JP2004246424A - Method for extracting skin color area - Google Patents

Method for extracting skin color area Download PDF

Info

Publication number
JP2004246424A
JP2004246424A JP2003032864A JP2003032864A JP2004246424A JP 2004246424 A JP2004246424 A JP 2004246424A JP 2003032864 A JP2003032864 A JP 2003032864A JP 2003032864 A JP2003032864 A JP 2003032864A JP 2004246424 A JP2004246424 A JP 2004246424A
Authority
JP
Japan
Prior art keywords
region
target image
pixel
gaussian
skin color
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
JP2003032864A
Other languages
Japanese (ja)
Inventor
Tai Kuwan Fuuintsuu
Masahide Kaneko
Mitsuhiko Meguro
タイ クワン フーインツー
光彦 目黒
正秀 金子
Original Assignee
Campus Create Co Ltd
Masahide Kaneko
株式会社キャンパスクリエイト
正秀 金子
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Campus Create Co Ltd, Masahide Kaneko, 株式会社キャンパスクリエイト, 正秀 金子 filed Critical Campus Create Co Ltd
Priority to JP2003032864A priority Critical patent/JP2004246424A/en
Publication of JP2004246424A publication Critical patent/JP2004246424A/en
Pending legal-status Critical Current

Links

Images

Abstract

An object of the present invention is to provide a method capable of accurately extracting a skin color region even when a target image includes various skin colors.
A histogram is generated for a hue H and a saturation S for each pixel in a skin color region in a sample image. Next, the histogram is approximated by a Gaussian mixture model having a plurality of Gaussian models. Next, the color difference signals (H, S) for the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model. Next, the value of the one-dimensional information is changed. At this time, the amount of increase in the number of pixels corresponding to the Gaussian model is calculated. The value of the one-dimensional information at which the amount of increase becomes substantially minimum is set as a threshold value. Pixels having one-dimensional information included in the range determined by the threshold are selected from the target image. Thereby, a skin color area can be extracted from the target image.
Further, by using the structure information of the extracted skin color region, a face region can be selected from the skin color region.
[Selection diagram] Fig. 1

