JP2008199146A - Photographing apparatus, method and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To further improve the detection accuracy of a face from an image, in a photographing apparatus. <P>SOLUTION: In the photographing apparatus, an image pickup system 6 acquires an image by photographing, a face detection part 37 calculates the degree of matching between a detection frame and an image and detects the position of the detection frame of which the degree of matching is a prescribed threshold and more as a face candidate. A face component detection part 38 detects candidates of at least one face component included in the face candidate in each face component. A judgment part 39 judges whether each face candidate is a real face or not on the basis of the number and/or positions of face component candidates detected in each face component. A threshold setting part 42 sets a threshold to be compared with the matching degree at the time of face detection to a first value for prescribed photographing, and for the other photographing, sets the threshold to a second value larger than the first value. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a photographing apparatus and method such as a digital camera for obtaining an image by photographing, and a program for causing a computer to execute the photographing method.

  In shooting with a digital camera, for example, an object such as a face is detected from an image acquired by shooting, and the conditions of image processing applied to the image are changed according to the detection result of the object, or the shooting conditions at the time of shooting are changed. It has been done. In particular, when the object is a face, the number of detected faces is counted, or the detected faces are trimmed and recorded.

  Thus, in order to detect an object from an image and perform various processes, it is necessary to accurately detect the object from the image. For this reason, various methods for accurately detecting an object have been proposed. For example, a face image of a person to be authenticated is photographed, a feature amount of the face of the authentication target person is extracted from the face image, a similarity between the extracted feature amount and a reference feature amount is calculated, and the similarity obtained by this calculation By comparing the threshold with the threshold value and authenticating whether or not the person to be authenticated is the principal, by changing the threshold according to whether or not the authentication person is frequently used There has been proposed a technique for improving the authentication rate in a time zone where the use frequency of the authentication target person is high (see Patent Document 1).

In addition, face candidates are detected from the image, and face candidates that do not satisfy a predetermined condition, such as when the color dispersion value of the face candidates is small or when the occupation rate of the skin color area is large, are excluded from the detected face candidates. A technique to do this has also been proposed (see Patent Document 2).
JP 2002-183734 A JP-A-2005-78376

  Although the face authentication accuracy or the face detection accuracy can be improved by the methods described in Patent Documents 1 and 2, it is desired to further improve the accuracy.

  The present invention has been made in view of the above circumstances, and an object thereof is to further improve the accuracy of detecting a face from an image.

An imaging apparatus according to the present invention includes imaging means for continuously acquiring images by continuous imaging,
A detection frame of a predetermined size is moved on the image, a feature amount is calculated from the image in the detection frame for each moved position, and a matching degree between the feature amount and a predetermined face feature amount is calculated. , A face detection unit that detects an image at the position of the detection frame as a face candidate when the matching degree is equal to or greater than a predetermined threshold;
Face component detection means for detecting at least one face component candidate included in the face candidate for each face component;
Determining means for determining whether or not the face candidate is a true face based on at least one of the number and position of the face component candidates detected for each face component;
When the face candidate is detected from a predetermined shooting image acquired at the time of predetermined shooting, the predetermined threshold is set to a first value and acquired by shooting after the predetermined shooting. When detecting the face candidate from the captured image, the face candidate having the lowest matching degree among the face candidates detected from the predetermined photographing image and determined to be the true face Threshold setting means for setting a second value capable of detecting a candidate to the predetermined threshold value is provided.

  “Face components” refers to components included in the face, specifically the eyes of the eyes, the corners of the eyes, the sides of the left and right nostrils, the left and right mouths, and the center of the mouth. It can be a component. Here, when the face candidate is a true face, only one face component candidate is detected at a position where the face component is located, and a plurality of face component candidates are detected in a manner that varies around the face component. There are many. For this reason, in the present invention, one or more face component candidate candidates are detected for one face component.

  As the “second threshold value that can detect the face candidate with the lowest matching degree”, for example, the face candidate with the lowest matching degree among the face candidates determined as the true face detected from the predetermined shooting image A value that is smaller than the matching degree at the time of detecting the image and that is greater than the matching degree for a face candidate having a relatively high matching degree among face candidates that are not determined to be true faces in the predetermined shooting image. do it.

  In the photographing apparatus according to the present invention, the predetermined photographing may be initial photographing or photographing at a predetermined interval.

  Further, in the photographing apparatus according to the present invention, when the determination unit determines whether or not the face candidate is the true face based on the position, each face component in the face candidate region is determined. It is good also as a means which calculates the positional likelihood with respect to the said corresponding face component of a candidate, and determines whether the said face candidate is the said true face based on this positional likelihood.

  Further, in the photographing apparatus according to the present invention, when the determination unit determines whether or not the face candidate is the true face based on the position, each face component in the face candidate region is determined. The likelihood of the positional relationship with respect to other face components other than the corresponding face component is calculated, and it is determined whether or not the face candidate is the true face based on the likelihood of the positional relationship. It is good also as a means to do.

  In the photographing apparatus according to the present invention, when the determination unit determines whether or not the face candidate is the true face based on the position, each face component in the face candidate region is determined. And a means for determining whether or not the face candidate is the true face based on the normalized position of each face component.

  “Normalize a face candidate” means to position a face component candidate at a position where it should be in a face candidate region. Specifically, by performing an affine transformation on the image in the face candidate region, each face component can be scaled, translated, and rotated, so that the position of each face component candidate can be positioned as it should be. it can.

The shooting method according to the present invention continuously acquires images by continuous shooting,
A detection frame of a predetermined size is moved on the image, a feature amount is calculated from the image in the detection frame for each moved position, and a matching degree between the feature amount and a predetermined face feature amount is calculated. , Detecting the image at the position of the detection frame as a face candidate when the matching degree is equal to or greater than a predetermined threshold value,
Detecting at least one face component candidate included in the face candidate for each face component;
In determining whether or not the face candidate is a true face based on at least one of the number and position of the face component candidates detected for each face component,
When the face candidate is detected from a predetermined shooting image acquired at the time of predetermined shooting, the predetermined threshold is set to a first value and acquired by shooting after the predetermined shooting. When detecting the face candidate from the captured image, the face candidate having the lowest matching degree among the face candidates detected from the predetermined photographing image and determined to be the true face A second value capable of detecting a candidate is set to the predetermined threshold value.

  In addition, you may provide as a program for making a computer perform the imaging | photography method by this invention.

