JP2005208760A - Person image extraction device and computer program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a person image extraction device and a computer program, extracting a person image from a person image including a background image by minimum operation. <P>SOLUTION: This person image extraction device has: a means for acquiring coordinates of points showing a jaw part and a parietal part of the person image; a means calculating dispersion of a pixel value of a pixel positioned inside a rectangular area including the acquired coordinates showing the parietal point inside a color space; a means deciding whether the calculated dispersion of the pixel value is smaller than a threshold or not; a means averaging the color space when the dispersion of the pixel value is smaller than the threshold; a contour extraction means extracting a contour of the person image as a portion wherein a degree of a change of the pixel value is larger than the threshold; and a background image removal part removing the background image. The background image removal part stores a multi-valued template copying shape of a person, fits the coordinates of the jaw part and the parietal part of the person image to a face portion of the multi-valued template, and generates a composition image of the person image and the multi-valued template. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a person image extraction apparatus and a computer program for extracting a person image by erasing the background from a person image taken together with a general background image showing an obstacle.

  Conventionally, for taking photographs of licenses, passports, portrait photographs for various certificates, funeral remains, portrait photographs for criminal investigation, etc., the background is white or blue and the size originally required Take a picture of a large area and crop it so that the face of the person meets the required conditions, or cut the printed photo to the required size, then portrait of the required size I was making a photo.

  The conventional portrait photography method described above requires a dedicated screen and appropriate lighting to obtain a white or blue background, and additionally trims the printed photo to meet the required conditions. It was necessary to cut, and it took time and effort.

  On the other hand, various image processing software has been developed along with the recent improvement and popularization of personal computers, and the user designates an arbitrary area, for example, a whole image of a person, a face, etc. from one image. Many softwares have been developed that have the function of extracting and synthesizing them with other images, such as landscape images. However, it is difficult to automatically extract an arbitrary area from a single image, and software that requires the user to trace the boundary of the image with a pointing device such as a mouse is the mainstream. Even software that says that it is possible to carry out contouring is actually a software that requires a contour instruction by a user's manual operation.

  In order to cope with such a problem, for example, in Patent Document 1, although the background image needs to be white or blue, by specifying the top of the person, the position of the eyes, and the position of the chin, An automatic face image processing apparatus is disclosed in which it is not necessary to trace the outline of a person with a pointing device such as a mouse.

  In Patent Document 2, foreground image extraction is performed in which only a background image without a person is captured, and a person is imaged on the same background, thereby effectively eliminating the background image and automatically trimming the outline of the person. A method is disclosed.

In Patent Document 3 or Patent Document 4, a face image is automatically extracted by extracting an image centered on a color, particularly a skin color that is the color of a human face, from a person image shown in a general background image. A face image extraction method that can be extracted is disclosed.
JP 2001-005977 A JP 2000-036032 A JP 9-0050528 A JP-A-10-334213

  The conventional techniques described above have the following problems. In other words, the face image processing method disclosed in Patent Document 1 eventually requires dedicated equipment (screen, lighting, etc.) to take a portrait photograph, and the head and eye positions of the person in the image. It is necessary to specify the position of the chin. Further, since the target person image is required to be brighter than the background image and to have a uniform background, it may be difficult to extract the person image depending on the state of the background image including an obstacle. There was a problem that there was.

  Further, in the foreground image extraction method disclosed in Patent Document 2, it is possible to extract the upper body of a person as a portrait image as well as a face image, but it is necessary to capture a background image in advance. In addition, there is a problem in that it is not easy to photograph a person with the same background image.

  Further, in the face image extraction method disclosed in Patent Document 3 or Patent Document 4, only a person's face image can be extracted, and it is difficult to extract the upper body of a person as a portrait image. there were.

  The present invention has been made in view of such circumstances, and extracts a person image by erasing the background image from a person image taken together with a general background image showing an obstacle with minimal operations. An object of the present invention is to provide a person image extraction device and a computer program.