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a method for extracting a skin color region from an image.
[0002]
[Background Art]
Techniques for detecting a face area from an image such as a photograph are already known. It is considered that this detection makes it possible to perform personal authentication and search for an image including a face.
[0003]
Conventionally proposed detection techniques use face structure information and skin color information. The detection technique using skin color information is excellent as a means for selecting face area candidates. However, in the related art, there is a problem that if there is a color change based on race or shadow, the detection accuracy of the face region is considerably reduced.
[0004]
[Problems to be solved by the invention]
The present invention has been made in view of the above circumstances. An object of the present invention is to provide a method capable of accurately extracting a flesh-colored area even when conditions of a target image are various.
[0005]
[Means for Solving the Problems]
The method for extracting a flesh color region according to the present invention includes the following steps.
(1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
(2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
(3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting a threshold of the value of the one-dimensional information whose amount is substantially minimum; and
(4) extracting a skin color region from the target image by selecting, from the target image, pixels having the one-dimensional information included in the range determined by the threshold value.
[0006]
The one-dimensional information may be a value in the Gaussian model, and a value corresponding to two color difference signals in a pixel of the target image may be a maximum.
[0007]
The amount of increase in the number of pixels in the step (3) is assumed to be for pixels converted into one-dimensional information by the same Gaussian model,
Associating the threshold with information indicating the Gaussian model;
A pixel selected from the target image in the step (4) is selected from pixels converted into one-dimensional information by the same Gaussian model by a threshold value associated with the Gaussian model.
be able to.
[0008]
The method for extracting a skin color region according to the present invention may include the following steps.
(1) decomposing each pixel in all or a part of the target image into a plurality of clusters using the values of Cb and Cr in the pixel; and
(2) A step in which a region where a pixel belonging to a cluster having a larger Cr and a smaller Cb than other clusters is located as the skin color region, among the plurality of clusters.
Part of the area in step (1) may be a skin color area extracted by the method of the present invention described above.
The target image may be obtained by converting from the RGB space to the YCbCr space.
The number of the plurality of clusters is, for example, two.
[0009]
Using the structure information of the flesh-colored area extracted by the above-described extraction method of the present invention, a flesh-colored area that is a face area can be extracted.
[0010]
The method for extracting a skin color region according to the present invention may include the following steps.
(1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
(2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
(3) generating a threshold based on a value in the Gaussian model corresponding to the color difference signal at a pixel in a target image; and
(4) extracting a skin color region from the target image by selecting the pixels included in the range determined by the threshold value from the target image.
[0011]
A computer program according to the present invention is for causing a computer to execute the steps in any of the above-described extraction methods.
The recording medium according to the present invention records the computer program and is readable by a computer.
[0012]
The skin color region extraction device according to the present invention includes a threshold value generation unit and a skin region candidate extraction unit, and the threshold value generation unit executes the following steps (1) to (3). The skin region candidate extraction unit performs the following step (4).
(1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
(2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
(3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting the threshold value candidate when the amount is substantially minimum as a threshold value;
(4) extracting a skin color region from the target image by selecting, from the target image, pixels having the one-dimensional information included in the range determined by the threshold value.
[0013]
An apparatus for extracting a skin color region according to the present invention includes a threshold value generation unit that generates a threshold value for determining a pixel in a target image, wherein the threshold value generation unit includes the following (1) to (3). May be performed.
(1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
(2) approximating the histogram by a Gaussian mixture model having a plurality of Gaussian models; and
(3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting the threshold value candidate when the amount is substantially minimum as a threshold value.
[0014]
The skin color region extraction device according to the present invention may be configured to include a face region candidate extraction unit that executes the following steps.
(1) decomposing each pixel in the entire or partial region of the target image into a plurality of clusters using the values of Cb and Cr in each pixel; and
(2) A step in which a region where a pixel belonging to a cluster having a larger Cr and a smaller Cb than other clusters is located as the skin color region, among the plurality of clusters.
[0015]
The extraction device may further include a face area extraction unit. The face area extraction unit extracts a skin color area, which is a face area, using the structure information of the skin color area.
[0016]
BEST MODE FOR CARRYING OUT THE INVENTION
A method for extracting a skin color region according to an embodiment of the present invention will be described with reference to the accompanying drawings. First, the configuration of an apparatus used in the extraction method of the present embodiment will be described with reference to FIG. In this specification, as the concept of the term, a skin color region includes a skin region candidate, a skin region candidate includes a face region candidate (that is, a skin region), and a face region candidate includes a face region.
[0017]
This device includes an image information acquisition unit 1, a color space conversion unit 2, a threshold generation unit 3, a skin region candidate extraction unit 4, a face region candidate extraction unit 5, and a face region extraction unit 6. ing. The image information acquisition unit 1 is a unit that acquires information of a target image as input. The target image information is usually an RGB signal in an RGB space. The color space conversion unit 2 is a part that converts the target image into a color space having a luminance signal and two color difference signals. Any color space may be used as long as it has two color difference signals. In the present embodiment, an example will be described in which an HSV space is used as such a color space and H (hue) and S (saturation) are used as two color difference signals. As a color space having two color difference signals, in addition to HSV, CIE LAB (L*a*b*), CIE LUV (L*u*v*) Or [Y, RY, BY], [Y, I, Q], etc. can be used.