  According to the photographing apparatus and method of the present invention, when a face candidate is detected from a predetermined photographing image acquired at a predetermined photographing, a predetermined threshold value for comparing the matching degree is set to the first value. Is done. In addition, when detecting a face candidate from an image acquired by shooting after a predetermined shooting, among the face candidates detected from the image at the time of the predetermined shooting and determined to be a true face, The second value that can detect the face candidate with the lowest matching degree is set to a predetermined threshold value. For this reason, it is possible to reduce the number of face candidates detected from images acquired by photographing after predetermined photographing.

  Here, the face includes face components such as eyes, nose and mouth, and when the face candidate is a true face, there are many face component candidates detected for one face component. Become. In addition, when the face candidate is a true face, the face component candidate exists at the position of the corresponding face component. Therefore, by determining whether or not the face candidate is a true face based on at least one of the number and position of face component candidates for each face component included in the face candidate, the true face is determined from the face candidate. It can be detected with high accuracy.

  However, it takes a long time to calculate whether or not the face is a true face based on the number and / or positions of the face component candidate candidates.

  In the present invention, since the number of face candidates subjected to a determination process with a large amount of calculation can be reduced for images obtained by shooting after a predetermined shooting, the face can be accurately obtained while reducing the amount of calculation. A true face can be detected from the candidates.

  Note that by normalizing each face candidate so that the position of each face component candidate becomes the position of the corresponding face component, a true face can be detected from the face candidate with higher accuracy.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a schematic block diagram showing the configuration of a digital camera to which the photographing apparatus according to the first embodiment of the present invention is applied. As shown in FIG. 1, the digital camera 1 according to the present embodiment is for operating system 2 such as an operation mode switch, zoom lever, up / down / left / right button, release button, and power switch, and for transmitting operation contents of the operating system 2 to the CPU 40. It has the operation system control part 3 which is an interface part.

  The imaging system 6 includes a focus lens 10a and a zoom lens 10b that constitute the photographing lens 10. Each lens can be moved in the optical axis direction by a focus lens driving unit 11 and a zoom lens driving unit 12 each including a motor and a motor driver. The focus lens drive unit 11 controls the movement of each lens based on the focus drive amount data output from the AF processing unit 30 and the zoom lens drive unit 12 based on the operation amount data of the zoom lever.

  The diaphragm 14 is driven by a diaphragm driving unit 15 including a motor and a motor driver. The aperture drive unit 15 adjusts the aperture diameter based on aperture value data output from the AE / AWB processing unit 31.

  The shutter 16 is a mechanical shutter and is driven by a shutter drive unit 17 including a motor and a motor driver. The shutter driving unit 17 controls the opening and closing of the shutter 16 according to a signal generated by pressing the release button and the shutter speed data output from the AE / AWB processing unit 31.

  A CCD 18 which is an image pickup device is provided behind the optical system. The CCD 18 has a photocathode in which a large number of light receiving elements are two-dimensionally arranged, and subject light that has passed through the optical system forms an image on the photocathode and is photoelectrically converted. In front of the photocathode, a microlens array for condensing light on each pixel and a color filter array in which filters of R, G, and B colors are regularly arranged are arranged. The CCD 18 outputs the charges accumulated for each pixel as a serial analog photographing signal line by line in synchronization with the vertical transfer clock and the horizontal transfer clock supplied from the CCD controller 19. The time for accumulating charges in each pixel, that is, the exposure time, is determined by an electronic shutter drive signal given from the CCD controller 19. The gain of the CCD 18 is adjusted by the CCD control unit 19 so that an analog imaging signal having a predetermined size can be obtained.

  Note that the photographing lens 10, the diaphragm 14, the shutter 16, and the CCD 18 constitute the imaging system 6.

  The analog photographing signal captured from the CCD 18 is input to the analog signal processing unit 20. The analog signal processing unit 20 includes a correlated double sampling circuit (CDS) that removes noise from the analog signal, an auto gain controller (AGC) that adjusts the gain of the analog signal, and an A / D that converts the analog signal into a digital signal. It consists of a converter (ADC). Note that the processing performed by the analog signal processing unit 20 is referred to as analog signal processing. The image data converted into the digital signal is CCD-RAW data having R, G, and B density values for each pixel.

  The timing generator 21 generates a timing signal. By supplying this timing signal to the shutter drive unit 17, the CCD control unit 19, and the analog signal processing unit 20, the release button is operated, the shutter 16 is opened and closed, The capture of the charge of the CCD 18 and the processing of the analog signal processing unit 20 are synchronized.

  The flash control unit 23 causes the flash 24 to emit light during shooting.

  The image input controller 25 writes the CCD-RAW data input from the analog signal processing unit 20 in the frame memory 26.

  The frame memory 26 is a working memory used when various image processing (signal processing) described later is performed on the image data. For example, an SDRAM (Synchronous) that performs data transfer in synchronization with a bus clock signal having a fixed period. Dynamic Random Access Memory) is used.

  The display control unit 27 displays the image data stored in the frame memory 26 on the monitor 28 as a through image, or displays the image data stored in the recording medium 35 on the monitor 28 in the reproduction mode. It is. Note that through images are continuously photographed by the CCD 18 at predetermined time intervals while the photographing mode is selected.

  The AF processing unit 30 and the AE / AWB processing unit 31 determine shooting conditions based on the pre-image. This pre-image is an image represented by image data stored in the frame memory 26 as a result of the CPU 40 having detected a half-press signal generated by half-pressing the release button causing the CCD 18 to perform pre-photographing. is there.

  The AF processing unit 30 detects a focal position based on the pre-image and outputs focus drive amount data (AF processing). As a focus position detection method, for example, a passive method that detects a focus position using a feature that the contrast of an image is high when a desired subject is in focus can be considered.

  The AE / AWB processing unit 31 measures subject brightness based on the pre-image, determines ISO sensitivity, aperture value, shutter speed, and the like based on the measured subject brightness, and ISO sensitivity data, aperture value data, and shutter speed data. Is determined as an exposure setting value (AE process), and white balance at the time of shooting is automatically adjusted (AWB process). The exposure and white balance can be set manually by the photographer of the digital camera 1 when the shooting mode is set to the manual mode. Even when the exposure and white balance are set automatically, the photographer can manually adjust the exposure and white balance by giving an instruction from the operation system 2.

  The image processing unit 32 performs image quality correction processing such as gradation correction, sharpness correction, and color correction on the image data of the main image, CCD-RAW data as Y data that is a luminance signal, and Cb data that is a blue color difference signal. Then, YC processing for converting into YC data composed of Cr data which is a red color difference signal is performed. The main image is an image based on image data that is captured from the CCD 18 by main shooting performed when the release button is fully pressed and stored in the frame memory 26 via the analog signal processing unit 20 and the image input controller 25. is there. Although the upper limit of the number of pixels of the main image is determined by the number of pixels of the CCD 18, for example, the number of recording pixels can be changed by setting such as fine and normal. On the other hand, the number of images of the through image and the pre-image is smaller than that of the main image.