  In order to achieve the above object, a human image extracting apparatus according to a first aspect of the present invention is a human image extracting apparatus that extracts a human image taken in the foreground of the image, and indicates the top and chin of the human image. Means for acquiring the coordinates, means for calculating the dispersion in the color space of the pixel values of the pixels located in the rectangular area including the coordinates indicating the acquired top, and the dispersion of the calculated pixel values from the threshold Means for determining whether or not the variance of the pixel values is smaller than a threshold value, and means for averaging the color space and a change in the pixel value of the pixel adjacent to the contour of the person image. A contour extraction unit that extracts a portion having a degree greater than a threshold; and a background image removal unit that removes a background image. The background image removal unit stores a multi-value template imitating the shape of a person; The top of the person image and The coordinates of the jaws is adapted to the face portion of the multi-value template, characterized in that are no to produce a composite image of the said person image and the multi-value template.

  In the person image extraction device according to the first aspect of the invention, only the coordinates of the top and chin of the person image to be extracted are acquired, and the color of the pixel value of the pixel located in the rectangular area including the coordinates indicating the acquired top By calculating the dispersion in the space, the color space can be averaged as necessary, and the head portion in the person image can be clearly specified, and the multi-value template can be used for the top and chin of the person image. The region to be extracted as a human image is clarified by stepwise gradation display. This makes it possible to reliably extract a person image by specifying only the top and chin of the person image, regardless of the type and state of the background image, and for licenses, passports, and various certificates. Portrait photos, funeral remains, and criminal investigation portrait photos.

  In the human image extraction device according to a second aspect of the present invention, in the first aspect, the background image removal unit distributes pixel values of pixels located in a rectangular area including coordinates indicating the top of the head in a color space. The hair portion is extracted based on the above, and the boundary between the hair image and the background image is specified.

  In the human image extracting apparatus according to the second aspect of the invention, based on the coordinates of the top of the human image to be extracted, the hair portion that has been difficult to extract by the conventional face image extracting method based on the color is reliably extracted. Can do. This makes it possible to specify the hair portion of a person image regardless of the type and state of the background image, and for licenses, passports, portrait photographs for various certificates, funeral remains, and criminal investigations. It is possible to create portrait photographs with high accuracy.

  In the human image extraction device according to a third aspect of the present invention, in the first or second aspect, the background image removal unit obtains an average value of pixel values of the facial image based on the coordinates acquired as the jaw. The boundary between the neck portion of the person image and the background image is specified by determining whether or not the difference in pixel value from the obtained average value is larger than a predetermined threshold value. .

  In the human image extracting apparatus according to the third aspect of the invention, based on the coordinates of the top of the human image to be extracted, the portion around the neck that is difficult to extract by the conventional face image extracting method based on the color is surely extracted. be able to. This makes it possible to specify the area around the neck of a person image regardless of the type and state of the background image, and for licenses, passports, portrait photographs for various certificates, funeral remains, and criminal investigations. It is possible to create a portrait photograph of the person with high accuracy.

  In the human image extraction device according to a fourth aspect of the present invention, in the first through third aspects, the background image removal unit extracts a contour line in the vicinity of the shoulder of the human image, and the inclination of the contour line is within a predetermined range. It is characterized in that an outline that is not inside is erased.

  In the human image extracting apparatus according to the fourth aspect of the present invention, based on the coordinates of the top of the human image to be extracted, obstacles in the vicinity of the shoulder that are difficult to erase by the conventional face image extracting method based on the color are included. Unnecessary edges generated based on the contour can be surely erased. As a result, it is possible to securely erase unnecessary edges on the shoulder portion of a person image regardless of the type and state of the background image, and for licenses, passports, portrait photographs for various certificates, funeral use This makes it possible to accurately create portraits for ashes and criminal investigations.