[0018]
The threshold value generation unit 3 generates a threshold value for distinguishing pixels constituting the image information into those in a flesh color region and those not in a skin color region. The method of generating this threshold will be described later.
[0019]
The skin region candidate extraction unit 4 executes a function of extracting a skin color region from the target image by selecting a pixel included in the range defined by the threshold from the target image. Details of this operation will also be described later.
[0020]
The face area candidate extracting section 5 narrows down the skin area candidates obtained by the skin area candidate extracting section 4 and distinguishes between the face area and the hair area. The method will be described later.
[0021]
The face area extraction unit 6 extracts (selects) a face area from the face area candidates (that is, skin areas) obtained by the face area candidate extraction unit 5 based on structural features. A specific extraction method will be described later.
[0022]
Next, a method for extracting a flesh-colored area in the present embodiment will be described. First, the image information acquisition unit 1 acquires a target image as an input. In this embodiment, it is assumed that the target image is represented by an RGB signal in an RGB space. Next, the color space conversion unit 2 converts the RGB signals into HSV signals. Thus, two color difference signals (H, S) can be generated from the target image. The conversion formula from the RGB space to the HSV space is well known, but is shown below just in case.
[0023]
Next, the skin region candidate extraction unit 4 extracts a skin region candidate using the target image as the HSV signal. In the present embodiment, a threshold is generated by the threshold generator 3 before the extraction. Thus, first, a method of generating a threshold will be described with reference to FIG.
[0024]
(Generate threshold)
First, a plurality of sample images in the RGB space are obtained (step 2-1). An image including a human face is used as the sample image. In addition, it is preferable that conditions such as race, age, gender, degree of sunburn, and shading are as diverse as possible in the face included in the sample image. Next, the RGB signals are converted into HSV signals (step 2-2). This conversion formula is the same as that described above.
[0025]
Next, pixels in the face area (that is, the skin color area) are selected from the sample images (step 2-3). In the sample image, the face area is known.
[0026]
Next, for each pixel in the face area in the sample image, a histogram is generated for two color difference signals, that is, hue (H) and saturation (S) (step 2-4). FIG. 3 shows an example of the histogram. 3 (b) and 3 (c) show the graph of FIG. 3 (a) in one dimension only with hue or saturation.
[0027]
Next, the histogram is approximated by a Gaussian mixture model (step 2-5). The Gaussian mixture model is expressed as follows.
here,
k: the number of Gaussian model mixtures (4 in this embodiment),
θ: Parameter α corresponding to one Gaussian modeli, ΜiEtc. (4 sets in this embodiment),
It is.
[0028]
FIG. 4 shows the result of approximation of the histogram with a Gaussian mixture model. In this example, the number of mixtures of the Gaussian model is four. After generating a Gaussian mixture model (see FIGS. 4B and 4C) for H or S, by combining them, a three-dimensional Gaussian mixture model as shown in FIG. 4A can be obtained. .
The Gaussian mixture model is, for example,
(1) S. McKenna, S.M. Gong, and Y. Raja. Modeling Facial Color and Identity with Gaussian Mixtures. Pattern Recognition, 31 (12): 1883-1892, 1998.
(2) H. P. Graf, E .; Cosato, D.M. Gibbon, M .; Kocheisen, and E.A. Petajan. Multimodel System for Locating Heads and Faces. In Processeds of the Second IEEE International Conference on Automatic Face and Gesture Recognition, pages 88-93, 1996.
, Is well known and will not be described in further detail.
[0029]
Then, each pixel x in the target imageiAre converted into one-dimensional information using the corresponding values in the Gaussian model (step 2-6). Specifically, pixel xiIs the largest value φ in the Gaussian mixture model for the set of color difference signals (H, S) atiAsk for. This φiAre the one-dimensional information in this embodiment. φiBecomes a frequency value if the value of the histogram is used as it is. Where φiMay be another value obtained from the frequency value. For example, the value obtained by normalizing the frequency value with the maximum value of the histogram is φiYou may use as. In this case, φiRepresents the probability. φiMeans, in effect, a likelihood value indicating the probability of being flesh-colored, whether it is a frequency value or a normalized probability. Therefore, in this specification, φiMay be referred to as a likelihood value. Furthermore, this φi(I.e., a label of a Gaussian model) l (1 is an integer from 1 to k). In this embodiment, as described above, k = 4. Thereby, each pixel xiFor the likelihood value φiAnd a label l as information (step 2-7). Specifically, for example, φiAnd l and xiAnd store it in a table.
[0030]
Next, the threshold value candidate when the amount of increase in the number of pixels when the value of the threshold value candidate in the one-dimensional information is changed is substantially the minimum is set as the threshold value (step 2-8). This step is described in detail below with reference to FIGS.
[0031]
First, the pixel (group) x associated with the label 1 of the Gaussian modelithink of. Then these pixels xiLikelihood value φ foriMaximum value (or a value close to it) φi maxFrom ΔTlA threshold value T smaller byl(See FIG. 5). Next, the threshold candidate TlThe above likelihood value φiPixel x with label liCount the number of. For example, an example of the pixel at this time is as shown in FIG. Next, the threshold candidate TlTo ΔTlAnd then again, threshold candidate TlThe above likelihood value φiA pixel x with label liCount. For example, an example of the pixel at this time is as shown in FIG. ΔTlIs a constant amount in this embodiment. ΔTlOnly threshold candidate TlThe above calculation is performed while shifting. Thus, the one-dimensional information φi, The threshold candidate TlCan be obtained for each pixel. For example, the number of pixels in FIG.1, The number of pixels in FIG.2Then the increase is A2-A1Can be obtained as FIG. 5 shows the relationship between the threshold value candidate and the amount of increase in the pixel thus obtained.
[0032]
Next, the likelihood value (that is, the value of one-dimensional information) at which the amount of increase in the number of pixels is substantially minimized is set to10And In this embodiment, the threshold candidate TlBetween the threshold value and the pixel increase amount. As an equivalent method, a threshold candidate TlOf change ΔTlAnd the number of pixels included in the range, the threshold T10Can also be requested. In this case, the change amount ΔTlMeans the amount of increase in the number of pixels described above.