  The compression / decompression processing unit 33 performs a compression process in a compression format such as JPEG on the image data of the main image that has been corrected and converted by the image processing unit 32 to generate an image file. A tag storing incidental information such as shooting date and time is added to the image file based on the Exif format or the like. In the reproduction mode, the compression / decompression processing unit 33 reads a compressed image file from the recording medium 35 and performs decompression processing. The decompressed image data is output to the monitor 28, and an image of the image data is displayed.

  The media control unit 34 controls the writing and reading of the image file by accessing the recording medium 35.

  The internal memory 36 stores various constants set in the digital camera 1, programs executed by the CPU 40, and the like.

  The face detection unit 37 detects all face candidates included in an image acquired by shooting. Note that the image may be any of a through image, a pre-image, and a main image, but in this embodiment, face candidates are detected from the through image. Here, as a method for detecting a face, a detection frame having a certain size is moved little by little on the image, a feature amount is calculated from an image in the detection frame for each moved position, and a predetermined face feature amount is obtained. Is used, and a detection frame position where the matching degree is equal to or greater than the threshold Th0 is detected as a face candidate. It should be noted that face candidates having different sizes can be detected by changing the size of the detection frame. The threshold value Th0 is set by a threshold value setting unit 42 described later.

  Then, the face detection unit 37 can detect the face candidates F1 to F5 surrounded by the rectangular detection frame from the image G1 as shown in FIG. In FIG. 2, since the face candidates are detected, an area surrounded by the detection frame is included even in a portion where no face exists.

  The face component detection unit 38 detects a face component candidate that is a candidate for a plurality of face components included in the face candidate. In the present embodiment, the nine facial components of the eye corners K1 and K2, the eyes of the eyes K3 and K4, the sides of the right and left nostrils K5 and K6, the left and right mouths K7 and K8, and the central part K9 of the mouth Assume that face component candidate candidates for K1 to K9 are detected. Specifically, the pattern of each rectangular face component is moved little by little on the image within the region of the face candidate to be processed, the matching degree is calculated for each moved position, and the matching degree is determined in advance. The coordinates of the position of the pattern that is equal to or greater than the threshold Th1 are detected as face component candidate. The coordinates are the coordinates in the face candidate when the upper left corner of the area in the face candidate is the origin.

  Here, when the face candidate is a true face, if a position of a pattern having a matching degree equal to or greater than the threshold Th1 is detected as a face component candidate, the face component candidates are positions of corresponding face components K1 to K9. Are often detected in a distributed manner around the corresponding face components K1 to K9. Therefore, the face component detection unit 38 detects one or more face component candidates for each face component.

  Here, when all nine face components K1 to K9 are included in the face candidate, as shown in FIG. 3A, the corners K1 and K2 of the eyes, the heads K3 and K4 of the eyes, and the right and left nose Face component candidate candidates corresponding to the nine face component parts of the hole sides K5 and K6, the left and right mouths K7 and K8, and the center part K9 of the mouth are detected. Further, for example, a plurality of face component candidate candidates are detected for the left eye as indicated by the crosses in FIG.

  When no face component candidate with a matching degree equal to or greater than the threshold Th1 is detected, it is assumed that no corresponding face component candidate has been detected.

  The determination unit 39 determines whether or not all face candidates detected by the face detection unit 37 are true faces based on the number of face component parts candidates for each face component detected by the face component detection unit 38. A face candidate determined to be a true face is detected as a true face. Specifically, for the face candidates to be processed among all the face candidates, the total number N1 to N9 of face component candidates for each of the nine face component parts K1 to K9 is calculated. Nsum which is an addition value of N9 is calculated. When the addition value Nsum is equal to or greater than the threshold value Th2, the face candidate to be processed is determined to be a true face, and the face candidate is detected as a true face. When the addition value Nsum is less than the threshold value Th2, the face candidate to be processed is determined as a non-face.

  The total number N1 to N9 of face component candidates for each of the nine face component parts K1 to K9 is plotted in a 9-dimensional space, and a hyperplane or hypersurface that defines a threshold in the 9-dimensional space is set. Depending on whether the plotted total number N1 to N9 is on the side of the hyperplane or hypersurface that defines the threshold, it is determined whether the face candidate to be processed is a true face. Good. Here, for the sake of simplicity, when the face components used for the determination are only the left and right mouths K7 and K8 and the center part K9 of the mouth, the total number N7 to N9 is plotted in a three-dimensional space. FIG. 4 is a diagram showing a state in which the total number N7 to N9 is plotted in a three-dimensional space. First, if the total number N7 to N9 is plotted as shown in FIG. 4A, the position X1 (N7, N8, N9) of the plot is above the hyperplane A1 (that is, the value) On the larger side). Therefore, when a plot is made as shown in FIG. 4A, the face candidate to be processed is determined as a true face.

  On the other hand, if the total number N7 to N9 is plotted as shown in FIG. 4B, the position X2 (N7, N8, N9) of the plot is lower than the hyperplane A1 for setting the threshold (that is, The value is on the smaller side. Therefore, when a plot is made as shown in FIG. 4B, it is determined that the face candidate to be processed is not a true face.

  It is determined whether the total number N1 to N9 exceeds the threshold value Th3, and when the total number exceeding the threshold value Th3 further exceeds the threshold value Th4, the face candidates to be processed are determined. You may determine that it is a true face.

  The threshold setting unit 42 sets a threshold Th0 to be compared with the matching degree when the face detection unit 37 detects a face candidate. First, when detecting face candidates from an image (through image) acquired by the first shooting, the threshold value Th0 is set to a relatively low value Z0 so that more face candidates are detected. Thereby, the face detection part 37 can detect the face candidates F1-F10 enclosed by the rectangular detection frame from the image G1, for example, as shown to Fig.5 (a). In FIG. 5A, since face candidates are detected, a region surrounded by a detection frame is included even in a portion where no face exists.

  On the other hand, when detecting face candidates from images acquired by the second and subsequent shootings, the threshold Th0 is set to a value larger than the value Z0 in order to reduce the number of face candidates detected compared to the first shooting. Set to Z1. Specifically, it is smaller than the matching degree Z2 when the face candidate with the lowest matching degree is detected from the face candidates determined as the true face detected from the image acquired by the first shooting. Among the face candidates that are detected from the image acquired by the second shooting and are not determined to be true faces, the threshold value is set to a value Z1 that is greater than the matching degree Z3 for the face candidate having the highest matching degree. Set the value Th0.

