as color images. M is a person who is in the shooting range of the video camera 11. F is the face of the person M.
[0029]
FIG. 1B is a functional block diagram of the computer 12. The image acquisition unit 13 acquires an image from the video camera 11. The dividing unit 14 divides an image obtained from the video camera 11 into a plurality of small areas. FIG. 2A shows an example of the acquired image G. Fo in FIG. 2A represents a face area. FIG. 2B shows a divided image obtained by dividing the acquired image into a plurality of small regions having the same shape and the same size. The small area in FIG. 2B is a rectangular area partitioned by vertical and horizontal lines. In the illustrated example, 120 (= 10 × 12) small areas are partitioned.
[0030]
The first determination unit 15 determines whether or not a part of the acquired image area satisfies the set skin color area standard. That is, the first determination unit 15 determines whether a part of the region of the image captured by the video camera 11 as the imaging unit satisfies the set skin color region standard, and determines whether the part of the region is satisfied. Either a non-satisfaction determination that the skin color region standard is not satisfied is performed. The skin color region criterion in the present embodiment is a preset number (hereinafter, referred to as the number of pixels of a color included in the effective skin color range shown in FIG. 3A, FIG. , Threshold value N0) or more. In other words, the skin color region reference is defined as the skin color effective range set as the reference skin color and the existence ratio that the skin color included in the skin color effective range exists in a region of a predetermined size or more within a certain size region. It is combination data. The skin color area reference data is stored in the storage unit 16. In the present embodiment, the color data is represented using an RG space color system.
[0031]
A curve E1 in the graph of FIG. 3A represents a frequency distribution of R values among RG values relating to skin color, and a curve E2 in a graph of FIG. 3B represents a frequency of G values among RG values relating to skin color. Represents the distribution. The horizontal axis in FIG. 3A represents the R value, and the vertical axis represents the number of accumulated pixels. The horizontal axis in FIG. 3B represents the G value, and the vertical axis represents the number of accumulated pixels. For example, the number of pixels having an R value of Rx is represented as Nx of the total number of pixels (for example, 40 × 40 = 1600 pixels) in one small area.
[0032]
The initial frequency distribution curves E1 and E2 are set as normal distribution curves. The R value represented by m1 shown in FIG. 3A represents the average value of the R values for the curve E1, and σ1 represents the variance of the R values for the curve E1. The G value represented by m2 shown in FIG. 3B represents the average value of the G values for the curve E2, and σ2 represents the variance of the G values for the curve E2. K in FIGS. 3A and 3B is a preset coefficient. k is 1, for example.
[0033]
The graph of FIG. 4 represents a reference effective skin color range set using the average values m1 and m2 and the variance values σ1 and σ2 in the frequency distribution curves E1 and E2 of FIGS. 3A and 3B. In the graph of FIG. 4, the horizontal axis represents the G value, and the vertical axis represents the R value. The area of the rectangle H represents the skin color effective range. The point P in the area of the rectangle H is an average coordinate in the RG space specified by the average value m1 of the R value and the average value m2 of the G value. The effective skin color range H can be changed by adjusting the coefficient k.
[0034]
When the first determination unit 15 determines that the part of the acquired image G satisfies the skin color region criterion, the search region setting unit 17 determines the image specified by the skin color region criterion. A search area including a skin color area in G is set. That is, when a certain small area is a skin color small area, the search area setting means 17 sets a search area composed of a small area which is not a skin color adjacent to the small skin color area and a skin color small area. This search area is temporarily stored in the storage unit 16.
[0035]
The second determination unit 18 determines whether a face area exists in the search area set by the search area setting unit 17 using pattern matching. The storage unit 16 stores a face pattern, and the second determination unit 18 compares the registered face pattern with an image in the set search area.
[0036]
When the second determination unit 18 determines that the face area does not exist in the search area, the auxiliary determination unit 19 determines whether the captured image includes a dynamic small area. I do. In the present embodiment, the auxiliary determination unit 19 determines the acquired image G and the image g acquired immediately before this [for example, the chain line diagram shown in FIG. ], The presence or absence of a dynamic small area is determined on the basis of the luminance difference at the same pixel. In the present embodiment, a small area in which the number of pixels whose luminance difference is equal to or greater than a predetermined value is equal to or greater than a predetermined number is defined as a dynamic small area (dynamic area). The reference data of the dynamic small area indicating that the number of pixels whose luminance difference is equal to or more than a predetermined value is equal to or more than a predetermined number is stored in the storage unit 16.
[0037]
The auxiliary search area setting means 20 sets the auxiliary search area including the dynamic small area when the auxiliary determination means 19 determines that the acquired image G includes the dynamic small area. . In other words, when a certain small area is a dynamic small area, the auxiliary search area setting means 20 sets the auxiliary search including the small area which is not the dynamic small area adjacent to the dynamic small area and the dynamic small area. Set the area. This auxiliary search area is temporarily stored in the storage unit 16.