[0033]
In the present embodiment, after all, the value φ in the Gaussian model corresponding to the color difference signal (H, S) at the pixel in the target imagei, A threshold value has been generated.
[0034]
(Extraction of skin candidate area)
Next, using this threshold value, the skin region candidate extraction unit 4 (see FIG. 1) extracts a skin region candidate from the target image. This extraction has a threshold T10Pixel x having the label l and having one-dimensional information (likelihood value) included in the range determined byiIs selected from the target image.
[0035]
This extraction procedure is shown in FIG. First, the pixel x associated with the label li, A threshold T corresponding to the label llIs applied (step 7-1). Here, pixel xiIs the threshold TlBinarization is performed by outputting 1 if larger and 0 if smaller. Also, for example, the pixel xiIs the threshold TlIf it is larger, 1 instead of 1 can be output. This has the advantage that label information can be output simultaneously with binarization.
[0036]
Next, it is determined whether or not the binarization result is 1 (step 7-2). Of course, whether or not it is 0 may be determined. If it is 1, the pixel is set as a skin region candidate pixel (step 7-3). If it is not 1, it is not regarded as a skin region candidate (step 7-4). This determination is made for each pixel.
[0037]
According to the present embodiment, by selecting a pixel from the target image based on the threshold value, a skin color region (skin region candidate) can be extracted from the target image based on the pixel. Conventionally, since the threshold value for determining whether or not the image is a skin region is fixed, there is a problem that the detection accuracy is deteriorated when the condition in the target image changes. On the other hand, in the method of this embodiment, the threshold is adaptively generated from the target image as described above. For example, in this embodiment, if the target images are different, the threshold candidate TlOf change ΔTlAre different. Therefore, another threshold value can be determined for each target image corresponding to each Gaussian model l. As a result, even if the brightness of the target image fluctuates or a person with various skin colors exists, a threshold value can be generated according to those conditions. As a result, in this embodiment, it is possible to appropriately extract the skin color region.
[0038]
FIG. 8B shows the result of selecting the pixels in this manner. FIG. 3A shows a target image. In FIG. 8B, selected pixels are shown in white, and unselected pixels are shown in black.
[0039]
(Face area candidate extraction ... narrowing down)
Next, the face area candidate extraction unit 5 narrows down skin area candidates. The procedure for narrowing down will be described with reference to FIG. First, isolated pixels are removed by a median filter (not shown) (step 9-1). As the median filter, for example, a filter having a range of 3 × 3 pixels can be used. Thereby, noise can be removed from the image of the skin region candidate. FIG. 8C shows an example of the processing result. Next, labeling is performed to assign a label to each area (step 9-2). Next, the number of pixels is calculated for each region, and a region having a small area is removed (step 9-3). Specifically, for example, the area (for example, the number of pixels) in each region is obtained. Then, their average is calculated. Next, a region having an area smaller than a% (a is, for example, 3) of the average value is removed. FIG. 8D shows an example of the result of removing the small area.
[0040]
Next, morphological processing is performed on the region (step 9-4). The morphology processing is to remove a narrow recess and smooth the contour. FIG. 8E shows an example of the processing result. Next, a filling process is performed (step 9-5). The hole filling process is a process of filling a hole (corresponding to, for example, an eye or a nose hole) in an area. FIG. 8F shows an example of the processing result. FIG. 8G shows an image of the region extracted in this manner.
Since each of the processes shown in FIG. 9 is well known as itself, further description is omitted.
[0041]
Through these processes, the skin region candidates can be made into regions of a certain size or more (that is, narrowed down). In this embodiment, the image of the narrowed-down skin region candidate is stored as an RGB signal.
[0042]
(Extraction of face area candidate: separation of hair area and skin area)
Next, the face region candidate extraction unit 5 further separates the hair region and the skin region in the narrowed-down skin region candidate image. The separation procedure will be described with reference to FIG. The reason for performing such separation is that it is difficult to separate a hair region from a skin region by a separation method using hue and saturation. For example, in the image examples shown in FIGS. 11 (a), 12 (a) and 13 (a), when hue and saturation are used, when the above-described procedure is applied, FIG. 11 (b), FIG. An area as shown in FIG. 13B and FIG. 13B is extracted. In these figures, white lines around the face indicate regions. The relationship between hue and saturation in these images is shown in FIGS. 11 (c), 12 (c) and 13 (c). In the figure, the pixels of the hair part are indicated by crosses, and the pixels of the skin part are indicated by dots (in the figure, crushed). These results also show that it is difficult to separate hair and skin when using hue and saturation.
[0043]
Therefore, in this embodiment, first, the image of the skin region candidate in the RGB space is converted into an image in the YCbCr space (step 10-1). That is, the RGB signal is converted into the YCbCr signal. Since the conversion formula is well known, the description is omitted.
[0044]
Next, a histogram in Cb and Cr is generated for each pixel (step 10-2). For example, FIGS. 14 to 16 show examples of histograms in the images shown in FIGS. 11, 12, and 13. FIG. Here, the pixels of the hair portion are indicated by crosses, and the pixels of the skin portion are indicated by dots. As described above, by using the histograms of Cb and Cr, it can be decomposed into a plurality of (here, two) clusters.
[0045]
Next, one cluster is selected (step 10-3). As is clear from the examples of FIGS. 14 to 16, in general, the skin color region has a large Cr and a small Cb, and therefore a cluster having the property is selected.
[0046]
Next, pixels belonging to the selected cluster are extracted (step 10-4). Next, noise is removed by removing isolated pixels from the extracted pixels (step 10-5). Thus, face area candidates can be extracted (step 10-6).
[0047]
As described above, according to the present embodiment, since the face region and the hair region are classified using the clusters corresponding to the values of Cb and Cr, it is possible to accurately extract the candidate image of the face region. Become.
[0048]
(Extraction of face area)