  FIG. 6 is a diagram for explaining the setting of the threshold value Th0. In FIG. 6, the horizontal axis indicates the face candidates F1 to F10 detected from the image acquired by the first shooting, and the vertical axis indicates the matching degree. As shown in FIG. 6, among the face candidates F1 to F10, those determined as true faces are the face candidates F1 to F3. Of the face candidates F1 to F3 determined to be true faces, the face candidate F3 has the smallest matching degree, and the matching degree is Z2. On the other hand, the face candidates F4 to F10 are determined to be non-faces. Among the face candidates F4 to F10 determined to be non-faces, the face candidate F8 has the largest matching degree, and the matching degree is Z3.

  Therefore, when the face detection unit 37 detects a face candidate from the images acquired by the second and subsequent photographing, the threshold setting unit 42 determines the matching degree Z2 of the face candidate F3 and the matching degree Z3 of the face candidate F8. The threshold value Th0 is set to an intermediate value Z1 (for example, Z1 = (Z2 + Z3) / 2). Thus, by setting the threshold value Th0 to the value Z1, only the face candidates F1 to F3 are detected from the image acquired by the first shooting. Further, as shown in FIG. 5B, for example, face candidates F1 ′ to F3 ′ corresponding to the face candidates F1 to F3 shown in FIG. 5A are detected from the images acquired by the second and subsequent shootings. Is done. Note that the face candidate F4 ′ corresponding to the face candidate F8 may be detected.

  Note that, in the face candidates F4 to F10 determined as non-faces in the above, the threshold value Th0 of the threshold value Th0 is set using the matching degree Z3 of the face candidate F8 having the highest matching degree. In the obtained face candidates F4 to F10, the threshold value Th0 may be set to an intermediate value between the matching degrees of the face candidates F9 and F10 having the second or third largest matching degree and the matching degree Z2 of the face candidate F3. Good.

  In addition, during shooting a through image, there is a high possibility that the subject is changed, and in this case, the face candidates to be detected change. In the present embodiment, the threshold setting unit 42 sets the threshold Th0 to a value Z0 every time a predetermined shooting interval has elapsed since the first shooting, and a face from an image acquired by the next shooting after that shooting. A new threshold value Th0 for candidate detection is reset. For example, when 30 through images are taken per second, the threshold Th0 is set to the value Z0 every 30 shots, and face candidates are detected from images acquired by subsequent shootings. Reset the threshold Th0.

  The CPU 40 controls each part of the main body of the digital camera 1 in accordance with signals from various processing units such as the operation system 2 and the AF processing unit 30. Further, the CPU 40 controls the face detection unit 37, the face component detection unit 38, and the determination unit 39 so as to detect a true face from each through image during shooting of the through image. When the determination unit 39 detects a true face, the CPU 40 instructs the display control unit 27 to display a through image by surrounding the detected true face with rectangular areas A1 to A3 as shown in FIG. I do. The rectangular area corresponds to a detection frame for face candidates detected by the face detection unit 37.

  The data bus 41 is connected to various processing units, the frame memory 26, the CPU 40, and the like, and exchanges digital image data and various instructions.

  Next, processing performed in the first embodiment will be described. FIG. 8 is a flowchart showing the processing performed in the first embodiment. When the operation mode of the digital camera 1 is set to the shooting mode, the CPU 40 starts processing and takes a through image (step ST1). Then, the threshold setting unit 42 performs threshold setting processing (step ST2).

  FIG. 9 is a flowchart of the threshold setting process. First, it is determined whether or not a through image is captured for the first time (step ST11). If step ST11 is affirmed, a threshold value Th0 is set to a value Z0 (step ST12), and the process ends. If step ST11 is negative, it is determined whether or not the through image has been shot after a predetermined shooting interval has elapsed since the previous threshold Th0 was set to the value Z0 (step ST13). Is affirmed, the process proceeds to step ST12, the threshold value Th0 is set to the value Z0, and the process ends.

  On the other hand, if step ST13 is negative, the threshold Th0 is set to the value Z1 (step ST14), and the process ends.

  Returning to FIG. 8, following step ST2, the face detection unit 37 detects all face candidates included in the through image (step ST3). Next, the face component detection unit 38 detects a face component candidate for each face component from the face candidates to be processed using the i-th face candidate as a process target face candidate (step ST4). The initial value of i is 1. Further, the processing order may be performed in order from the face candidate existing on the left side toward the right side on the through image, for example.

  Then, the determination unit 39 determines whether or not the addition value Nsum of the total number of face component candidates for each face component detected by the face component detection unit 38 is greater than or equal to a threshold Th2 (step ST5). If step ST5 is positive, the face candidate to be processed is determined as a true face and detected (step ST6). On the other hand, if step ST5 is negative, the face candidate to be processed is determined as a non-face (step ST7).

  Subsequent to steps ST6 and 7, the CPU 40 determines whether or not the determination unit 39 has completed the determination for all face candidates (step ST8). If step ST8 is negative, 1 is added to i (step ST8). ST9), the process returns to step ST4. If step ST8 is affirmed, a through image in which the true face is surrounded by a rectangular area is displayed on the monitor 28 (step ST10), and the process returns to step ST1.

  As described above, in the first embodiment, a true face is detected from face candidates based on the number of detected face component candidate candidates. Here, the face includes face components such as eyes, nose and mouth, and when the face candidate is a true face, there are many face component candidates detected for one face component. Become. Therefore, it is possible to accurately detect the true face from the face candidates by determining whether the face candidate is a true face based on the number of face component candidates for each face component included in the face candidate. Can do.

  However, it takes a long time to determine whether or not the face is a true face based on the number of face component candidates. In the first embodiment, the value of the threshold value Th0 for face detection is increased for images acquired by shooting after the first shooting and after shooting every time a predetermined shooting interval elapses. The number of detected face candidates can be reduced. Therefore, it is possible to reduce the number of face candidates that are subjected to a determination process with a large amount of calculation, and thereby it is possible to detect a true face with high accuracy while reducing the amount of calculation.

