JPH05165119A - Method for extracting data on person's face - Google Patents

Abstract

PURPOSE:To obtain the method for extracting the data on a person's face which can automatically extract only the data on the person's face with high accuracy from a color original image. CONSTITUTION:An original image is divided by each of color regions in accordance with the data obtd. by photometry. One region is selected as a noticed region from the color regions. The nucleus of the noticed region is determined by subjecting the noticed region selected to contraction processing (162). The circle with which the area inscribing the noticed region centering the determined nucleus is largest is determined (164 to 166) and the end of the division concerning the entire region of the noticed region is judged (168 to 170). The noticed region is divided in order of the circles of the larger areas by repeating the processing. The circle of the divided max. area is judged as the region of the face after the end of the division of the noticed region. The photometric data of R, G, B of the region judged to be the face are outputted to a correct exposure calculating circuit (172).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a human face data extraction method, and more particularly, to a human face density data used when copying an original color image onto a color copying material or a black-and-white copying material. Method for extracting face data of a person.

[0002]

2. Description of the Related Art When viewing a portrait photograph, the most noticeable part is the person's face, and in order to create a high quality photograph, the person's face color should be adjusted appropriately. It is necessary to burn it in a different color.

Conventionally, a face area in an original image on a color film is designated by a light pen to extract density data of a person's face, and the color of the face is properly printed based on the extracted density data. The amount of exposure is determined. Examples of such a technique include JP-A-62-115430, JP-A-62-115431, and JP-A-62-1154.
32, JP-A-62-189456, JP-A-62-189457, and JP-A-63-138340.
JP-A-63-178222.

However, in the above-mentioned conventional technique, the operator has to specify the face area with a light pen for each image, and therefore, there is a problem that the printing operation takes time. Further, since the operator has to visually specify the face area, it is difficult to unmanned.

Further, Japanese Patent Laid-Open No. 52-156624,
JP-A-52-156625, JP-A-53-123
30, JP-A-53-145620, JP-A-53-145621, and JP-A-53-145622.
The publication describes the following method for extracting the data of a person's face by extracting the skin color data. That is, the color original image is divided into a large number of photometric points, and each photometric point is decomposed into three colors of R (red), G (green), and B (blue) for photometry, and each photometric point calculated from the photometric data. It is determined whether the color of is within the skin color range. Then, the cluster (group) of photometric points determined to be in the skin color range is used as face density data.

However, in this method, since the color within the skin color range is assumed to be the density data of the face, the parts other than the face having the skin color such as the ground, the trunk of the tree, and the clothes or the color close to the skin color of the face are also included. It is extracted as density data.

Further, even if the skin color such as the ground or a tree trunk or a part similar to the skin color can be excluded and extracted, in an image of the sea bathing season, pool, etc., an area with the same skin as the face with a lot of exposed skin is the face. There are a large number around the area, and the area around the area may be combined with the facial area. In addition, as a person image, there is an image with a cheek stick, and the image of the same color as the face is combined with the face area. Therefore, when trying to extract a face area from the color features of the original image, it is not possible to extract the area of only the face, and the combined area of the same color is also extracted. As a result, the density data of only the face cannot be automatically extracted.

The present invention has been made to solve the above problems, and it is possible to automatically extract only the face data of a person, which is characteristic image data, from a color original image such as a negative film with high accuracy. It is an object of the present invention to provide a method of extracting face data of a person who can do it.

[0009]

In order to achieve the above object, the invention of claim 1 divides a color original image into a large number of pixels and decomposes each pixel into three colors of red light, green light and blue light. Photometric, obtain a color region having the same or similar hue on the color original image based on the data obtained by photometry, and assume a circle or an ellipse inscribed at the boundary part of the obtained color region, and A color region obtained by a circle or an ellipse is sequentially divided from a large circle or an ellipse, at least one of the divided regions is selected, and the data of the selected region is extracted as the face data of a person.

According to a second aspect of the invention, the color original image is divided into a large number of pixels, each pixel is decomposed into three colors of red light, green light and blue light, and photometry is performed. Based on the data obtained by the photometry. Then, a color area having the same or similar hue on the color original image is obtained, and a circle or an ellipse inscribed in the boundary portion of the obtained color area is assumed, and an area having the largest circle or an ellipse is obtained, A convex region having a predetermined area or more that protrudes outward from the outer circumference of a circle or an ellipse and is in contact with the obtained region is obtained, and the color region is divided by a circular region inscribed in the boundary region of the color region included in the convex region. Then, at least one of the divided areas is selected and the data of the selected area is extracted as the face data of the person.