[0038]
The third determination unit 21 determines whether or not a face area exists in the set auxiliary search area by using pattern matching.
When the second determination unit 18 determines that the face area exists in the set search area or the auxiliary search area, the resetting unit 22 performs the skin color area reference based on the color data in the face area. Resets the effective range of skin color that is part of. That is, the resetting unit 22 resets the skin color area reference. The color data in the face area is data represented by an RG value and the number of pixels (corresponding to the frequency distribution curves E1 and E2 in FIGS. 3A and 3B). Resetting the effective skin color range (that is, resetting the skin color area reference) is to change the skin color effective range that is a part of the skin color area reference stored in the storage unit 16. In the present invention, the resetting of the flesh color area reference means resetting the flesh color effective range.
[0039]
5 and 6 are flowcharts showing a face detection program executed by the face detection device 10. Hereinafter, face detection by the face detection device 10 will be described with reference to this flowchart.
[0040]
In step S1, first, the image acquisition unit 13 determines whether to acquire an image from the video camera 11. That is, in the present embodiment, acquisition of an image from the video camera 11 is performed at predetermined time intervals t (for example, t = 0.1 second), and the image acquisition unit 13 performs image acquisition. It is determined whether it is time. If it is determined that it is time to acquire an image (step S1 is YES), the image acquiring unit 13 acquires an image from the video camera 11 (step S2). Step S2 is an imaging step. On the other hand, if the image acquisition unit 13 determines that it is not time to acquire an image (step S1 is NO), step S1 is repeated.
[0041]
After obtaining the image from the video camera 11, the dividing unit 14 divides the obtained image into a plurality of small areas (Step S3). After dividing the acquired image G into a plurality of small areas, the first determination unit 15 determines the skin color small based on the color data (RG value and the number of pixels) in the small area and the skin color area reference. An area search is performed (step S4).
[0042]
In step S4, the first determination unit 15 determines the relationship between the RG value of the color data (data represented by the RG value and the number of pixels) in each small region and the effective skin color range indicated by the rectangular H region shown in FIG. Make a comparison. Then, the first determination unit 15 checks whether or not the number of pixels of a color included in the effective skin color range indicated by the square H area is equal to or larger than the threshold value N0 in the small area.
[0043]
The first determination unit 15 determines whether each small region is a skin color region (step S5). Step S5 is a first determination step. If the number of pixels of the color included in the skin color effective range indicated by the square H area is equal to or larger than the threshold value N0 in the small area, the first determination unit 15 determines that the small area is a skin color small area. It is determined that there is (sufficiency determination) (YES in step S5). If the number of pixels of a color included in the effective skin color range indicated by the square H does not reach the threshold value N0 in the small area, the first determination unit 15 determines that this small area is a small area of skin color. (Step S5: NO).
[0044]
It is assumed that the image of at least one small area satisfies the skin color area standard. That is, when it is determined that at least one small area is a skin color small area (YES in step S5), the first determination unit 15 determines that the skin color small area adjacent to the skin color small area is not a skin color small area. A search area D1 (see FIG. 2 (b)) is set and stored in the storage means 16 (step S6). Step S6 is a search area setting step. In FIG. 2B, assuming that only the small area A1 is a skin color small area, the search area D1 is adjacent to the small area A1 which is a skin color area and the small area A1 as shown in FIG. 2C. This is an area including the small areas A2, A3, A4, A5, A6, A7, A8, and A9. Note that the state where the corner of the small area (for example, A1) is in contact with the corner of the small area (for example, A2) is also the state where the small area A1 and the small area A2 are adjacent.
[0045]
After setting the search area D1, the second determination unit 18 scans the search area D1 using a face pattern (step S7). The storage unit 16 stores regions having a set shape (for example, a rectangle) and a plurality of sizes, and a known face pattern. In step S7, the second determination unit 18 cuts out the candidate area from the search area D1 by using one area K (see FIG. 2C) among the areas of a plurality of sizes, and registers the candidate area. Check with the face pattern. This collation is performed by scanning all positions in the search area D1 with all the registered rectangle sizes. Then, the second determination unit 18 determines whether or not the region cut out by the size of the region K is a face region using a pattern matching method (step S8). Step S8 is a second determination step.
[0046]
In the present embodiment, a statistical method such as linear discriminant analysis, SVM (Support Vector Machines), or SNow (Sparse Network of Windows) is used to determine whether there is a face area by pattern matching.
[0047]
It is assumed that it has been determined that the face area is present in one of the cutout areas (YES in step S8). In this case, the resetting unit 22 sets a new skin color effective range using the color data obtained from the acquired image G and the stored skin color effective range, and sets the new skin color effective range to the previous one. The storage means 16 is stored in place of the skin color effective range (step S9). Step S9 is a resetting step. In setting a new skin color effective range, for example, a curve obtained by multiplying the frequency distribution curve E1 in FIG. 3A by a predetermined coefficient (for example, 0.5) and an R value of a skin color obtained from a detected face area are used. A frequency distribution curve of the sum with the frequency distribution curve is obtained. Similarly, the frequency of the sum of the curve obtained by multiplying the frequency distribution curve E2 of FIG. 3B by a predetermined coefficient (for example, 0.5) and the frequency distribution curve relating to the G value of the skin color obtained from the detected face area Find the distribution curve. A new effective skin color range corresponding to the effective skin color range shown in the graph of FIG. 4 is set from these new frequency distribution curves.