  In the first embodiment, the corners K1 and K2 of the eyes, the eyes K3 and K4 of the eyes, the sides K5 and K6 of the right and left nostrils, the left and right mouths K7 and K8, and the center 9 of the mouth K9. Although individual face components are detected, it is not necessary to detect all of them, and one or more face component candidates of these face components may be detected. In this case, the threshold value Th2 to be compared with the total addition value Nsum may be changed according to the number of face components to be detected. When only one face component is detected, it is preferable to detect any one of the corners of both eyes and the eyes of both eyes. The face components to be detected are not limited to the corners of both eyes, the eyes of both eyes, the sides of the left and right nostrils, the left and right mouths, and the center of the mouth. Any components can be used as long as they are components.

  Next, a second embodiment of the present invention will be described. In the second embodiment, the process performed by the determination unit 39 is only different from that in the first embodiment, and a detailed description of the configuration is omitted here.

  In the second embodiment, the determination unit (39A because it is different from the first embodiment) determines the positional likelihood of the face component candidate for each face component detected by the face component detection unit 38. It is calculated, and it is determined whether the face candidate is a true face based on the positional likelihood. Here, the positional likelihood is a probability representing how much the detected face component candidate is located at the position of the corresponding face component that should be in the face candidate region.

  Here, in the present embodiment, probability distributions representing the existence probabilities in the face candidates of each face component for nine types of face components are obtained in advance.

  FIG. 10 is a diagram showing a probability distribution representing the presence probability of a face component. The probability distribution shown in FIG. 10 shows that when the detection frame in which the face candidate is detected is normalized to a predetermined size, the eye corners K1 and K2, the eye heads K3 and K4 of the eyes in the detection frame, 9 represents the probability distribution of the existence probabilities of the nine face components of the nostril sides K5 and K6, the left and right mouths K7 and K8, and the central part K9 of the mouth. Note that the circles B1 to B9 in FIG. 10 are the eye corners K1 and K2 of the eyes of the face candidates, the eyes K3 and K4 of the eyes, the sides of the right and left nostrils K5 and K6, the left and right mouths K7 and K8, and the mouth. FIG. 10 is a probability distribution representing the existence probability of the central portion K9, and when the paper plane in FIG. 10 is the XY plane and the direction perpendicular to the paper plane is the Z direction, the Z direction indicates each face as shown in the probability distribution profile of FIG. It indicates the existence probability of the component. Therefore, the closer to the center of each circle in FIG. 10, the higher the probability of existence of each face component.

  The probability distribution may be obtained in advance using a large number of face sample images.

  The determination unit 39A normalizes each face candidate detected by the face detection unit 37 to the certain size, and the probability of the corresponding face component for each face component candidate for each face component in the normalized face candidate The existence probability is calculated as a positional likelihood with reference to the distribution. Specifically, for each face component candidate, a position in the vicinity of the probability distribution representing the existence probability of the corresponding face component is obtained, and the existence probability at that position is calculated as a positional likelihood. Thereby, for example, when the left eye corner candidates 1 to 4 are at positions C1 to C4 in the vicinity of the probability distribution B1 shown in FIG. 12, the likelihood of the left eye eye candidate 1 at the position C1 as shown in FIG. For each face, 0%, likelihood of left eye corner candidate 2 at position C2 is 2%, likelihood of left eye corner candidate 3 at position C3 is 9%, and likelihood of left eye corner candidate 4 at position C4 is 17%. The positional likelihood of the component candidate is determined.

  Further, the determination unit 39A calculates an average value of the positional likelihood of the face component candidate for each face component. FIG. 14 is a diagram illustrating an average value of the positional likelihood of face component candidates for each face component for two face candidates. Then, for the face candidate to be processed, it is determined whether or not the number of face components whose average positional likelihood is equal to or greater than the threshold Th5 is equal to or greater than the threshold Th6, and this determination is affirmed. In such a case, the face candidate to be processed is determined to be a true face and detected. For example, if 13% is used as the threshold Th5 and 5 is used as the threshold Th6 since nine face components are used in the present embodiment, the position candidate 1 shown in FIG. There are six face components whose average likelihood value is equal to or greater than the threshold Th5, the left eye corner, the left eye head, the right eye head, the left nasal side, the right nose side, and the right mouth, and the number is greater than the threshold Th6. Therefore, the candidate face 1 to be processed is determined as a true face and detected. On the other hand, since face candidate 2 has 0 face components whose average likelihood is equal to or greater than threshold value Th5, face candidate 2 is not determined to be a true face.

  Next, processing performed in the second embodiment will be described. FIG. 15 is a flowchart showing processing performed in the second embodiment. When the operation mode of the digital camera 1 is set to the shooting mode, the CPU 40 starts processing and takes a through image (step ST21). Then, the threshold setting unit 42 performs threshold setting processing (step ST22). Subsequently, the face detection unit 37 detects all face candidates included in the through image (step ST23). Next, the face component detection unit 38 detects a face component candidate for each face component from the face candidates to be processed using the i-th face candidate as a process target face candidate (step ST24).

  Then, the determination unit 39A calculates the positional likelihood of the facial component candidate for each facial component (step ST25), and determines the facial component of which the average positional likelihood is equal to or greater than the threshold Th5. It is determined whether or not the number is greater than or equal to threshold value Th6 (step ST26). If step ST26 is affirmed, the face candidate to be processed is determined as a true face and detected (step ST27). On the other hand, if step ST26 is negative, the candidate face to be processed is determined as a non-face (step ST28).

  Subsequent to steps ST27 and 28, the CPU 40 determines whether or not the determination unit 39A has finished the determination for all face candidates (step ST29). If step ST29 is negative, 1 is added to i (step ST29). ST30), the process returns to step ST24. If step ST29 is affirmed, a through image in which the true face is surrounded by a rectangular area is displayed on the monitor 28 (step ST31), and the process returns to step ST21.

  As described above, in the second embodiment, a true face is detected from each face candidate based on the position of the detected face component candidate, particularly the positional likelihood. Here, the face includes face components such as eyes, nose and mouth, and when the face candidate is a true face, the face component candidate exists at the position of the corresponding face component. It will be. Therefore, the true face can be detected from the face candidate with high accuracy by determining whether or not the face candidate is a true face based on the position of the face component candidate included in the face candidate.

  Further, as in the first embodiment, since the number of face candidates to be subjected to determination processing with a large amount of calculation can be reduced, a true face is detected from the face candidates with high accuracy while reducing the amount of calculation. be able to.

  In the second embodiment, the determination unit 39A calculates the positional likelihood of the face component candidate for each face component, and based on this, determines whether the face candidate is a true face. Although it is determined, the likelihood of the positional relationship of the facial component candidate for each facial component may be calculated, and it may be determined whether the facial candidate is a true face based on the likelihood of the positional relationship. . Hereinafter, this will be described as a third embodiment.