In the above steps, a flesh color region was extracted based on the color information. However, it is difficult to exclude a skin region (for example, an arm) other than the face only by the above processing. Thus, the face area detection unit 6 extracts (selects) a face area. The procedure will be described with reference to FIG.
[0049]
First, it is determined whether or not the shape of the region is an ellipse within a predetermined range (step 17-1). That is, it is determined whether (short axis length / long axis length) is within a predetermined value range (for example, greater than 0.4). If it is out of the range, the area is not set as a face area (step 17-2).
[0050]
If the region shape is an ellipse within a predetermined range, it is determined whether or not both eyes are present (step 17-3). Here, first, an eye candidate area is obtained from the color information. Normally, using the above-described clustering in the YCbCr signal, the eye region can be distinguished from the skin color region in the same manner as the hair region. Further, the eye region can be separated from the hair region using structural information such as the size. It is determined whether or not the obtained eye region matches a predetermined rule. The rules are, for example, structural rules such as whether the eye region is above the short axis, whether the eye region is on a line almost parallel to the short axis, whether both eyes are not on one side of the long axis, etc. It is. In this manner, the presence / absence of an eye region is checked using the region extracted based on the color information as a candidate. If there is an eye area, the face area candidate is set as a face area (step 17-4).
[0051]
If the eye region cannot be detected based on the color information, in step 17-3, the eye region is further detected based on the luminance information. Here, a region with low luminance is extracted as a candidate, and the determination is made based on structural information such as the size and arrangement of the region. As a result, if the eye area can be detected, the face area candidate is set as a face area (step 17-4).
[0052]
If the eye area cannot be detected by the above procedure, the face area candidate is not set as a face area (step 17-2).
[0053]
Thus, according to the present embodiment, a face region can be extracted from a target image. Once the face region is determined, a face image can be obtained from the pixels included in the region.
[0054]
Execution of each of the above-described embodiments can be easily performed by those skilled in the art using a computer. The program for that can be stored in any computer-readable recording medium, for example, HD, FD, CD, MO and the like.
[0055]
The description of each of the above embodiments is merely an example, and does not show a configuration essential to the present invention. The configuration of each part is not limited to the above as long as the purpose of the present invention can be achieved.
Further, as specific means of each unit (including functional blocks) for realizing each embodiment, hardware, software, a network, a combination thereof, or any other means may be used. It is obvious to those skilled in the art. Further, the functional blocks may be combined into one functional block. Further, the function of one functional block may be realized by cooperation of a plurality of functional blocks.
[0056]
【The invention's effect】
According to the present invention, it is possible to provide a method capable of accurately extracting a flesh-tone area even when conditions of a target image are various.
[Brief description of the drawings]
FIG. 1 is a block diagram schematically showing an apparatus for implementing a skin color region extraction method according to an embodiment of the present invention.
FIG. 2 is a flowchart showing a procedure for generating a threshold value in the skin color region extraction method according to one embodiment of the present invention.
3A and 3B are graphs showing a histogram for each pixel with respect to hue (H) and saturation (S), wherein FIG. 3A is two-dimensional, FIG. 3B is one-dimensional only of hue, and FIG. Is a one-dimensional representation of only the saturation.
FIGS. 4A and 4B are graphs showing a state in which a histogram for each pixel regarding hue (H) and saturation (S) is approximated by a Gaussian mixture model, where FIG. 4A is two-dimensional and FIG. (C) is a one-dimensional representation of only the saturation.
FIG. 5 is a graph for explaining a threshold value generation method, in which the vertical axis represents the amount of pixel increase and the horizontal axis represents the threshold value (likelihood value).
6A and 6B are diagrams for explaining the number of pixels corresponding to threshold candidates. FIGS. 6A and 6B show examples of pixels having likelihood values equal to or greater than different threshold candidates. Is shown.
FIG. 7 is a flowchart illustrating a pixel extraction method based on a threshold.
8A and 8B are image examples for explaining a process of extracting a skin color region from a target image, wherein FIG. 8A shows a target image, and FIG. 8B shows a white region extracted using a threshold. , FIG. (C) is a diagram showing a region after being processed by a median filter, FIG. (D) is a diagram after removing a small region, FIG. (E) is a diagram after morphological processing, and FIG. () Shows the figure after filling in the holes, and FIG. (G) shows the image in the narrowed area.
FIG. 9 is a flowchart illustrating a procedure for narrowing down skin area candidates.
FIG. 10 is a flowchart illustrating a procedure for separating a hair part and a face part from each other.
11A and 11B are explanatory diagrams for explaining a procedure for separating a hair part and a face part from each other. FIG. 11A is an example of a target image, FIG. 11B is an extracted skin area candidate, and FIG. c) is a histogram of the pixels in the skin region candidate, in which the vertical axis represents the saturation and the horizontal axis represents the hue.
12A and 12B are explanatory diagrams for explaining a procedure for separating a hair part and a face part from each other. FIG. 12A is another example of a target image, FIG. 12B is an extracted skin area candidate, FIG. 7C is a histogram of the pixels in the skin region candidate, in which the vertical axis represents the saturation and the horizontal axis represents the hue.
13A and 13B are explanatory diagrams for explaining a procedure of separating a hair part and a face part from each other, wherein FIG. 13A shows still another example of the target image, and FIG. (C) is a histogram of pixels in the skin region candidate, in which the vertical axis represents saturation and the horizontal axis represents hue.
14 is a histogram of pixels in the skin region candidate shown in FIG. 11, in which the vertical axis is Cr and the horizontal axis is Cb.
15 is a histogram of the pixels in the skin region candidate shown in FIG. 12, where the vertical axis is Cr and the horizontal axis is Cb.
FIG. 16 is a histogram of pixels in the skin region candidate shown in FIG. 13, where the vertical axis is Cr and the horizontal axis is Cb.
FIG. 17 is a flowchart illustrating a procedure for extracting a face area from face area candidates.