  Next, in order to achieve the above object, a computer program according to a fifth aspect of the present invention is a computer program for causing a computer to execute a process of extracting a person image photographed in the foreground of the image. A step of acquiring coordinates of a point indicating a part, a step of calculating a dispersion in a color space of pixel values of pixels located in a rectangular area including the acquired coordinates indicating the top of the head, and Determining whether the variance is less than a threshold; if determining that the variance of pixel values is less than the threshold; averaging the color space; and changing a pixel value of a pixel adjacent to the contour of a person image And extracting the background image as a portion having a degree greater than a threshold, and removing the background image. The multi-value template stored is stored, the coordinates of the top and chin of the person image are adapted to the face portion of the multi-value template, and a composite image of the person image and the multi-value template is generated. It is characterized by being.

  In the computer program according to the fifth aspect of the present invention, only the coordinates of the top and chin of the person image to be extracted are acquired, and within the color space of the pixel value of the pixel located in the rectangular region including the coordinates indicating the acquired top By calculating the variance at, the color space can be averaged as necessary to clearly identify the head portion of the person image, and the multi-value template can be used for the coordinates of the top and chin of the person image. The region to be extracted as a human image is clarified by stepwise gradation display. This makes it possible to reliably extract a person image by specifying only the top and chin of the person image, regardless of the type and state of the background image, and for licenses, passports, and various certificates. Portrait photos, funeral remains, and criminal investigation portrait photos.

  According to the first and fifth aspects of the invention, the person image can be reliably extracted by specifying only the top and chin of the person image, regardless of the type and state of the background image. Certificates, passports, portrait photographs for various certificates, funeral remains, and portrait photographs for criminal investigations can be created.

  In addition, according to the second invention, it is possible to specify the hair portion of the person image regardless of the type and state of the background image, and to obtain a license, passport, portrait photograph for various certificates, funeral use This makes it possible to accurately create portraits for ashes and criminal investigations.

  In addition, according to the third invention, it is possible to specify a portion around the neck of the person image regardless of the type and state of the background image, and to obtain a license, passport, portrait photograph for various certificates, funeral use It is possible to accurately create portraits for crimes and criminal investigations.

  Further, according to the fourth invention, an unnecessary edge in the shoulder portion of the person image can be surely erased regardless of the type and state of the background image, and the license image, passport and various certificates can be used. It is possible to accurately create portrait photographs, funeral remains, and criminal investigation portrait photographs.

  Hereinafter, the present invention will be described in detail with reference to the drawings illustrating embodiments thereof. FIG. 1 is a schematic configuration diagram of a computer that embodies a human image extraction apparatus according to an embodiment of the present invention. As shown in FIG. 1, a computer that embodies a person image extracting apparatus includes at least a CPU (central processing unit) 11, a storage unit 12, a RAM (memory) 13, a communication unit 14 connected to an external communication unit, a mouse. , An input means 15 such as a keyboard, a display means 16 such as a monitor, an image input means 17 such as a camera and a CCD, and an auxiliary storage means 18.

  The CPU 11 is connected to the above-described hardware units via a bus, controls the above-described hardware units, and executes various software functions according to programs stored in the storage unit 12. The image input unit 16 captures an image using, for example, a CCD, and inputs the captured image data.

  The display means 16 is a display device such as a liquid crystal display device or a CRT display, and displays captured image data or image data extracted as a person image.

  The auxiliary storage means 18 is a portable recording medium 19 in which a program used by a computer that embodies the person image extraction apparatus is recorded, and corresponds to a DVD, a CD-ROM, or the like. Further, a portable recording medium 19 for recording data used by a person image extraction device such as multi-value template data used for background removal of a person image is also included.

  FIG. 2 is a flowchart of pre-processing in CPU 11 of the person image extraction apparatus according to the embodiment of the present invention. In FIG. 2, an image input means 17 such as a camera captures a human image including a background image, and inputs the human image including the captured background image as image data. When image data is input, the CPU 11 stores the input image data in the person image storage unit 121 of the storage unit 12 (step S201).