  In the third embodiment, the determination unit (39B because it is different from the first embodiment) sets the face component candidate for each face component detected by the face component detection unit 38 for each face component candidate. Then, the existence probability with respect to the position of another face component is calculated as the likelihood of the positional relationship, and it is determined whether or not the face candidate is a true face based on the calculated likelihood of the positional relationship.

  FIG. 16 is a diagram showing probability distributions of existence probabilities for the other eight face components of the right eye, the eyes of both eyes, the eyes of the left eye, the sides of the left and right nostrils, the left and right mouths, and the central part of the mouth. It is. In FIG. 16, probability distributions B11 to B18 are respectively the right eye corner, the left eye corner, the right eye corner, the left eye corner, the side of the left nostril, the side of the right nostril, the left mouth, The probability distribution of the probability of existence for the right mouth and the central part of the mouth is shown.

  Here, when the target for calculating the likelihood of the positional relationship is the right eye, in the third embodiment, the determination unit 39B determines each face candidate detected by the face detection unit 37 as in the second embodiment. For each right eye-head candidate detected by the face component detection unit 38 in the normalized face candidate, the probability of existence is determined as the likelihood of the temporary positional relationship with reference to the probability distributions B11 to B18. calculate. For example, the likelihood of the temporary positional relationship of the right eye to the left eye corner is 15%, the likelihood of the temporary positional relationship to the right eye corner is 12%, the likelihood of the temporary positional relationship to the left eye corner is 13%, the left 10% likelihood of the temporary positional relationship to the side of the nostril, 19% likelihood of the temporary positional relationship to the side of the right nostril, 13% likelihood of the temporary positional relationship to the left mouth, right The likelihood of the temporary positional relationship is calculated such that the likelihood of the temporary positional relationship with respect to the mouth is 17% and the likelihood of the temporary positional relationship with respect to the central portion of the mouth is 15%.

  Then, the determination unit 39B calculates an average value of the likelihoods of the eight calculated temporary positional relationships, and further calculates an average value for all face component candidate candidates of this average value as a final value of the face component component candidate. It is calculated as the likelihood of a correct positional relationship.

  In the third embodiment, not only the right eye corner, the left eye corner, the right eye corner, the left eye corner, the side of the left nostril, the side of the right nostril, the left mouth, the right The probability distributions of the existence probabilities with respect to the other face components are also obtained for each of the mouth and the central portion of the mouth, and the determination unit 39B determines the likelihood of the positional relationship for the face component candidates of all nine face components. Calculate the degree. Then, the determination unit 39B determines whether or not the number of face component parts for which the likelihood of the positional relationship of the nine face component parts calculated for each face component part is equal to or greater than the threshold value Th7 is equal to or greater than the threshold value Th8. If this determination is affirmative, the processing target face candidate is determined to be a true face and detected.

  Next, processing performed in the third embodiment will be described. FIG. 17 is a flowchart showing processing performed in the third embodiment. When the operation mode of the digital camera 1 is set to the shooting mode, the CPU 40 starts processing and takes a through image (step ST41). Then, the threshold setting unit 42 performs threshold setting processing (step ST42). Subsequently, the face detection unit 37 detects all face candidates included in the through image (step ST43). Next, the face component detection unit 38 detects a face component candidate for each face component from the face candidates to be processed, using the i-th face candidate as a face candidate to be processed (step ST44). The initial value of i is 1.

  Then, the determination unit 39B calculates the likelihood of the positional relationship between the facial component candidates for each facial component (step ST45), and determines the number of facial components whose positional relationship likelihood is equal to or greater than the threshold Th7. It is determined whether or not the threshold value is Th8 or more (step ST46). If step ST46 is affirmed, the face candidate to be processed is determined as a true face and detected (step ST47). On the other hand, if step ST46 is negative, the face candidate to be processed is determined as a non-face (step ST48).

  Subsequent to steps ST47 and 48, the CPU 40 determines whether or not the determination unit 39B has finished the determination for all face candidates (step ST49). If step ST49 is negative, 1 is added to i (step ST49). ST50), the process returns to step ST44. If step ST49 is affirmed, a through image in which the true face is surrounded by a rectangular area is displayed on the monitor 28 (step ST51), and the process returns to step ST41.

  As described above, in the third embodiment, a true face is detected from each face candidate based on the detected position of the face component, particularly the likelihood of the positional relationship. Here, the face includes face components such as eyes, nose and mouth, and when the face candidate is a true face, the face component candidate exists at the position of the corresponding face component. In addition, the positional relationship between the face components is substantially determined. Therefore, by determining whether or not the face candidate is a true face based on the positional relationship of the face component candidate included in the face candidate, the true face can be detected from the face candidate with high accuracy.

  Further, as in the first embodiment, since the number of face candidates to be subjected to determination processing with a large amount of calculation can be reduced, a true face is detected from the face candidates with high accuracy while reducing the amount of calculation. be able to.

  In the third embodiment, the likelihoods of all the positional relationships of the nine types of face components are calculated, and the facial components whose positional relationship likelihood is equal to or greater than the threshold value Th7 are threshold values Th8. Whether or not the face candidate is a true face is determined based on whether or not it is above, but it is not necessary to use all nine types of face components, and the positional relationship of at least one face component You may make it determine whether a face candidate is a true face based on likelihood.

  In the second and third embodiments, if the face component candidate of the detected face candidate is located on the probability distribution of each corresponding face component, the positional likelihood and Although the likelihood of the positional relationship can be calculated, as shown in FIG. 18, the position of each face component candidate of the face candidate (indicated by x in the figure) is different from the probability distribution of the position of the face component that should be originally located. The likelihood cannot be calculated with high accuracy, and as a result, it cannot be accurately determined whether or not the face candidate is a true face. For this reason, it is preferable to normalize face candidates so that the detected face component candidate is located in the probability distribution. Hereinafter, this will be described as a fourth embodiment.