Claims (15)

  1. A method for extracting a skin color region, comprising the following steps:
    (1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
    (2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
    (3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting the value of the one-dimensional information whose amount is substantially minimum as a threshold value;
    (4) extracting a skin color region from the target image by selecting, from the target image, pixels having the one-dimensional information included in the range determined by the threshold value.
  2. The extraction method according to claim 1, wherein the one-dimensional information is a value in the Gaussian model, and a value corresponding to two color difference signals in a pixel of the target image is a maximum. .
  3. The increasing amount of the number of pixels in the step (3) is a value of a pixel converted into one-dimensional information by the same Gaussian model,
    The threshold is associated with information indicating the Gaussian model,
    The pixel selected from the target image in the step (4) is selected from pixels converted into one-dimensional information by the same Gaussian model according to a threshold value associated with the Gaussian model. The extraction method according to claim 1.
  4. A method for extracting a skin color region, comprising the following steps:
    (1) decomposing each pixel in all or a part of the target image into a plurality of clusters using the values of Cb and Cr in the pixel;
    (2) A step in which a region where a pixel belonging to a cluster having a larger Cr and a smaller Cb than other clusters is located as the skin color region, among the plurality of clusters.
  5. The extraction method according to claim 4, wherein the partial area is a skin color area extracted by the extraction method according to any one of claims 1 to 3.
  6. The method according to claim 4, wherein the target image is obtained by converting an RGB space into a YCbCr space.
  7. The method according to claim 4, wherein the number of the plurality of clusters is two.
  8. A method for extracting a skin-colored area, which is a face area, using the structure information of the skin-colored area extracted by the extraction method according to claim 1.
  9. A method for extracting a skin color region, comprising the following steps:
    (1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
    (2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
    (3) generating a threshold based on a value in the Gaussian model corresponding to the color difference signal in a pixel in the target image;
    (4) extracting a skin color region from the target image by selecting the pixels included in the range determined by the threshold value from the target image.
  10. A computer program for executing the steps according to claim 1 on a computer.
  11. A computer-readable recording medium on which the computer program according to claim 10 is recorded.
  12. A threshold value generation unit; and a skin region candidate extraction unit. The threshold value generation unit executes the following steps (1) to (3). An apparatus for extracting a flesh-colored area, characterized by performing the step of 4):
    (1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
    (2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
    (3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting the threshold value candidate when the amount is substantially minimum as a threshold value;
    (4) extracting a skin color region from the target image by selecting, from the target image, pixels having the one-dimensional information included in the range determined by the threshold value.
  13. A threshold generation unit that generates a threshold for determining a pixel in the target image, wherein the threshold generation unit executes the following steps (1) to (3); Skin color region extraction device:
    (1) For each pixel in the skin color region in the sample image, generating a histogram for two color difference signals;
    (2) approximating the histogram with a Gaussian mixture model having a plurality of Gaussian models;
    (3) The two color difference signals of the pixels in the target image are converted into one-dimensional information using the corresponding values in the Gaussian model, and then the number of pixels increases when the value of the one-dimensional information is changed Setting the threshold value candidate when the amount is substantially minimum as a threshold value.
  14. An apparatus for extracting a skin color region, comprising a face region candidate extraction unit that executes the following steps:
    (1) decomposing each pixel in all or a part of the target image into a plurality of clusters using the values of Cb and Cr in each pixel;
    (2) A step in which a region where a pixel belonging to a cluster having a larger Cr and a smaller Cb than other clusters is located as the skin color region, among the plurality of clusters.
  15. The extraction device according to any one of claims 12 to 14, further comprising a face region extraction unit, wherein the face region extraction unit extracts a skin color region, which is a face region, using structure information in the skin color region. An extraction device characterized by performing the following steps.
JP2003032864A 2003-02-10 2003-02-10 Method for extracting skin color area Pending JP2004246424A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2003032864A JP2004246424A (en) 2003-02-10 2003-02-10 Method for extracting skin color area