[0048]
After setting the new effective skin color range, the second determination unit 18 causes the storage unit 16 to store the image data determined to have a face area (step S10). After storing the image data determined to have a face area, the computer 12 returns to step S1.
[0049]
It is assumed that it is determined in step S8 that the face area is not located in any of the cutout areas (NO in step S8). In this case, the auxiliary determination unit 19 determines whether the acquired image G and the image g acquired immediately before the image G (for example, the chain line diagram illustrated in FIG. 2B represents a part of the image g). A search for a dynamic small area based on the luminance difference in the same pixel is performed (step S11). And the auxiliary | assistant determination means 19 determines the presence or absence of a dynamic small area (step S12). Step S12 is an auxiliary determination step.
[0050]
If a dynamic small area exists (YES in step S12), the auxiliary search area setting unit 20 determines whether the dynamic small area is a small area that is not a dynamic small area and is adjacent to the dynamic small area. The auxiliary search area D2 (see FIG. 2D) is set and stored in the storage unit 16 (step S13). Step S13 is an auxiliary search area setting step. As shown in FIG. 2D, it is assumed that small areas A1, A2, A3, A5, A10, A11, A12, and A13 are dynamic small areas. Then, the auxiliary search area D2 is adjacent to the dynamic small areas A1, A2, A3, A5, A10, A11, A12, A13 and the dynamic small areas A1, A2, A3, A5, A10, A11, A12, A13. This is a region that is combined with a small region (a small region with oblique lines in FIG. 2D).
[0051]
After setting the auxiliary search area D2, the third determination unit 21 scans the auxiliary search area D2 with a face pattern (step S14). In step S14, the third determination unit 21 cuts out and registers a candidate area from the auxiliary search area D2 by using one area K (see FIG. 2C) among the areas of a plurality of sizes. Collation with the face pattern is performed. This collation is performed by scanning all positions in the auxiliary search area D2 with all the registered rectangle sizes. Then, the third determination unit 21 determines whether or not the region cut out by the size of the region K is a face region by using a pattern matching method (step S15). Step S15 is a third determination step.
[0052]
It is assumed that the determination is made that the face area is also in one of the cutout areas (YES in step S15). In this case, the resetting unit 22 sets a new skin color effective range using the color data obtained from the acquired image G and the stored skin color effective range, and sets the new skin color effective range to the previous one. The storage means 16 is stored in place of the skin color effective range (step S9). After setting the new effective skin color range, the third determination unit 21 causes the storage unit 16 to store the image data determined to have a face area (step S10). After storing the image data determined to have a face area, the computer 12 returns to step S1.
[0053]
If the images in all the small areas do not satisfy the flesh color area standard, the first determination unit 15 determines that there is no flesh color small area (NO in step S5). If there is no skin-colored small region, the auxiliary determination unit 19 performs a search for a dynamic small region based on the luminance difference between the acquired image G and the image g acquired immediately before the image G. Perform (Step S11). And the auxiliary | assistant determination means 19 determines the presence or absence of a dynamic small area (step S12).
[0054]
If a dynamic small area exists (YES in step S12), the auxiliary search area setting unit 20 determines whether the dynamic small area is a small area that is not a dynamic small area and is adjacent to the dynamic small area. The auxiliary search area D2 is set and stored in the storage unit 16 (step S13).
[0055]
After setting the auxiliary search area D2, the third determination unit 21 scans the auxiliary search area D2 with a face pattern (step S14). In step S14, the third determination unit 21 cuts out and registers a candidate area from the auxiliary search area D2 by using one area K (see FIG. 2C) among the areas of a plurality of sizes. Collation with the face pattern is performed. This collation is performed by scanning all positions in the auxiliary search area D2 with all the registered rectangle sizes. Then, the third determination unit 21 determines whether or not the region cut out by the size of the region K is a face region by using a pattern matching method (step S15).
[0056]
It is assumed that the determination is made that the face area is also in one of the cutout areas (YES in step S15). In this case, the resetting unit 22 sets a new skin color effective range using the color data obtained from the acquired image G and the stored skin color effective range, and sets the new skin color effective range to the previous one. The storage means 16 is stored in place of the skin color effective range (step S9). After setting the new effective skin color range, the third determination unit 21 causes the storage unit 16 to store the image data determined to have a face area (step S10). After storing the image data determined to have a face area, the computer 12 returns to step S1.
[0057]
The present embodiment has the following effects.