[0011]

Since the most noticeable part of a person's photograph is the person's face, it is preferable to extract the data of the person's face area from the original color image and use that data as the characteristic image data. .. In addition, most of the human face shapes are circles or ellipses, and there is a possibility that the human face region will be extracted by preferentially extracting the circular region from the color region of the similar color of the color original image. Is high.

Therefore, in the invention described in claim 1, the color original image is divided into a large number of pixels, each pixel is decomposed into three colors of red light, green light and blue light, and photometry is performed, and data obtained by photometry is obtained. Based on this, color regions having the same or similar hue are obtained on the original color image. Next, a circle or an ellipse inscribed in the boundary portion of the obtained color region is assumed, and the color region obtained by the circle or the ellipse region is divided in order from a circle or an ellipse having a large area. As a result, the finally divided area is likely to be an independent area of only the face. Therefore, each of the divided color regions of the color original image is further divided into regions that are likely to be extracted as a face region, and if at least one region that represents the features of this image is selected, it is selected. Since the data of the area represents the characteristic image data, the characteristic image data can be extracted by selecting the area.

When the face region is in contact with a part other than the face (for example, when the face region is buried in the body when photographed obliquely from above in the pool), the inscribed circle of the color region If it is divided first, the face area may be divided and extracted. Therefore, in the invention described in claim 2, a color region having the same or similar hue is obtained on the color original image based on the data obtained by photometry as described in claim 1. Next, a circle or an ellipse inscribed in the boundary part of the obtained color region is assumed, and a region of a circle or an ellipse having the largest area is obtained. Then, when the obtained area is in contact with a convex area having a predetermined area or more protruding outward from the outer circumference of the circle or the ellipse, a circle inscribed in the boundary part of the color area included in the convex area The color area is divided by the area. As a result, even if similar color areas are in contact with each other on the original color image, it is possible to divide a circle or ellipse area in which a person's face area is likely to be extracted as a face area with priority. By selecting at least one of the divided areas, the data in this area can be extracted as the characteristic image data.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the present invention will be described in detail below with reference to the drawings. In this embodiment, the present invention is applied to an auto printer. As shown in FIG. 1, the auto printer of this embodiment includes a conveyance roller 12 that conveys a color negative film 10. A light source 14, a color correction filter 16 such as a dimming filter, and a diffusion box 18 are sequentially arranged below the color negative film 10 conveyed by the conveyance roller 12. Further, above the negative film 10, a distribution prism 20 that distributes the light rays transmitted through the negative film 10 in two directions is arranged. A projection optical system 22, a black shutter 23 and a color paper (printing paper) 24 are sequentially arranged on one optical path distributed by the distribution prism 20, and a projection optical system 26 and a CCD image sensor are arranged on the other optical path. 28
Are arranged in order. This CCD image sensor 28
Divides the entire screen (one frame) of the negative film 10 into a large number of pixels (for example, 256 × 256 pixels) and divides each pixel into three colors of R (red), G (green), and B (blue). Disassemble and measure light. The CCD image sensor 28 is connected to a 3 × 3 matrix circuit 34 for sensitivity correction of the CCD image sensor via an amplifier 30 that amplifies the output of the CCD image sensor and an analog-digital (A / D) converter 32. The 3 × 3 matrix circuit 34 is connected to an appropriate exposure amount calculation circuit 40 via a face extraction circuit 36 composed of a microcomputer that stores a program of a routine described below, and at the same time, calculates the average density of one entire screen. It is connected to a proper exposure amount calculation circuit 40 via an average density calculation circuit 38 for calculation. The appropriate exposure amount calculation circuit 40 is connected to the color correction filter 16 via a driver 42 that drives the color correction filter.