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP2003032864A JP2004246424A (en) 2003-02-10 2003-02-10 Method for extracting skin color area

Publications (1)

Publication Number Publication Date
JP2004246424A true JP2004246424A (en) 2004-09-02

Family

ID=33019062

Family Applications (1)

Application Number Title Priority Date Filing Date
JP2003032864A Pending JP2004246424A (en) 2003-02-10 2003-02-10 Method for extracting skin color area

Country Status (1)

Country Link
JP (1) JP2004246424A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2006313468A (en) * 2005-05-09 2006-11-16 Sony Corp Image processor and method, recording medium and program
KR100680278B1 (en) 2005-12-28 2007-02-07 고려대학교 산학협력단 Method for lip shape extraction and apparatus thereof
JP2007221408A (en) * 2006-02-16 2007-08-30 Victor Co Of Japan Ltd Color-difference signal correcting device
JP2007257087A (en) * 2006-03-20 2007-10-04 Univ Of Electro-Communications Skin color area detecting device and skin color area detecting method
JP2007336538A (en) * 2006-06-09 2007-12-27 Lg Innotek Co Ltd Camera module and mobile communication terminal having the same
JP2010055468A (en) * 2008-08-29 2010-03-11 Nikon Corp Image processing apparatus, and camera
JP2016061692A (en) * 2014-09-18 2016-04-25 株式会社メガチップス Human body detection device
CN107633252A (en) * 2017-09-19 2018-01-26 广州市百果园信息技术有限公司 Skin color detection method, device and storage medium
JP2018049564A (en) * 2016-09-23 2018-03-29 カシオ計算機株式会社 Detection device and detection method