(1) If it is determined that the image of at least one small area satisfies the skin color area standard (YES in step S5), the skin color small area specified by the skin color area standard [FIG. In the example, a search area (small area A1 to A9) including the small area A1] is set. The set search area is narrower than the entire area of the acquired image, and the narrowed search area is a target of face area search by pattern matching. The identification of a small skin color region using the skin color region reference indicates that there is a possibility that a face region exists in the set search region. The method using pattern matching, which is hardly affected by changes in the environment and changes in camera parameters, is used to determine with high accuracy whether or not a search area including a small area of the specified flesh color actually includes a face area. This is a preferred method.
[0058]
Attempts to search for a face region using a color-based method lack accuracy. It takes time to search for a face region over the entire image by a method using pattern matching, and it is difficult to speed up face detection. The technique of narrowing the search area by a face detection technique using colors and determining the presence or absence of a face area in the narrowed search area by pattern matching improves the speed and accuracy of face detection.
[0059]
(2) When the environment (brightness or darkness) changes, the effective skin color range may not be appropriate. In the present embodiment, the search area is specified using the color (step S6), and the skin color effective range is reset every time the presence of a face area is determined by pattern matching (YES in steps S8 and S15). (Step S9). The effective skin color range obtained by the resetting (that is, the resetting of the skin color region reference) can be adjusted by selecting a coefficient to be applied to the frequency distribution curve. Therefore, if the coefficient is appropriately selected, the effective range of the skin color can be appropriately adjusted according to the environmental change.
[0060]
(3) In the case where the search area cannot be set using the skin color, in the present embodiment, the search area is determined based on the luminance difference between the same pixel in the acquired image G and the image g acquired earlier. Search for a dynamic small area. If there is a dynamic small area, an auxiliary search area D2 including the dynamic small area is set. The set auxiliary search area D2 is a target of a face area search by pattern matching. The acquired dynamic small area indicates that there is a possibility that a face area exists in the set auxiliary search area D2. Such a search for a dynamic small area enables the face area to be detected even when the face area cannot be detected because the effective skin color range is not properly set. That is, a face detection method that combines a face detection method using color, a face detection method using pattern matching, and a face detection method using dynamic regions further improves the accuracy of face detection.
[0061]
(4) When there are a plurality of faces as viewed from the video camera 11 and a part of those areas overlaps, it is difficult to detect a face accurately. For example, as shown in FIG. 2E, when the face regions F1 and F2 overlap in the image, the face regions F1 and F2 are detected in a connected state only by the face region detection method using color. In the present embodiment, since the inside of the search area D1 is scanned by the face pattern, even when the face areas overlap in the image as shown in FIG. It can be detected accurately.
[0062]
In the present invention, the following embodiments are also possible.
As a method of setting the effective range of skin color, a Euclidean distance to each RG value is calculated from the average value in the normalized RG space corresponding to the average value of RG values related to skin color, and a normalized RG space having a variance as a radius is calculated. May be set as the skin color effective range.
[0063]
As a method of setting the effective range of the skin color, the average value m, the variance value σ, and the coefficient k are used.
[0064]
In the above-described embodiment, the search for the dynamic small area may be performed simultaneously with the search for the skin color small area. That is, a search for a dynamic small area may be performed between step S3 and step S4 in FIG. 5 or between step S4 and step S5.
[0065]
In the above-described embodiment, when the auxiliary search area is set, the search area set when it is determined that a part of the image area satisfies the skin color area standard is set as the auxiliary search area. It may not be included. By doing so, the auxiliary search area is reduced, and the time for scanning the auxiliary search area with the face pattern to search for the face area can be shortened.
[0066]
In the above-described embodiment, the color data and the graph indicating the effective range of the skin color are represented using the color system of the normalized RG space. However, other color systems such as Luv space and XYZ space may be used. Good.
[0067]
The skin color effective range may be reset using color data in an area excluding the eye area and the lip area from the detected face area. In this way, the color of the relatively homogeneous skin color portion (for example, the cheek region) in the face region can be mainly used for resetting the effective skin color range, and the appropriateness of the setting of the effective skin color range is improved. .
[0068]
The detected face area may be stored, and the size and position of the stored face area may be used to detect the face area in the next acquired image. The size of the cutout area can be limited by the size of the stored face area, and the types of cutout areas having different sizes for scanning the search area can be reduced. The smaller the number of types of cutout areas having different sizes, the shorter the scanning time, and high-speed face area detection can be expected. Also, if the position of the detected face area is used, the search area in the next acquired image can be set more accurately.
[0069]
In the first embodiment, the function of the third determination unit 21 may be shared by the second determination unit 18.
In the above-described embodiment, the skin color effective range is set to be the same for all regions of the image. However, the skin color effective range may be set for each small region. If a different skin color effective range is used for each small area, it is possible to absorb a change in skin color caused by a difference in lighting conditions depending on a location in the photographing range.
[0070]
In resetting the skin color effective range for each small area, the position information of the face detection result may be used to reset the skin color effective range of only the small area detected as including part or all of the face area. . For example, in the case of FIG. 2D, the skin color effective range is reset only for the small areas A1, A2, A3, and A5, and the skin color effective range is not reset for the other small areas. I do.