Next, the operation of this embodiment will be described. Light source 14
The light beam emitted from the laser beam passes through the color correction filter 16, the diffusion box 18, and the color negative film 10, is distributed by the distribution prism 20, and is received by the CCD image sensor 28 via the projection optical system 26. At this time, the black shirt 23 is closed. By this light reception, the CCD image sensor 28 divides the entire screen into a large number of pixels, decomposes each pixel into R, G, and B colors, and performs photometry, and outputs a photometric data signal. The photometric data signal is amplified by the amplifier 30, converted into a digital signal by the A / D converter 32, the sensitivity of the image sensor is corrected by the 3 × 3 matrix circuit 34, and the face extraction circuit 36 and the average density calculation circuit 38. Entered in. The average density calculation circuit 38 calculates the average density of one screen. The face extraction circuit 36 estimates the human face part in one screen as described below, and outputs R, G, B three-color photometric data of the part estimated to be the face. The exposure amount calculation circuit 40 calculates the exposure amount using the three-color photometric data output from the face extraction circuit 36 and the average density calculated by the average density calculation circuit 38, and the color correction filter 16 is calculated via the driver 42. Is controlled and the black shutter 23 is opened / closed for printing. When the average density calculated by the average density calculation circuit 38 is used, the exposure correction amount for the average density can be calculated. When the exposure correction amount is not calculated, the average density calculation circuit 38 is not necessarily required, and the exposure amount may be calculated directly from the three-color photometric data output from the face extraction circuit 36.

FIG. 2 shows a face extraction routine by the face extraction circuit 36, which performs noise removal, that is, smoothing, on the three-color photometric data input in step 100. In the next step 102, the following (1) to (3)
The R, G, B three-color photometric data is converted into H (hue value),
Convert to L (lightness value) and S (saturation value).

L = (R + G + B) / 3 ... (1) S = 1-min (r ', g', b ') ... (2) H = H' / 2Pi .... ( 3) However, R, G, and B are standardized so that the minimum value is 0 and the maximum value is 1 as shown in the three-dimensional color coordinates in FIG.
Color photometric data, min () is the minimum value of the numerical values in (), r ', g', b'is r '= R / L, g' = G / L,
It represents b '= B / L. Further, H ′ is given by the following expression (4), and Pi (i is one of R, G, and B) is P in FIG.
Is.

[0018]

[Equation 1]

However,

[0020]

[Equation 2]

In step 104, as shown in FIG. 4A, a two-dimensional histogram of the hue value and the saturation value is obtained using a coordinate system composed of a hue value axis, a saturation value axis and a pixel number axis which are orthogonal to each other. Then, as will be described later in step 106, the obtained two-dimensional histogram is divided into mountains, that is, clustering of the two-dimensional histogram is performed. In the next step 108, clustering of a large number of pixels is performed based on the peaks of the clustered two-dimensional histogram, the screen is divided based on this clustering, and areas that are candidates for human faces are extracted from the divided areas. .. In the next step 110, at least the face region is divided into other regions by further dividing the color region extracted as a face candidate into a circle or an ellipse region, and based on this divided region. The area of the face is estimated and the R, G, B three-color photometric data of the area estimated as the face is output. Then, in step 112, it is judged whether or not the printing of all the frames is completed, and when it is judged that the printing is completed, this routine is ended.

Next, details of steps 106 to 110 will be described. FIG. 5 shows the details of step 106. In step 120, an area to be evaluated is cut out from the two-dimensional histogram of the hue value and the saturation value.
In FIG. 4, one frame is set as the evaluation area for simplification of description. In step 122, it is determined whether there is an evaluation area. When the evaluation area cannot be cut out in step 120, that is, when the evaluation of all areas is completed, there is no evaluation area, and this routine is ended. If there is an evaluation area, in step 124, the X and Y axes for creating the mountain cutting histogram are determined.
That is, the evaluation area is rotated about an axis parallel to the pixel number axis, and when the peaks of the histogram are viewed from the side, priority is given to multimodality and the position where the peaks are the sharpest is obtained. Determine the X and Y axes. When the processing time needs to be shortened, the accuracy is somewhat lowered, but the axes that maximize the variance of the histogram may be used as the X and Y axes. In the example of FIG. 4 (1), when the four mountains with the symbols 1 to 4 are viewed from the side, priority is given to multimodality and the peaks are the most sharp positions where the three mountains are visible. Therefore, the X axis is defined in the direction orthogonal to the viewing direction, and the Y axis is defined in the direction orthogonal to the X axis.