  Note that the person image captured by the image input means 17 is usually represented in the RGB color space. The color space is not limited to the RGB color space, but may be an HSV color space, a CIELAB (L * a * b) color space, or the like.

  The CPU 11 reads out the image data stored in the person image storage unit 121 and displays it on the display unit 16 (step S202). The user designates the crown and chin in the human image including the background image displayed on the screen by using a pointing device such as a mouse or a tablet as the input unit 15. FIG. 3 is a view showing an example of a person image including a background image displayed on the display means 16. As indicated by the circles in FIG. 3, when the top 31 and the jaw 32 are designated on the screen on the human image including the background image, the CPU 11 displays the display coordinate value of the point selected on the screen. Is calculated and stored in the RAM 13 (step S203).

  CPU11 sets the predetermined | prescribed rectangular area surrounding the display coordinate value of the point designated as a top part, and each R value, G value, or B value is set about all the pixels located in the coordinate corresponded in this rectangular area. Find the average value of. The CPU 11 compares the obtained average value of the R value, G value, or B value for each pixel in the color space, and calculates the variance of the R value, G value, or B value in the color space (step S204). ).

  Based on the variance of the R value, G value, or B value in the calculated color space, the CPU 11 determines that it is difficult to discriminate between the background image and the top of the person image, for example, for a portion that is judged to have a small variance. Judgment can be made. As a specific example, this corresponds to the case where the background image is white while white hair is present at the top of the person image.

  Therefore, when the CPU 11 determines that the variance of the calculated R value, G value, or B value in the color space is smaller than a predetermined threshold (step S205: YES), the CPU 11 averages the color space. (Step S206), and conversion is performed so that a color-sensitive difference is likely to occur between the background image and the top of the person image.

  FIG. 4 is a diagram showing an overview of the averaging process regarding the lightness in the color space. The CPU 11 averages the color space in which the lightness frequency distribution is not uniform as shown in FIG. 4A and the color space so that the lightness frequency distribution is uniform as shown in FIG. Stretch). By performing the same processing not only for lightness but also for saturation and hue, the color band can be extended (stretched), and it is easy to extract the contour at the top of the person image in the contour extraction processing described later. At the same time, it is possible to extract a contour line as a head portion of a person image even for a hair portion including gray hair, which has been difficult in the past.

  Next, the CPU 11 smoothes a portion in the person image that has a gentle color change as an area of the same color (step S207). FIG. 5 is an explanatory diagram of the principle of the smoothing process. For all the pixels of the acquired human image, a difference in R value, G value, or B value with a pixel located in the vicinity of a predetermined pixel, for example, within 10 pixels in the vertical and horizontal directions is obtained. When the obtained difference is equal to or less than a predetermined threshold, the R value, G value, or B value of the predetermined pixel is set to the average value of the R value, G value, or B value with the pixel located in the vicinity of the predetermined pixel. Replace with. In FIG. 5, since the R value, G value, or B value changes gently in the skin color region located on the left, the skin color region is replaced with a substantially uniform R value, G value, or B value. However, in the vicinity of the outline 51, the color changes greatly to the color of the background image, so the R value, G value, or B value is not replaced with the average value, and the R value of the pixel in which the outline 51 exists , G value or B value protrudes compared to R value, G value, or B value of adjacent pixels, and as a result, the contour line 51 is emphasized.

  After performing the above preprocessing, the CPU 11 performs contour extraction processing for extracting a portion where the R value, G value, or B value of the pixel changes rapidly as the contour of the person image. FIG. 6 is a flowchart of the contour extraction process in the CPU 11 of the person image extraction apparatus according to the embodiment of the present invention. The contour extraction process shown in FIG. 6 is a process for emphasizing coordinates at which the R value, G value, or B value of a pixel changes rapidly using a general filter, for example, a sobel filter.