[0071]
Alternatively, in consideration of the fact that the person moves, a small area detected as including part or all of the face area, and a skin color effective range of a small area adjacent to the detected small area are reset. Is also good. For example, in the case of FIG. 2D, the small areas A1, A2, A3, and A5 and the small areas adjacent to the small areas A1, A2, A3, and A5 [the small areas A1, A2, and A2 in FIG. A3, A5, a shaded small area surrounding A3) may be reset.
[0072]
In the above-described embodiment, the skin color area reference is based on the case where the number of pixels of the color included in the effective skin color range among the pixels in the small area is equal to or larger than the threshold value N0. An area reference may be used. In this case, a pixel of a color included in the skin color effective range becomes a skin color area. For example, in the case of FIG. 2D, the face part indicated by Fo is a skin color area formed by a group of pixels of a color included in the skin color effective range. The search area may be an area obtained by combining a small area including a part or all of the skin color area and a small area adjacent to the small area. In the case of FIG. 2D, the small areas A1, A2, A3, and A5 are small areas including part or all of the skin color area. The search areas related to the small areas A1, A2, A3, and A5 are the small areas A1, A2, A3, and A5 and the small areas adjacent to the small areas A1, A2, A3, and A5 [the small area in FIG. A1, A2, A3, and A5).
[0073]
When only the skin color effective range is used as the skin color region reference, the search region may be set without dividing the image into small regions. In this case, the search region may be a region (for example, a rectangular shape) that is larger than the flesh color region and that has a shape including the flesh color region.
[0074]
(Second embodiment)
Next, a face detection device according to a second embodiment will be described with reference to FIGS. The second embodiment basically modifies the skin color region reference based on the result of face detection by pattern matching as in the first embodiment as shown in the flowcharts of FIGS. Adapting to change. Specifically, the resetting unit 22 resets the skin color area reference. At this time, in setting a new effective skin color range, for example, a curve obtained by multiplying the frequency distribution curves E1 and E2 of FIGS. 3A and 3B by a predetermined coefficient (for example, 0.5) and a detection face A frequency distribution curve of the sum with the frequency distribution curve relating to the R value and G value of the skin color obtained from the region is obtained. A new skin color effective range corresponding to the skin color effective range shown in the graph of FIG. 4 is set from these new frequency distribution curves (step 9 (hereinafter, “step” is abbreviated as S)).
[0075]
However, in the first embodiment, the parameter setting method based on the flesh color region is a method of adding a uniform ratio (0.5 in the first embodiment) to the histogram. Is difficult. In addition, when an erroneous detection is performed at the time of face detection by pattern matching, addition is performed at the same ratio as that when detection is correctly performed, and a skin color parameter is set. Therefore, parameter setting may be inappropriate.
[0076]
Therefore, the face detection method and apparatus of the second embodiment are different from the flowcharts of the first embodiment shown in FIGS. 5 and 6 in the process of resetting the effective range of skin color (S9).
[0077]
From the image acquired from the camera as shown in FIG. 7 (S1 to S2), a skin color area is detected (S3, S4, S5; YES). As shown in FIG. 8, the vicinity of the skin color area is set as the face search area (S6). Then, as shown in FIG. 9, a face area is detected using pattern matching in the vicinity of the skin color area (S7). If the face area is detected (S8; YES) and the reliability is high, the skin color effective range resetting (S9) uses the pixel values in the face area to reset the skin color area reference, which is the skin color detection parameter. Make settings (adjustments).
[0078]
However, when the skin color region reference is not properly set, such as immediately after an initial frame or an environmental change, detection using color information may fail (S5; NO). Therefore, if the face cannot be detected by the face detection using the skin color, a dynamic small area is searched (S11). If there is a dynamic small area (S12; YES), the dynamic small area (all frames) is searched. (Using an image) is set as a search range, and a face area is searched and detected using pattern matching (FIG. 10).
[0079]
At this time, when the face area is detected, the reliability of the histogram (FIG. 11) of the skin color pixels accumulated so far is calculated using the pixel values in the face area as shown in the following equation (1). By multiplying by the corresponding coefficient a and adding it up, the detection parameters of the flesh color are set. Here, the coefficient “a” is a fixed value (for example, 0.5) in the first embodiment. However, in the second embodiment, the coefficient “a” is a coefficient having a different value depending on the condition and a range of 0 <a <1. Take the value with “Ht” indicates a histogram before updating, and “Hc” indicates a histogram of a detected face area. “Ht + 1” indicates the updated histogram.
[0080]
Ht + 1 = aHt + (1-a) Hc (1)
As shown in FIG. 11A, the peak of the histogram E3 has an R value of m3 and a height of 1. This height corresponds to the number of pixels in the first embodiment, but is referred to as the accumulation degree in the second embodiment. On the other hand, in the histogram E4 obtained by multiplying the histogram E3 by the coefficient a, the peak R value is at the same m3, and the height is a / 1.