In the next step 126, the two-dimensional histograms are projected on the X and Y axes to create one-dimensional histograms. In the example of FIG. 4 (1), when viewed from the direction orthogonal to the X axis, the peaks with the reference numerals 1 and 2 appear to overlap, so the one-dimensional histogram for the X axis has the peaks and the reference numerals with the reference numeral 3. Three mountains, a mountain with 1 and 2 and a mountain with symbol 4, appear, and when viewed from the direction orthogonal to the Y axis, the mountains with symbols 1 to 4 appear to overlap, so one dimension about the Y axis. One mountain appears in the histogram. In the next step 128, the histogram is converted into an evaluation function H (a) by the following equation (5), and mountains are cut out from the histogram on the X axis based on this evaluation function.

[0024]

[Equation 3]

However, f (a) is the number of pixels when the value (feature amount) in the X-axis direction is a, and x is the displacement from the feature amount a.

That is, the average value T of the evaluation function H (a) is obtained, and the range (valley, skirt existence range) below the average value T of the evaluation function H (a) is obtained. Next, the position where the histogram is the smallest within this range is set as the valley or skirt of the histogram. Then, a histogram is cut out at the obtained valley or skirt.

Explaining the cutting of the mountain with reference to FIG. 6, when the evaluation function H (a) is obtained from the histogram represented by the solid line SI, it becomes as shown by the broken line in the figure. The range of the average value T or less with respect to the negative portion of the evaluation function H (a) is the range of feature amounts v0 to v1 and v2 to v3. The position where the frequency of the histogram in this range is the minimum is the range v
Av0 = v0 in 0 to v1, av1 in the range v2 to v3
Then, av0 is obtained as a skirt and av2 is obtained as a valley, and the histogram is cut out at this position.

In step 130, the mountain peak of the Y axis is cut out in the same manner as the mountain peak cut of the X axis. Next Step 13
In 2, the area where the peaks of the one-dimensional histogram for the X-axis and the Y-axis cut out as described above overlap is obtained on the two-dimensional histogram, and the mountain is cut out from the two-dimensional histogram for the hue value and the saturation value. .. An area E1 in FIG. 4 (1) shows an example of a mountain cut out as described above.

In the next step 134, it is judged whether or not the mountain cut out from the two-dimensional histogram is a single peak, and if it is not a single peak, steps 124 to 134 are repeated until the mountain cut out from the two-dimensional histogram becomes a single peak. repeat. Region E2 in FIG. 4C shows an example of a single-peak mountain cut out as described above.

In the next step 136, a process (labeling) for attaching a label for identifying the cut-out single-peak mountain is performed. In step 138, the labeled mountain is masked and the process returns to step 120. Then, the above steps are repeated to divide the entire area of the two-dimensional histogram for the hue value and the saturation value into single peaks.

FIG. 7 shows the details of step 108 in FIG. 2. In step 140, the range XR (FIG. 4 (3)) and Y in the X-axis direction of the single-peaked mountain divided as described above is obtained. An axial range YR (FIG. 4 (3)) is obtained for each single peak, and it is determined whether the hue value and the saturation value of each pixel of the original image belong to these ranges, and pixel clustering is performed. At the same time, the pixels belonging to the range surrounded by the ranges XR and YR are collected, and the original image is divided so that the collected pixels form one area on the original image. Also, the divided areas are numbered. FIG. 4 (2) shows an example in which the original image is divided, and the pixels in each region denoted by reference numerals 1 to 4 are
It corresponds to the pixels included in the unimodal peaks denoted by reference numerals 1 to 4 in FIG. Pixels belonging to the same single-peak mountain in FIG. 4 (1) are divided into different regions in FIG. 4 (2). This is due to the hue value range of the single-peak mountain in FIG. 4 (1). This is because the pixel has a saturation value range, but the region is divided in FIG. 4 (2).

At the next step 142, minute areas are removed by judging the areas of the divided areas, and the numbering is performed again. In the next step 144, a contraction process for deleting all the boundary pixels of the region to remove one skin and an expansion process for expanding the boundary pixels in the direction of the background pixel to enlarge the skin by one converse to the contraction process are performed. It separates the busy small area from the large area. Next step 146
Then, in the same manner as in step 142, minute areas are removed and renumbering is performed, and in step 148, in order to separate the weakly bonded areas, the same contraction and expansion processing as described above is performed. Similarly, the minute area is removed and renumbering is performed.