  In the fourth embodiment, in order to normalize a face candidate, one of the face component candidates in the face candidate is set to the center of the probability distribution of the corresponding face component (that is, the most probable probability). Affine transformation is performed on the face candidate image so that the image matches the position of the face. The affine transformation is a transformation in which any three points on the plane are moved to any three points by enlarging / reducing, translating and rotating, and is specifically represented by the following formula (1).

x ′ = a1 · x + b1 · y + d1
y ′ = a2 · x + b2 · y + d2 (1)
In order to calculate the coefficients a1, a2, b1, b2, d1, d2 of the affine transformation from the equation (1), the coordinates of three points corresponding to each other in the probability distribution of the face candidate and the face component are required. Become. Here, in the probability distribution of the face candidates and the face component parts, considering an XY coordinate system with the lower left corner as the origin as shown in FIG. 18, the face component candidate candidates P1 to P9 are positioned at the centers of the probability distributions B1 to B9. It is necessary to set the coefficient of affine transformation so that it does. In the fourth embodiment, the face component candidate with the highest matching degree among the at least one face component part detected by the face component detection unit 38 for each face component is the face representing the face component candidate. Select as the component candidate candidates P1 to P9, and match the top three face component candidates with the highest matching degree among the selected nine face component candidates P1 to P9 with the center of the probability distribution of the corresponding face component The affine transformation coefficients a1, a2, b1, b2, d1, and d2 may be calculated so that

  For example, when the matching degrees of the face component candidate candidates P1 to P9 shown in FIG. 18 are P1> P2> P3> P4> P5..., The face component candidate candidates P1, P2, P3 are assigned to the corresponding face component parts. Affine transformation coefficients a1, a2, b1, b2, d1, d2 are calculated so as to coincide with the centers of the probability distributions B1, B2, B3, respectively.

  In order to calculate the coefficient of affine transformation, not only the coordinates of three points but also the coordinates of four or more points may be used. For example, the affine transformation coefficient may be calculated so that all nine face component candidate candidates P1 to P9 coincide with the centers of the corresponding face component probability distributions B1 to B9. In this case, the coefficient of the affine transformation using the least square method so that the error between the coordinates of the nine face component candidates P1 to P9 after the transformation and the coordinates of the center positions of the probability distributions B1 to B9 is minimized. May be calculated.

  Next, processing performed in the fourth embodiment will be described. FIG. 19 is a flowchart showing processing performed in the fourth embodiment. In addition, although the process at the time of applying 4th Embodiment to 2nd Embodiment is demonstrated here, it can apply similarly also to 3rd Embodiment.

  When the operation mode of the digital camera 1 is set to the shooting mode, the CPU 40 starts processing and takes a through image (step ST61). Then, the threshold setting unit 42 performs threshold setting processing (step ST62). Subsequently, the face detection unit 37 detects all face candidates included in the through image (step ST63). Next, the face component detection unit 38 detects the face component candidate for each face component from the face candidates to be processed, using the i-th face candidate as the face candidate to be processed (step ST64). The initial value of i is 1.

  Then, the determination unit 39A normalizes the face candidate to be processed (step ST65), and after normalization, calculates the positional likelihood of the face component candidate for each face component (step ST66). It is determined whether or not the number of face components whose average likelihood value is equal to or greater than threshold value Th5 is equal to or greater than threshold value Th6 (step ST67). If step ST67 is positive, the face candidate to be processed is determined as a true face and detected (step ST68). On the other hand, if step ST68 is negative, the face candidate to be processed is determined as a non-face (step ST69).

  Subsequent to steps ST68 and 69, the CPU 40 determines whether or not the determination unit 39A has finished the determination for all face candidates (step ST70). If step ST70 is negative, 1 is added to i (step ST70). ST71), the process returns to step ST64. If step ST70 is affirmed, a through image in which the true face is surrounded by a rectangular area is displayed on the monitor 28 (step ST72), and the process returns to step ST61.

  Thus, in the fourth embodiment, face candidates are affine transformed and normalized so that the position of each face component candidate is located at the position of the corresponding face component in the face candidate region. Therefore, the true face can be detected from the face candidates with higher accuracy.

  Further, as in the first embodiment, since the number of face candidates to be subjected to determination processing with a large amount of calculation can be reduced, a true face is detected from the face candidates with high accuracy while reducing the amount of calculation. be able to.

  In the fourth embodiment, the affine transformation coefficient is calculated for each face candidate and the affine transformation is performed. For all face candidates, the average of the face component candidate selected for each face component is selected. The position may be calculated, and the affine transformation coefficient may be calculated so that the calculated average position matches the center of the probability distribution. Even in this case, the coefficient of affine transformation may be calculated from three face component candidates among the face component candidates selected from the nine face component parts, and the affine transformation coefficient may be calculated from four or more face component candidates. A coefficient may be calculated.

  Further, in the fourth embodiment, the temporary positional likelihood or the temporary positional relationship likelihood is calculated for the face component candidate for each face component before normalization, and the temporary positional likelihood is calculated. Face candidates so that the top predetermined number of face component candidates with the highest likelihood of degree or tentative positional relationship match the position of the corresponding face component (that is, the position at which the existence probability peaks). Normalization may be performed by performing affine transformation.

  Next, a fifth embodiment of the present invention will be described. In the fifth embodiment, the process performed by the determination unit 39 is only different from that in the first embodiment, and a detailed description of the configuration is omitted here.

  In the fifth embodiment, the determination unit (39C because it is different from the first embodiment) determines the face based on the number of face component candidates for each face component detected by the face component detection unit 38. First determination processing for detecting a true face is performed by determining whether the candidate is a true face, a non-face, or an ambiguous face, and the face candidate determined as an ambiguous face by the first determination processing As in the second, third, or fourth embodiment, the second method of detecting the true face by determining whether the face candidate is a true face based on the position of the face component candidate. This is different from the first embodiment in that the determination process is performed.

  In the first determination process, the determination unit 39C according to the fifth embodiment uses the total number of face component parts candidates for each of the nine face component parts K1 to K9 in the same manner as the determination unit 39 according to the first embodiment. N1 to N9 are calculated, and Nsum which is an added value of the total number N1 to N9 is calculated. When the addition value Nsum is equal to or greater than the threshold Th9, the face candidate to be processed is determined to be a true face, and the face candidate is detected as a true face. Further, when the addition value Nsum is greater than or equal to the threshold Th10 and less than the threshold Th9, the face candidate to be processed is determined to be an ambiguous face, and when the addition value Nsum is less than the threshold Th10, the face candidate to be processed. Is determined to be non-face. In addition, the process in the second, third, or fourth embodiment for the face candidate determined as an ambiguous face is performed as the second determination process.

  Next, processing performed in the fifth embodiment will be described. FIG. 20 is a flowchart showing processing performed in the fifth embodiment. When the operation mode of the digital camera 1 is set to the shooting mode, the CPU 40 starts processing and takes a through image (step ST81). Then, the threshold setting unit 42 performs threshold setting processing (step ST82). Subsequently, the face detection unit 37 detects all face candidates included in the through image (step ST83). Next, the face component detection unit 38 detects the face component candidate for each face component from the face candidates to be processed, using the i-th face candidate as the face candidate to be processed (step ST84). The initial value of i is 1.