[0081]
FIG. 12 is a flowchart showing in detail the skin color area reference setting process in the skin color effective range resetting step (S9) of the second embodiment.
Here, the computer 12 executes a detection area determination step (S21) of determining whether the determination that the face area is present is the search area or the auxiliary search area. Note that the computer 12 executing this step functions as a detection area determination unit. When the detected face area is the search area set by the skin color (S21; YES), it can be estimated that the setting of the skin color area reference for detecting the face in the frame is appropriate. Therefore, the coefficient a to be applied to the histogram stored as flesh color pixels is set to “a = a0 (for example, 0.5)” (S22), and the flesh color detection criterion (skin color parameter) is set by adding to the histogram of the detected pixels. (S23).
[0082]
For example, when the accumulated histogram E3 shown in FIG. 11A is multiplied by the coefficient a0 (= 0.5) in this case, the accumulated accumulation degree N is halved as a whole. , And a curve like a histogram E4. In this case, the average values m3 and m4 of the R values do not change.
[0083]
Similarly, when (1-a) = 0.5 is applied to a histogram (not shown) having an average R value of m5 generated from the detected skin color data, a histogram shown in FIG. E5.
[0084]
Next, the histogram E4 and the histogram E5 are added. That is, a new histogram E6 is generated by adding the accumulation degrees N of the two. In this case, the average value m6 of the histogram E6 is intermediate between the average value m4 of the histogram E4 and the average value m5 of the histogram E5. The position of the new average value changes depending on the weighting of the accumulated histogram data accumulated in the past and the newly detected detection histogram data, in other words, the manner of setting the coefficient a.
[0085]
If the detected face area is not the search area set by the skin color (S21; NO), but is the auxiliary search area set by the movement, it is considered that "the environment has changed". When an environmental change occurs, the pixel value of the entire screen changes, so that a motion area is detected on the entire screen by motion detection. Therefore, when the ratio of the motion area to the entire screen is equal to or more than a certain threshold (for example, 80% or more) (S24; YES), it can be inferred that the environment has changed. Small “a = a1 (for example, 0.2)” is set (S25), and the detection pixels are added (S23). By reducing the coefficient applied to the accumulation histogram, the influence of the detected pixel on the setting of the flesh color area reference is increased, so that the detection pixel can quickly adapt to the environment. As shown in FIG. 11C, in this case, the average value m9 of the new histogram E9 is smaller than the average value m7 of the histogram E7 generated from the accumulated histogram, that is, the average value of the histogram E8 generated from the detected data. It is close to m8.
[0086]
Here, when it is determined that a face region exists in the auxiliary search region, the computer 12 determines whether the ratio of the detected dynamic region is equal to or greater than a predetermined threshold or less than a predetermined threshold. The step is executed (S24). Note that the computer 12 executing S24 functions as a dynamic area ratio determination unit. When the detected face area is not the search area set by the flesh color (S21; NO) and the ratio of the entire motion area is smaller than a certain threshold and small (S24; NO), it is unlikely that "the environment has changed". . On the other hand, the reason why the skin color region could not be detected from the candidate region may be a case where “the current setting of the skin color parameter is inappropriate” and a case where “an incorrect detection at the time of face detection by a pattern” is performed. The guess of which of the above is based on whether the previous frame (for example, 20 frames) in which the face has been successfully detected is the “search area by skin color” or the “auxiliary search area by motion”. (S26).
[0087]
Here, when it is determined that the face area exists in the auxiliary search area, the computer 12 determines whether the past face area is detected in the search area when the rate of the detected dynamic area is less than the predetermined threshold. Is executed (S26). The computer 12 executing S26 functions as a past history determination unit. If the result of the face detection so far is “candidate area based on skin color” (S26; YES), it is “detected from the moving area even though there is no environmental change and the skin color setting is appropriate”. , An erroneous detection at the time of face detection by a pattern ". Therefore, the coefficient a applied to the accumulation histogram is set to “a = a2 (for example, 0.8)” which is larger than a0 (S27), and the detection pixels are added (S23). In this case, the new average value m is closer to the average value of the accumulated histogram than the average value of the histogram generated from the detected data. By increasing the coefficient a applied to the accumulation histogram, the influence of the detected pixel on the flesh color setting can be reduced, and inappropriate setting of the flesh color parameter due to erroneous detection can be suppressed.
[0088]
Further, when a face image is detected from the “candidate region due to motion” in the past frame (S26; NO), it can be estimated that “the current skin color setting is inappropriate”, and thus the coefficient to be applied to the accumulation histogram is used. “a” is set to a small value “a = a3 (for example, 0.2, etc., may be the same as a1)” (S28), and the detection pixels are added (S23). By reducing the coefficient applied to the accumulation histogram, the influence of the detected pixel on the setting of the flesh color area reference is increased, so that the detection pixel can quickly adapt to the environment.