FIG. 8 shows the details of step 110. Further, in order to describe step 110 in detail, the image shown in FIG. 9A when the color region of the same color as the face or a similar color covers a wide range will be used.

First, in step 162, step 1
08, that is, one region is selected as a region of interest from the regions extracted by the routine of FIG. 7 (see FIG. 9 (1)). Next, a contraction process is performed on the selected attention area to obtain a nucleus for decomposing the attention area. That is, the contraction process of deleting the boundary pixels of the attention area is repeatedly performed, and one finally remaining dot-shaped or linear area is used as the nucleus. As shown in FIG. 9 (2), the linear region means a plurality of continuous points (pixels or a set of a plurality of pixels), that is, a line L ₁ . The target image is the original image C
Since it is based on the data obtained by photometry by the CD image sensor 28, it is not a continuous system but a discrete system.
The nucleus will have an area. Also, depending on the shape of the target image, there may be a plurality of nuclei that finally remain. In this case, the region having the smallest area is the core. Further, when a region having the same area remains, any one is selected.

At the next step 164, a circle or an ellipse having the maximum area inscribed in the region of interest centering on the nucleus obtained as described above is obtained. That is, by performing expansion processing centered on the nucleus as many times as contraction processing is performed to obtain the nucleus as described above, when the nucleus is a point-like shape, it is an inscribed circle, and when the nucleus is a linear shape, It can be obtained as an inscribed ellipse (Morphologic filter).

After the circle or ellipse having the largest area inscribed in the region of interest is obtained, the process proceeds to step 166, and step 1
At 66, a process (labeling) is performed to attach a label for identifying the obtained circle having the largest area (or an ellipse formed by this circle). In the next step 168, the labeled circle or oval region BL1 is masked and the step 1
Proceed to 70. In the next step 170, it is judged whether or not the division by the circle or the oval area is completed for all the extracted areas, and if it is not completed, the step 16 is executed.
2 to step 168 are repeated. As a result, FIG.
As shown in (2) to (6), the regions BL1 to BL10 are divided in the order of circles having a large area.

When the division of the attention area is completed, step 17
Go to 2. In step 172, although details will be described later, at least one of the divided circles or the ellipses is selected to estimate the face area, and R, G, B of the area estimated to be the face are selected.
The photometric data is output to the proper exposure amount calculation circuit 40.

As described above, even when the color area having the same color as the face covers a wide area including the area other than the face, the color area is further divided into circular areas having a high probability as a face area. Therefore, by selecting this circular region, it is possible to extract data of only the face region with a high probability even from an image in which faces, grounds, trees, and the like having similar hues and saturations are mixed.

The kernel L _i and the area X _{i of the} inscribed circle described above are expressed by the equations (6) and (7) of the set operation shown below.
Can be expressed as

[0040]

[Equation 4]

Here, the shape B _i when contracting the color area and obtaining the inscribed circle of the color area centered on the nucleus is expressed by the following equation (8).
Can be determined by

[0042]

[Equation 5]

However, B: circular component Also, this n _i can be determined by the following equation (9).

[0044]

[Equation 6]

However, k = 1, 2, ... M _i ² = Area (X−X ′ _i−1 ) Area (X): Area of X region Size to be decomposed by optimally selecting this k value. N _i can be defined. The above formula (9)
May be Equation (10) shown below.

[0046]

[Equation 7]

However, k = 1, 2, ..., R ₀ : radius of component B Further, the relationship between the circular component B and the X region is as follows.

[0048]

[Equation 8]

It can be represented by FIG. 15 shows Step 1
72 shows details of 72. In step 302, one area is selected from the areas divided into circles or ellipses as described above as a characteristic area, and the horizontal fillet diameter and the vertical fillet diameter of the characteristic area are set to predetermined values. The size of the characteristic region is standardized by enlarging / reducing the characteristic region so that the value becomes a value, and the density value or the luminance value is standardized according to the following equation (11).