  FIG. 7 is a view showing an example of a sobel filter used for the contour extraction process. 7A shows the filter matrix parameter fx in the x-axis direction of the screen, and FIG. 7B shows the filter matrix parameter fy in the y-axis direction of the screen.

  The CPU 11 converts the R value, the G value, or the B value into an R ′ value, a G ′ value, or a B ′ value as shown in (Equation 1) using the filter matrix value shown in FIG. S601). In (Equation 1), Rin, Gin, or Bin indicates the R value, G value, or B value of the pixel, respectively.

(Equation 1)
R ′ = 255− | fx (Rin) + fy (Rin) |
G ′ = 255− | fx (Gin) + fy (Gin) |
B ′ = 255− | fx (Bin) + fy (Bin) |

  Since the sobel filter is a filter that expresses the contour line in white with black as the tone, in (Equation 1), the black and white are inverted to improve visibility, and the contour line is expressed in black with white as the tone. Yes.

  FIG. 8 is a view showing an example of the human image after the outline is extracted by applying the sobel filter. In FIG. 8, although the contour line is emphasized, noise or the like during the preprocessing is left on the person image. In the conventional contour extraction process, the contour is binarized at this point. Although there is a merit that the labeling of the contour line becomes easy by performing the binarization process, the computer load increases. Therefore, in the present invention, the binarization process is not performed and the background image is removed by the multi-value template. Make it easy.

  The CPU 11 applies the multi-value template of the person image stored in advance in the storage unit 12 to the person image from which the contour line shown in FIG. 8 is extracted. That is, the person image is enlarged so that the head and jaw portions of the subject image that has been subjected to the target preprocessing match the positions indicating the crown and jaw portions in the multi-value template stored in the storage unit 12. Or reduce. And CPU11 performs the process which emphasizes the outline of a person image and weakens the outline of a background image by performing the calculation process using a multi-value template.

  FIG. 9 is an illustration of a multi-value template applied to a person image. As shown in FIG. 9, the central portion of the human image, that is, the highly accurate portion that is the human image is white, the peripheral portion that is far from the human image, that is, the high-accuracy portion that is the background image, is black, and the intermediate portion is This is stored in the multi-value template storage unit 122 of the storage unit 12 as a template generated as sequential gradation data.

  When the multi-value template stored in the multi-value template storage unit 122 is, for example, a 256-level gray scale multi-value template, black is represented as 0, white is represented as 255, and intermediate gray is represented as 128. The multi-value template is applied to a human image in which the outline is expressed in black with the above-described white as a base tone.

  Specifically, the CPU 11 enlarges / reduces the person image so that the coordinates of the top and chin of the person image match the multi-value template (step S602). When the coordinates of the top and chin match the multi-value template, the template value is T, and the R ′ value, B ′ value, or G ′ value of each pixel is R ″ value as shown in (Equation 2). , G ″ value or B ″ value (step S603). In (Expression 2), max () and min () indicate functions for obtaining the maximum value and the minimum value, respectively. The multi-value template may be enlarged or reduced so as to match the coordinates of the top and chin of the person image.

(Equation 2)
R ″ = max (0, min (255, R′−T + 128)
G ″ = max (0, min (255, G′−T + 128)
B ″ = max (0, min (255, B′−T + 128)

  By (Equation 2), the portion where the multi-value template is black is expressed in white and the contour can be removed, and the portion where the multi-value template is white is expressed in black and can be extracted as the contour. . Therefore, it is possible to surely remove the black portion of the multi-value template with high accuracy that is the background image, and reliably leave the contour line portion of the white portion of the multi-value template with high accuracy that is the human image. It becomes possible. FIG. 10 is a view showing an example of a person image after the multi-value template is applied. As shown in FIG. 10, it is possible to effectively erase noise and the like remaining in the entire image.