[0089]
As described above, when setting the skin color region reference, the coefficient a to be applied to the accumulated histogram is determined from “from which of the search regions or the auxiliary search region was detected”, “what was the overall change”, "What was the detection result so far?" Thus, a face detection method and apparatus that flexibly adapts to the environment can be realized by judging whether or not the detection result is correct and controlling parameters when setting the skin color area reference.
[0090]
The present embodiment has the following effects.
(1) In a system for monitoring a person, there is no need to manually set detection parameters according to camera parameters and environmental light, and a face region can be accurately detected at high speed. In addition, it is possible to flexibly adapt to erroneous detection in the case of detection by a pattern and environmental changes.
[0091]
(2) Further, by adjusting the coefficient a in addition to the adjustment of the coefficient k and the dispersion of the histogram in consideration of the sensitivity characteristics of the camera and the characteristics of the ambient light, the face is reduced in noise and the noise is reduced. Adjustments can be made for robust detection.
[0092]
The invention that can be grasped from the above-described embodiment will be described below.
[1] The method according to any one of claims 1 and 2, further comprising a dividing step of dividing the captured image into a plurality of small areas, wherein the first determining step includes setting an image in the small area. Determining whether or not an image in the small area satisfies the flesh color area criterion. A face detection method in which a search area is set so as to include a small area satisfying the following.
[0093]
[2] In the above item [1], the skin color area reference includes a skin color effective range set as a reference skin color and an area where a skin color included in the skin color effective range has a predetermined ratio or more within an area of a certain size. A face detection method that is data that is combined with the existence ratio of being present.
[0094]
[3] The method according to any one of claims 3 and 4, further comprising a dividing unit that divides the image taken by the imaging unit into a plurality of small areas, wherein the first determination unit is configured by the dividing unit. It is determined whether or not the image in the divided small area satisfies the set skin color area criterion, and the search area setting means determines that the image in the small area satisfies the skin color area criterion. A face detection apparatus that sets a search area so as to include a small area satisfying the skin color area criterion when the determination is made by the first determination unit.
[0095]
[4] In the above item [3], the skin color area reference includes a skin color effective range set as a reference skin color and an area in which a skin color included in the skin color effective range has a predetermined ratio or more within an area of a certain size. The face detection device is data that is combination data with the existence ratio of being present in.
[0096]
[5] an imaging step for imaging the face, a satisfaction determination that a part of the area of the captured image satisfies the set skin color area standard, and a part of the area satisfies the skin color area standard. Setting a search area including a part of an area of the image satisfying the skin color area criterion, wherein the first determination step performs any one of an unsatisfied determination that the skin color area is not satisfied; and A search area setting step, a second determination step of determining whether or not a face area exists in the set search area using pattern matching, and when the unsatisfaction determination is performed, or When it is determined that no face region exists in the search region, an auxiliary determination step of determining whether a dynamic region is included in the image, and that the image includes a dynamic region Was determined In this case, an auxiliary search area setting step of setting an auxiliary search area including the dynamic area and a third step of determining whether a face area exists in the set auxiliary search area by using pattern matching. A face detection program for causing a computer to execute the determining step.
[0097]
【The invention's effect】
As described in detail above, the present invention has an excellent effect that both the speed and accuracy of face detection can be improved.
[Brief description of the drawings]
FIG. 1A is a block diagram illustrating a face detection device according to an embodiment. (B) is a functional block diagram of a computer.
FIG. 2A is an explanatory diagram of image data captured by a video camera. (B) is an explanatory view of a divided image. (C) is an explanatory view showing a search area. (D) is an explanatory view showing an auxiliary search area. (E) is an explanatory view of an image.
FIG. 3A is a graph showing a frequency distribution curve of an R value. (B) is a graph showing a frequency distribution curve of a G value.
FIG. 4 is a graph showing a skin color effective range.
FIG. 5 is a flowchart showing a face detection program.
FIG. 6 is a flowchart illustrating a face detection program.
FIG. 7 is a diagram showing an example of an input image after small area division (S3).
FIG. 8 is a diagram showing a search area in which a search area is set (S6) based on skin color when the skin color area reference is appropriate.
FIG. 9 is a diagram showing a search method for evaluating the likelihood of a face region in a cut-out region.
FIG. 10 is a diagram showing a search area in which an auxiliary search area setting using motion (S13) is performed.
FIGS. 11A to 11C are graphs showing histograms.
FIG. 12 is a flowchart showing in detail a skin color area reference setting process in a skin color effective range resetting step (S9) according to the second embodiment.
[Explanation of symbols]
10. Face detection device. 11 Video camera (imaging means). 12 Computer (first determining means, search area setting means, second determining means, resetting means). 15: First determination means. 17 Search area setting means. 18 ... second determination means. 19 ... Auxiliary determination means. 20 ... Auxiliary search area setting means. 21. Third determination means. 22 ... resetting means. G ... Image. F ... face. Fo: face area. A1: Small area which is a skin color area. A1, A2, A3, A5, A10, A11, A12, A13... Small areas as dynamic areas. D1 ... Search area. D2: auxiliary search area. H: Effective range of skin color.