[0050]

[Equation 9]

Where dmax: maximum density value (or brightness value) in the area dmin: minimum density value (or brightness value) in the area ds: full-scale density value (or brightness value) of the image sensor d: density value before normalization (Or brightness value) dr: normalized density value (or brightness value) In step 304, a plurality of types (10 types in the present embodiment) of standard face images stored in advance (face image viewed from the front, horizontal direction) The correlation coefficient r of the characteristic region with respect to the face image (left and right), the downward face image, the upward face image, etc. viewed from
The calculation is performed according to the equation (2), and this correlation coefficient is used as the feature amount.
Even if this standard face image is only the data of the outline of the face, the internal structure of the face (eye, nose, mouth, etc.) is added to the data of the outline of the face.
It may be data to which data is added.

[0052]

[Equation 10]

However,

[0054]

[Equation 11]

Where T is the horizontal and vertical fillet diameters of the image (here, the fillet diameters are the same), f
(X, y) represents a characteristic region, and g (x, y) represents a standard face image.

Then, in step 306, it is determined whether or not the characteristic region is a human face by linear discriminant analysis using the characteristic amount as a variable, and R, G, B photometric data of the region determined to be a face is determined. Is output to the proper exposure amount calculation circuit 40. In step 308, it is determined whether or not the determination as to whether the face is a face has been completed for all the divided areas. If not, steps 302 to 308 are repeated.

In the above, the correlation coefficient is used as the feature amount used to determine whether the face is a person's face. However, the invariant derived from the normalized central moment around the center of gravity, which will be described below, Autocorrelation functions or geometric invariants may be used.

The central moment μ _pq about the (p + q) th centroid of the image f (x, y) is

[0059]

[Equation 12]

However,

[0061]

[Equation 13]

Then, the normalized central moment around the center of gravity is as follows.

[0063]

[Equation 14]

However, y = (p + q + 2) / 2 p + q = 2, 3, ... From the above, the following seven invariants ψ _i are derived from the normalized central moments about the second and third centroids. _, (I
= 1, 2, ..., 7) is derived.

[0065]

[Equation 15]

Further, the autocorrelation function R _f is expressed as follows.

[0067]

[Equation 16]

The geometrical invariant feature amount is expressed by the following equation.

[0069]

[Equation 17]

The appropriate exposure amount calculation circuit 40 is composed of the face extraction circuit 3
6, the R, G, B photometric data of the face area extracted as described above and the screen average density D _i of one frame calculated by the average density calculation circuit 38 (i = R, G, or B) ) And are used to calculate an appropriate exposure amount E _i according to the following equation, and output to the driver 42. The driver 42 controls the dimming filter 16 by calculating an exposure control value from the appropriate exposure amount E _i .

L _og E _i = LM _i · CS _i · (DN _i −D _i ) + PB _i + LB _i + MB _i + NB _i + K ₁ + K ₂ (11) However, each symbol represents the following.

LM: Magnification slope coefficient, which is preset according to the enlargement magnification determined by the type of negative and the print size.

CS: Color slope coefficient prepared for each type of negative, there are underexposure and overexposure, and it is determined whether the average density of the frame to be printed is under or over the standard negative density value. Either underexposure or overexposure is selected.

DN: Standard negative density value. D: Average density value of print frame.

PB: A correction balance value for a standard color paper, which is determined according to the type of color paper.

LB: For standard print lens. It is a correction lens balance value, and is determined according to the type of printing lens.

MB: A correction value (master balance value) with respect to a change in print light source and a change in paper developing performance.

NB: Negative balance (color balance) value determined by the characteristics of the negative film.

K ₂ : Color correction amount. K ₁ : A density correction amount represented by the following formula.

[0080]

[Equation 18]

Here, K _a and K _b are constants, and FD is the face area average density. Further, the density correction amount K ₁ in the above equation (8) may be used as the correction value obtained by the film inspection apparatus, and the color correction amount K ₂ may be expressed using the face area average density as follows.

[0082]

[Formula 19]

However, K _c is a constant. Further, the density correction amount K ₁ and the color correction amount K ₂ in the above equation (8) are used as the correction amounts obtained by the film inspection device, and the average density D _i of the print frame in the equation (8) is the average density FD of the face area. _i
Alternatively, the exposure amount may be obtained instead.

Next, a second embodiment of the present invention will be described with reference to the flowchart shown in FIG. In this embodiment, the face area is optimally divided from the area in contact with the face area. That is, in most cases, the face area in the image of the person is around the body. Therefore, when the color region including the face region has a plurality of convex regions, by selecting the divided circular region by dividing the color region from the circular region inscribed in the convex portion, The area in which the circular area is highly likely to be estimated as the face area is preferentially divided.