  Next, the CPU 11 performs a conversion process from the RGB color space to the HSV color space (step S604). Then, the degree of color change is calculated for each pixel (step S605), and when it is determined that the degree of color change is uniform in the vicinity of a predetermined pixel (step S606: YES), pixels in the vicinity of the pixel Is replaced with a uniform pixel value as a background image, for example, a pixel indicating gray (step S607). FIG. 11 is an explanatory diagram of the outline of the background image painting process at the periphery of the person image. As shown in FIG. 11, the CPU 11 can determine that everything above the coordinates of the top is a background image, and can also determine that the outside of the extracted outline is also a background image. Therefore, based on the coordinates of the top 31 and the chin 32, a rectangular area 111 containing the person image subjected to the contour extraction process is set, and the degree of color change in the rectangular area 111 is calculated for each pixel. Pixels having a uniform degree of change are replaced with pixels having uniform pixel values. This makes it possible to eliminate most of the noise around the person image.

  Hereinafter, the CPU 11 specifies a clearer outline of the person image by performing a detailed adjustment process on the outline. First, the CPU 11 reduces the person image after the background image filling process to remove the pointed obstacle, and then re-enlarges it to the original size. By reducing the person image, the pointed obstacle is also reduced. If the pointed obstacle is small, it is erased when it is re-enlarged. Further, even if not erased, the line width becomes narrower and the display becomes thinner. Using this, the CPU 11 compares the reduced / enlarged person image with the original person image, extracts the erased part or the part with a thin display as an unnecessary pointed obstacle, The pixel is replaced with a pixel having a uniform pixel value. Thereby, it becomes possible to erase the pointed obstacle in the human image with certainty.

  Further, the CPU 11 detects the face image of the person image by the skin color extraction process performed in the conventional face image extraction process, and whether the skin color is the skin color with respect to the surrounding rectangular area centered on the coordinates of the jaw part. Judge whether or not. A similar color part can be determined to be part of the face image or the neck part, and other parts can be distinguished from the clothes part, shoulder part, and background part according to the inclination of the extracted edge. It becomes.

  For example, in the shoulder portion, the direction of the extracted edge falls within a predetermined range. FIG. 12 is an explanatory diagram of edge angle extraction in the left shoulder portion. That is, in the case of the left shoulder portion, as shown in FIG. 12, when the horizontal direction is set to 0 degree, the contour line indicating the shoulder is always inclined downward by the angle θ. Therefore, when the edge 121 inclined upward in the left shoulder portion is detected, the CPU 11 can easily determine that the edge 121 is not the contour line indicating the shoulder of the person image. Therefore, when the CPU 11 extracts the inclination angle θ of the edge of the shoulder portion and determines that θ <0 when the inclination angle θ is not within the predetermined range, that is, when the horizontal downward direction is positive, the background image for the edge portion is obtained. As a background image, the pixel is replaced with a pixel having a uniform pixel value. As a result, the contour line of the shoulder portion of the person image can be made clearer and unnecessary edges can be deleted. FIG. 13 is a view showing an example of a person image after unnecessary edge portion deletion. As shown in FIG. 13, even if any background image appears as an obstacle in the first image, it is possible to extract a person image reliably by performing the above-described processing.

  As described above, according to the present embodiment, as long as the top and the chin are specified for the human image including the background image, the human image can be reliably detected even if the background image is not a uniform image. It is possible to define a contour line so that it can be extracted.

  The following is added.

(Supplementary Note 1) In a person image extraction apparatus that extracts a person image taken in the foreground in an image,
Means for acquiring the coordinates of a point indicating the crown and chin of the person image;
Means for calculating a dispersion in a color space of a pixel value of a pixel located in a rectangular area including a coordinate indicating the acquired top portion;
Means for determining whether or not the variance of the calculated pixel values is smaller than a threshold;
Means for averaging the color space if the means determines that the variance of the pixel values is less than a threshold;
A contour extraction unit that extracts a contour of a person image as a portion in which the degree of change in the pixel value of an adjacent pixel is larger than a threshold;
A background image removal unit for removing the background image,
The background image removal unit stores a multi-value template imitating the shape of a person, adapts the coordinates of the top and chin of the person image to the face part of the multi-value template, and A human image extracting apparatus characterized in that a composite image with a multi-value template is generated.