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2006313468A (en) * 2005-05-09 2006-11-16 Sony Corp Image processor and method, recording medium and program
KR100680278B1 (en) 2005-12-28 2007-02-07 고려대학교 산학협력단 Method for lip shape extraction and apparatus thereof
JP2007221408A (en) * 2006-02-16 2007-08-30 Victor Co Of Japan Ltd Color-difference signal correcting device
JP4715545B2 (en) * 2006-02-16 2011-07-06 日本ビクター株式会社 Color difference signal correction device
JP2007257087A (en) * 2006-03-20 2007-10-04 Univ Of Electro-Communications Skin color area detecting device and skin color area detecting method
JP2007336538A (en) * 2006-06-09 2007-12-27 Lg Innotek Co Ltd Camera module and mobile communication terminal having the same
JP2010055468A (en) * 2008-08-29 2010-03-11 Nikon Corp Image processing apparatus, and camera
JP2016061692A (en) * 2014-09-18 2016-04-25 株式会社メガチップス Human body detection device
JP2018049564A (en) * 2016-09-23 2018-03-29 カシオ計算機株式会社 Detection device and detection method
CN107871309A (en) * 2016-09-23 2018-04-03 卡西欧计算机株式会社 Detection method and detection means
US10621754B2 (en) 2016-09-23 2020-04-14 Casio Computer Co., Ltd. Method of detecting skin color area of human
CN107633252A (en) * 2017-09-19 2018-01-26 广州市百果园信息技术有限公司 Skin color detection method, device and storage medium
CN107633252B (en) * 2017-09-19 2020-04-21 广州市百果园信息技术有限公司 Skin color detection method, device and storage medium

Similar Documents

Publication Publication Date Title
JP2004246424A (en) Method for extracting skin color area
EP1434164B1 (en) Method of extracting teeth area from teeth image and personal identification method and apparatus using the teeth image
JP2010160793A (en) Method, system, and device for red-eye detection, computer readable medium, and image processing device
KR101875891B1 (en) apparatus and method for face detection using multi detection
US20180228426A1 (en) Image Processing System and Method
KR20150007880A (en) method and apparatus for improving quality of image and recording medium thereof
KR101853006B1 (en) Recognition of Face through Detecting Nose in Depth Image
JP4423333B2 (en) Background area specifying method, background area specifying system, background color determining method, control program, and recording medium
JP4158153B2 (en) Face image detection method
JP3902887B2 (en) Lip extraction method
Youlian et al. Face detection method using template feature and skin color feature in rgb color space
Das et al. Human face detection in color images using HSV color histogram and WLD
Garg et al. Color based segmentation using K-mean clustering and watershed segmentation
JP5979008B2 (en) Image processing apparatus, image processing method, and program
Felhi et al. Multiscale stroke-based page segmentation approach
KR20160017152A (en) Gender Classification Based on Binary Haar Cascade
CN110084135A (en) Face identification method, device, computer equipment and storage medium
Patil et al. A review paper on facial detection technique using pixel and color segmentation
Sun et al. Lip segmentation based on facial complexion template
KR100927642B1 (en) A face edge extraction method using visual division histogram analysis
Ojo et al. Illumination invariant face detection using hybrid skin segmentation method
Prakash et al. An identification of abnormalities in dental with support vector machine using image processing
Saranya et al. An approach towards ear feature extraction for human identification
JP2001331803A (en) Method for extracting character area in image
Hussain et al. Human Face Detection with skin color properties

Legal Events

Date Code Title Description
A621 Written request for application examination

Free format text: JAPANESE INTERMEDIATE CODE: A621

Effective date: 20060111

A711 Notification of change in applicant

Free format text: JAPANESE INTERMEDIATE CODE: A711

Effective date: 20060111

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A821

Effective date: 20060111

A977 Report on retrieval

Free format text: JAPANESE INTERMEDIATE CODE: A971007

Effective date: 20090317

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20090406

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20090601

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20090601

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20090706

A02 Decision of refusal

Free format text: JAPANESE INTERMEDIATE CODE: A02

Effective date: 20091104