First, in step 180, step 1
08, that is, one region is selected as a region of interest from the regions extracted by the routine of FIG. 7 (FIG. 11 (1)).
reference). Then, as in the first embodiment, the selected region of interest is contracted to obtain the nucleus L _a of the region of interest (see FIG. 11 (2)).

In the next step 182, a circle or an ellipse B _a having the largest area inscribed in the region of interest centering on the nucleus L _a thus obtained is obtained. Next Step 184
Then, by comparing the area of the circle or ellipse inscribed in the attention area and the area of the attention area obtained as described above, the area ER _i (i remaining around the inscribed circle of the attention area
= 1, 2, ...) Is greater than or equal to a predetermined value.
When the area of the remaining area ER _i is equal to or larger than the predetermined value, there is a possibility that a convex area exists in the remaining area ER _i, and therefore the process proceeds to step 186. When the area is less than the predetermined value, step 1 is performed.
Proceed to 92.

In step 186, first, FIG. 11 (3)
As shown in, the kernel L _a obtained as described above is excluded (masked) from the attention area. In the next step 188, contraction and expansion processing is performed in the same manner as described above, and the nucleus L
A circle or ellipse B _b having the largest area inscribed in the region of interest in which _a is masked is obtained (see FIG. 11 (4)).

In the next step 190, as described above,
Inscribed circle or ellipse B obtained by_a, B _bEach of the predetermined (1
~ 3) Expand in the range of about a pixel and
Of which the size of the area protruding from the area of interest is greater than or equal to a specified value
The color region of interest by determining whether or not
Judge the size of the convex portion on the outer side of. I.e. expansion
The area outside the attention area is large.
It is judged to be convex. If the obtained convex part is more than a predetermined value
If there is, go to step 194 and inscribe
Yen or ellipse B_bProcess to attach a label to identify
And proceed to step 196. When the value is below a certain value
Is the same as in the first embodiment, at step 192,
A circle with the largest area inscribed in the area of interest (or by this circle
The process of attaching a label to identify the oval)
Then, the process proceeds to step 196.

In step 196, the area of the circle or ellipse labeled in step 192 or step 194 is masked, and the flow advances to step 198. In step 198, it is determined whether or not the division by the circle or the oval area has been completed for all the extracted areas. If not completed, steps 180 to 198 are repeated. In the next step 200, at least one region is selected from the divided circles or ellipses, the face region is determined in the same manner as in the first embodiment, and the R of the region determined to be the face, The G and B photometric data are output to the proper exposure amount calculation circuit 40.

As described above, in the present embodiment, a convex portion around the color region and a large circular area inscribed in the convex portion, that is, a circular region having a high probability of being a face region is given priority. Since the region is divided, even in the color region where the body and the face are in contact with each other, it is possible to extract the data of the face region by dividing only the face region with higher accuracy.

In the above embodiment, an example in which the color area is divided based on the two-dimensional histogram of the hue value and the saturation value has been described, but the present embodiment is not limited to this, and the hue is not limited to this. You may divide into a color area based on the histogram only by a value. Further, the color area is not limited to the histogram, and the color region may be divided based on other feature amounts of the original image.

FIG. 12 shows a modified example in which the present invention is applied to an exposure amount determining device separate from a printer or a printer processor. Note that, in FIG. 11, portions corresponding to those in FIG. Further, the average density calculation circuit 38 is not always necessary, but instead of this, an integrated transmission density detection circuit for detecting LATD of the entire screen may be used.

In FIG. 13, the face extraction circuit of FIG. 12 is composed of a plurality of face extraction circuits 36 ₁ , 36 ₂ ... 36n, and the exposure amount is calculated by parallel processing. Face extraction circuit 36
_1, 36 ₂ ··· 36n reads an image according to the time chart of FIG. 14, calculates the exposure amount, and outputs the result. In FIG. 14, t ₁ is the image reading time for one frame, and t _1
2 is the exposure amount calculation time for one frame, t ₃ is the exposure amount calculation result transfer time for one frame, and t ₂ >> t ₁ and t ₃ . The face extraction circuit 36 ₁ reads one frame image at t ₁ hours, and
_The exposure amount is calculated in ₂ hours, and the calculation result is transferred in t ₃ hours. At the same time when one frame of image is read by the face extraction circuit 36 ₁ , one frame of film is fed and the face extraction circuit 36 1
1 frame image reading by ₂ is started, face extraction circuit 3
The exposure amount calculation of 6 ₁ and the image reading of the face extraction circuit 36 ₂ are performed in parallel, and the same applies to the face extraction circuits 36 ₃ and 36.
₄ ... 36n are processed in parallel.