(Appendix 2) The background image removal unit
The hair portion is extracted based on the dispersion in the color space of the pixel value of the pixel located in the rectangular area including the coordinates indicating the top of the head, and the boundary between the hair image and the background image is specified. The person image extraction device according to appendix 1, characterized by:

(Supplementary Note 3) The background image removing unit
Based on the coordinates acquired as the jaw, the average value of the pixel values of the facial image is obtained, and by determining whether the difference in pixel value from the obtained average value is greater than a predetermined threshold, The person image extraction device according to appendix 1 or 2, wherein a boundary between the neck portion and the background image is specified.

(Supplementary Note 4) The background image removing unit
The contour line near the shoulder of the person image is extracted, and a contour line whose inclination is not within a predetermined range is to be deleted. Human image extraction device.

(Additional remark 5) In the person image extraction method which extracts the person image currently image | photographed in the foreground in an image,
Obtaining coordinates of points indicating the top and chin of the person image;
Calculating a variance in a color space of pixel values of pixels located in a rectangular area including coordinates indicating the acquired top of the head;
Determining whether the calculated variance of pixel values is less than a threshold;
If it is determined that the variance of the pixel values is less than a threshold, averaging the color space;
Extracting a contour of a person image as a portion where the degree of change in pixel value of an adjacent pixel is larger than a threshold;
Removing a background image,
The step of removing the background image stores a multi-value template imitating the shape of a person, adapts the coordinates of the top and chin of the person image to the face portion of the multi-value template, and And a multi-value template for generating a composite image.

(Supplementary Note 6) The step of removing the background image includes:
A hair portion is extracted based on a dispersion in a color space of a pixel value of a pixel located in a rectangular area including coordinates indicating the top of the head, and a boundary between the hair image and a background image is specified. The person image extraction method according to appendix 5.

(Supplementary Note 7) The step of removing the background image includes:
Based on the coordinates acquired as the jaw, the average value of the pixel values of the facial image is obtained, and by determining whether the difference in pixel value from the obtained average value is greater than a predetermined threshold, The person image extraction method according to appendix 5 or 6, characterized in that a boundary between the neck periphery and the background image is specified.

(Supplementary Note 8) The step of removing the background image includes:
8. The person image extraction method according to any one of appendices 5 to 7, wherein a contour line near the shoulder of the person image is extracted, and a contour line whose inclination is not within a predetermined range is deleted. .

(Supplementary Note 9) In a computer program for causing a computer to execute a process of extracting a person image taken in the foreground in an image,
Obtaining coordinates of points indicating the top and chin of the person image;
Calculating a variance in a color space of pixel values of pixels located in a rectangular area including coordinates indicating the acquired top of the head;
Determining whether the calculated variance of pixel values is less than a threshold;
If it is determined that the variance of the pixel values is less than a threshold, averaging the color space;
Extracting a contour of a person image as a portion where the degree of change in pixel value of an adjacent pixel is larger than a threshold;
Removing a background image,
The step of removing the background image stores a multi-value template imitating the shape of a person, adapts the coordinates of the top and chin of the person image to the face portion of the multi-value template, and And a computer program for generating a composite image of the multi-value template.

(Supplementary Note 10) The step of removing the background image includes:
A hair portion is extracted based on a dispersion in a color space of a pixel value of a pixel located in a rectangular area including coordinates indicating the top of the head, and a boundary between the hair image and a background image is specified. The computer program according to appendix 9.

(Supplementary Note 11) The step of removing the background image includes:
Based on the coordinates acquired as the jaw, the average value of the pixel values of the facial image is obtained, and by determining whether the difference in pixel value from the obtained average value is greater than a predetermined threshold, The computer program according to appendix 9 or 10, wherein a boundary between a neck portion and a background image is specified.