Claims (12)

A face detection method executed by a computer, the imaging step for the computer to image a face,
Either a satisfaction determination that a part of the region of the captured image satisfies the set skin color region criterion or a non-satisfaction determination that a part of the region does not satisfy the skin color region criterion A first determination step of performing a determination;
When the satisfaction determination is made, a search area setting step of setting a search area including a part of the area of the image that satisfies the skin color area criterion,
A second determination step of determining whether or not a face area exists in the set search area by using pattern matching;
If the unsatisfied determination is made, or if it is determined that no face area is present in the search area, an auxiliary determination step of determining whether the image includes a dynamic area,
When it is determined that a dynamic area is included in the image, an auxiliary search area setting step of setting an auxiliary search area including the dynamic area,
A third determination step of determining whether a face area exists in the set auxiliary search area by using pattern matching. A resetting step of resetting the skin color area reference based on color data in the face area when it is determined that the face area exists in the set search area or in the set auxiliary search area. The face detection method according to claim 1, comprising: 3. The face detecting method according to claim 2, wherein the resetting step changes the resetting of the skin color region reference according to a predetermined condition. In the resetting step, it is determined that the face area exists in the search area based on the determination in the detection area determination step of determining whether the face area is determined to be the search area or the auxiliary search area. 4. The face detection method according to claim 3, wherein resetting of the skin color region reference is different between a case where the face region is determined to be present in the auxiliary search region and a case where the face region is determined to be present. 5. The resetting step includes, when it is determined that a face area is present in the auxiliary search area, determining whether a ratio of the detected dynamic area is equal to or greater than a predetermined threshold or less than a predetermined threshold. Based on the judgment of the step, the resetting of the skin color region reference is different between a case where the ratio of the detected dynamic region is equal to or more than a predetermined threshold value and a case where the ratio is less than the predetermined threshold value. The face detection method according to claim 4. In the resetting step, when it is determined that a face area is present in the auxiliary search area, a past face area is detected in the search area when a ratio of the detected dynamic area is less than a predetermined threshold. The resetting of the skin color area reference is different depending on whether the rate at which the past face area is detected in the search area by the past history determination step of determining the rate is equal to or greater than a predetermined threshold, and depending on whether the rate is less than the predetermined threshold. The face detection method according to claim 5, wherein A face detection device including a computer,
The computer, imaging means for imaging the face,
A satisfaction determination that a part of the region of the image captured by the imaging unit satisfies the set skin color region criterion, and a non-satisfaction determination that a part of the region does not satisfy the skin color region criterion First determining means for performing any of the following determinations:
A search region setting unit configured to set a search region including a part of the image region satisfying the skin color region standard, when the first determination unit has performed the satisfaction determination;
Second determining means for determining whether or not a face area exists in the set search area by using pattern matching;
When the first determination unit makes the non-satisfaction determination, or when the second determination unit determines that the face area does not exist in the set search area, the image is moved. Auxiliary determining means for determining whether a target area is included,
When the auxiliary determination unit has determined that a dynamic region is included in the image, an auxiliary search region setting unit that sets an auxiliary search region including the dynamic region,
A face detection device that functions as third determination means for determining whether or not a face area exists in the set auxiliary search area by using pattern matching. When the second determining means determines that a face area exists in the search area set by the search area setting means, or in the auxiliary search area set by the auxiliary search area setting means 8. The face detecting apparatus according to claim 7, further comprising resetting means for resetting the skin color area reference based on color data in the face area when the auxiliary determining means determines that a face area exists in the face area. 9. The face detection apparatus according to claim 8, wherein the resetting unit changes the reset of the skin color region reference according to a predetermined condition. The resetting means includes detection area determining means for determining whether the face area is determined to be the search area or the auxiliary search area. Based on the determination, a face area exists in the search area. 10. The face detection apparatus according to claim 9, wherein resetting the skin color area reference is different between a case where it is determined that the face area is present in the auxiliary search area and a case where it is determined that the face area exists in the auxiliary search area. The resetting means determines, when it is determined that a face area exists in the auxiliary search area, whether a ratio of the detected dynamic area is equal to or more than a predetermined threshold or less than a predetermined threshold. Based on the judgment of the means, the resetting of the skin color area reference is different between a case where the ratio of the detected dynamic area is equal to or more than a predetermined threshold value and a case where the ratio is less than the predetermined threshold value. The face detection device according to claim 10. The resetting means may determine that a face area exists in the auxiliary search area, and when a ratio of the detected dynamic area is less than a predetermined threshold, a past face area is detected in the search area. The resetting of the skin color area reference is different depending on whether the rate at which the past face area is detected in the search area by the past history determining means for determining the rate is equal to or greater than a predetermined threshold and when the rate is less than the predetermined threshold. The face detection apparatus according to claim 11, wherein the face detection is performed.

Images