The time Tp required for parallel processing of mxn frames is Tp = m (t ₁ + t ₂ + t ₃ ) + (n-1) t ₁ . On the other hand, the processing time T when parallel processing is not performed
s is Ts = m · n (t ₁ + t ₂ + t ₃ ). Therefore,

[0095]

[Equation 20]

Double speeding is possible. This parallel processing device can also be applied to the printer of FIG.

In addition to determining the exposure amount of the photographic printing apparatus, the present invention determines the exposure amount of the digital color printer, the copying condition of the copying machine, the exposure amount of the camera, the display condition of the CRT screen, the hard copy from the magnetic image data. It can also be applied to the determination of the amount of light when creating.

[0098]

As described above, according to the first aspect of the present invention, even if an image of the same color as or similar to the characteristic image is mixed in the area divided by the color area division,
Since the color area is divided by the circular area that is likely to be the person's face, it is possible to separate the person's face image with high probability and extract only the person's face data.

Further, according to the second aspect of the present invention, since the area having a large area in contact with the circle inscribed in the color area is preferentially divided, a part having the same hue or a similar hue as the human face is mixed. Even if it does, the face data of a person can be separated and extracted with a high probability.

[Brief description of drawings]

FIG. 1 is a schematic diagram showing a printer according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a face extraction routine of a face extraction circuit.

FIG. 3 is a diagram showing color coordinates.

FIG. 4 (1) is a diagram showing a two-dimensional histogram of hue values and saturation values. (2) is a diagram showing a state in which the original image is divided. (3) is a diagram showing a state in which a single peak is cut out from a two-dimensional histogram.

5 is a diagram showing details of step 106 of FIG. 2. FIG.

FIG. 6 is a diagram showing a histogram and an evaluation function.

FIG. 7 is a diagram showing details of step 108 in FIG.

FIG. 8 is a diagram showing details of step 110 of FIG.

FIG. 9 is a diagram showing a process of dividing a color area according to the first embodiment.

FIG. 10 is a flow chart of a face estimation routine according to the second embodiment of the present invention.

FIG. 11 is a diagram showing a process of dividing a color area according to the second embodiment.

FIG. 12 is a schematic diagram of an exposure amount calculation device to which the present invention is applied.

FIG. 13 is a schematic diagram of an exposure amount calculation device that performs parallel processing by a plurality of face extraction circuits.

FIG. 14 is a diagram showing a time chart of parallel processing.

15 is a diagram showing details of step 172 of FIG. 8. FIG.

[Explanation of symbols]

 28 CCD image sensor 30 amplifier 36 face extraction circuit

Claims (2)

[Claims] 1. A color original image is divided into a large number of pixels, each pixel is decomposed into three colors of red light, green light and blue light, and photometry is performed. Based on the data obtained by photometry, the original color image is displayed on the color original image. A color area that has the same or similar hue, is assumed to be a circle or ellipse that is inscribed in the boundary area of the obtained color area, and is a circle or an ellipse in order from the circle with the largest area or the ellipse. A method for extracting face data of a person, which comprises dividing at least one of the divided areas and selecting data of the selected area as data of the face of the person. 2. A color original image is divided into a large number of pixels, each pixel is decomposed into three colors of red light, green light and blue light, and photometry is performed. Based on the data obtained by photometry, the original color image is displayed on the color original image. Find a color area with the same or similar hue, assume a circle or ellipse that is inscribed in the boundary area of the obtained color area, and find the area of the circle or ellipse with the largest area. A convex region having a predetermined area or more that protrudes outward and is in contact with the obtained region is obtained, and the color region is divided by a circular region inscribed in the boundary portion of the color region included in the convex region. A method of extracting data of a person's face, wherein at least one is selected and the data of the selected area is extracted as the data of the person's face.