(Supplementary Note 12) The step of removing the background image includes:
The computer program according to any one of appendices 9 to 11, wherein a contour line in the vicinity of a shoulder of the person image is extracted and a contour line whose inclination is not within a predetermined range is deleted.

1 is a schematic configuration diagram of a computer that embodies a human image extraction apparatus according to an embodiment of the present invention. It is a flowchart of the pre-processing in CPU of the person image extraction apparatus which concerns on embodiment of this invention. It is an illustration figure of the person image containing the background image displayed by the display means. It is a figure which shows the outline | summary of the averaging process regarding the lightness in color space. It is explanatory drawing of the principle of a smoothing process. It is a flowchart of the contour extraction process in CPU of the person image extraction apparatus which concerns on embodiment of this invention. It is an illustration figure of a sobel filter used for an outline extraction process. It is an illustration figure of a person image after applying a sobel filter and extracting a contour line. It is an illustration figure of the multi-value template applied to a person image. It is an illustration figure of the person image after multi-value template application. It is a schematic explanatory drawing of the background image filling process of a person image periphery. It is explanatory drawing of edge angle extraction in a left shoulder part. It is an illustration figure of the person image after unnecessary edge part deletion.

Explanation of symbols

11 CPU
12 storage means 13 RAM
DESCRIPTION OF SYMBOLS 14 Communication means 15 Input means 16 Display means 17 Image input means 18 Auxiliary storage means 19 Portable recording medium 31 Top part 32 Jaw part 51 Contour line

Claims (5)

In a person image extraction device that extracts a person image taken in the foreground in the image,
Means for acquiring the coordinates of a point indicating the crown and chin of the person image;
Means for calculating a dispersion in a color space of a pixel value of a pixel located in a rectangular region including the acquired coordinates indicating the top of the head;
Means for determining whether or not the variance of the calculated pixel values is smaller than a threshold;
Means for averaging the color space if the means determines that the variance of the pixel values is less than a threshold;
A contour extraction unit that extracts a contour of a person image as a portion in which the degree of change in the pixel value of an adjacent pixel is larger than a threshold;
A background image removal unit for removing the background image,
The background image removal unit stores a multi-value template imitating the shape of a person, adapts the coordinates of the top and chin of the person image to the face part of the multi-value template, and A human image extracting apparatus characterized in that a composite image with a multi-value template is generated. The background image removing unit
The hair portion is extracted based on the dispersion in the color space of the pixel value of the pixel located in the rectangular area including the coordinates indicating the top of the head, and the boundary between the hair image and the background image is specified. The person image extraction device according to claim 1. The background image removing unit
Based on the coordinates acquired as the jaw, the average value of the pixel values of the facial image is obtained, and by determining whether the difference in pixel value from the obtained average value is greater than a predetermined threshold, 3. The person image extracting apparatus according to claim 1, wherein a boundary between the neck portion and the background image is specified. The background image removing unit
The contour line in the vicinity of the shoulder of the person image is extracted, and a contour line whose inclination is not within a predetermined range is to be erased. Human image extraction device. In a computer program for causing a computer to execute a process of extracting a person image taken in the foreground in the image,
Obtaining coordinates of points indicating the top and chin of the person image;
Calculating a variance in a color space of pixel values of pixels located in a rectangular area including coordinates indicating the acquired top of the head;
Determining whether the calculated variance of pixel values is less than a threshold;
If it is determined that the variance of the pixel values is less than a threshold, averaging the color space;
Extracting a contour of a person image as a portion where the degree of change in pixel value of an adjacent pixel is larger than a threshold;
Removing a background image,
The step of removing the background image stores a multi-value template imitating the shape of a person, adapts the coordinates of the top and chin of the person image to the face portion of the multi-value template, and And a computer program for generating a composite image of the multi-value template.

Images