  Then, the determination unit 39C performs a first determination process (step ST85). First, it is determined whether or not the addition value Nsum of the total number of face component candidates for each face component detected by the face component detection unit 38 is greater than or equal to the threshold Th9 (step ST86), and step ST86 is affirmed. Then, the face candidate to be processed is determined as a true face and detected (step ST87). On the other hand, if step ST86 is negative, it is determined whether or not the addition value Nsum is greater than or equal to threshold value Th10 and less than threshold value Th9 (step ST88). If step ST88 is negative, face candidate to be processed is determined. Is determined to be a non-face (step ST89). If step ST88 is affirmed, a second determination process is performed assuming that the face candidate to be processed is an ambiguous face (step ST90).

  First, as in the second embodiment, the determination unit 39C calculates the positional likelihood of the face component candidate for each face component (step ST91), and the average likelihood value for each face component is calculated. It is determined whether or not the number of face components that are equal to or greater than threshold Th5 is equal to or greater than threshold Th6 (step ST92). Note that the candidate face to be processed may be normalized before step ST91 as in the fourth embodiment. Moreover, you may perform the process of step ST91,92 using the likelihood of positional relationship similarly to the process of step ST45,46 of 3rd Embodiment. If step ST92 is affirmed, the face candidate to be processed is determined as a true face and detected (step ST93). On the other hand, if step ST92 is negative, the face candidate to be processed is determined as a non-face (step ST94).

  Subsequent to steps ST87, 89, 93, and 94, the CPU 40 determines whether or not the determination unit 39C has completed the determination for all face candidates (step ST95), and if step ST95 is negative, i is set to 1. Addition (step ST96) returns to step ST84. If step ST95 is affirmed, a through image in which the true face is surrounded by a rectangular area is displayed on the monitor 28 (step ST97), and the process returns to step ST81.

  Here, when determining whether or not the face candidate is a true face based on the number of face component candidate, and whether or not the face candidate is a true face based on the position of the face component candidate. In the case of determination, the former has less calculation amount. In addition, since face candidates become dark at the time of shooting a dark scene or backlight, the number of detected face component candidates is reduced even if the face candidate is a true face. As a result, the first embodiment If only the process is performed, the true face may be determined as a non-face. Therefore, as in the fifth embodiment, for only face candidates determined to be ambiguous based on the number of face component candidates, whether the face candidate is a true face based on the position of the face component candidate By determining whether or not, the amount of calculation can be further reduced, and the true face can be detected from the face candidates with high accuracy.

  In the fifth embodiment, as the first determination process, the total number N1 to N9 of face component parts candidates for each of the nine face component parts K1 to K9 is plotted in a 9-dimensional space. A hyperplane or hypersurface that defines a threshold value is set in a dimensional space, and the face candidate is true depending on which side of the hyperplane or hypersurface that defines the threshold value the total number N1 to N9 is plotted. You may make it determine whether it is a face, an ambiguous face, and a non-face.

  Moreover, in the said 5th Embodiment, although the 1st determination process and the 2nd determination process are performed in the same determination part 39C, two determination parts which respectively perform a 1st and 2nd determination process are provided. You may make it provide.

  The digital camera according to the embodiment of the present invention has been described above. The computer corresponds to the face detection unit 37, the face component detection unit 38, the determination units 39, 39A to 39C, and the threshold setting unit 42 described above. A program that functions as a means and performs processing as shown in FIGS. 8, 9, 15, 17, 19, and 20 is also one embodiment of the present invention. A computer-readable recording medium in which such a program is recorded is also one embodiment of the present invention.

1 is a schematic block diagram showing the configuration of a digital camera to which a photographing apparatus according to a first embodiment of the present invention is applied. Diagram for explaining detection of face candidates The figure for demonstrating the detection of a face component candidate The figure for demonstrating determination of whether a face candidate is a true face Diagram for explaining setting of threshold (part 1) Diagram for explaining setting of threshold (part 2) Diagram showing a through image with a true face surrounded by a rectangle The flowchart which shows the process performed in 1st Embodiment. Threshold setting process flowchart Diagram showing probability distribution of face component existence probability Diagram showing probability distribution profile The figure which shows the example of the position of the face component candidate in the probability distribution vicinity The figure which shows the positional likelihood calculated about each face component candidate The figure which shows the average value of the positional likelihood of the face component candidate for every face component about two face candidates. The flowchart which shows the process performed in 2nd Embodiment. The figure which shows the probability distribution of the existence probability with respect to the other eight face components of the right eye's eyes, the eyes of both eyes, the eyes of the left eye, the sides of the right and left nostrils, the left and right mouths, and the central part of the mouth. The flowchart which shows the process performed in 3rd Embodiment The figure for demonstrating the shift | offset | difference of the position of a face component The flowchart which shows the process performed in 4th Embodiment The flowchart which shows the process performed in 5th Embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Digital camera 2 Operation system 3 Operation system control part 6 Imaging system 28 Monitor 35 Recording medium 37 Face detection part 38 Face component detection part 39 Determination part 40 CPU
42 Threshold setting section

Claims (8)

Photographing means for continuously acquiring images by continuous photographing;
A detection frame of a predetermined size is moved on the image, a feature amount is calculated from the image in the detection frame for each moved position, and a matching degree between the feature amount and a predetermined face feature amount is calculated. , A face detection unit that detects an image at the position of the detection frame as a face candidate when the matching degree is equal to or greater than a predetermined threshold;
Face component detection means for detecting at least one face component candidate included in the face candidate for each face component;
Determining means for determining whether or not the face candidate is a true face based on at least one of the number and position of the face component candidates detected for each face component;
When the face candidate is detected from a predetermined shooting image acquired at the time of predetermined shooting, the predetermined threshold is set to a first value and acquired by shooting after the predetermined shooting. When detecting the face candidate from the captured image, the face candidate having the lowest matching degree among the face candidates detected from the predetermined photographing image and determined to be the true face An imaging apparatus, comprising: threshold value setting means for setting a second value capable of detecting a candidate to the predetermined threshold value.   The imaging apparatus according to claim 1, wherein the predetermined imaging is an initial imaging.   The photographing apparatus according to claim 1, wherein the predetermined photographing is photographing at a predetermined interval.   When determining whether the face candidate is the true face based on the position, the determining unit determines whether each face component candidate in the face candidate region corresponds to the corresponding face component. The position likelihood is calculated, and based on the position likelihood, it is means for determining whether or not the face candidate is the true face. The imaging device according to 1.   When determining whether the face candidate is the true face or not based on the position, the determining unit determines whether each face component candidate in the face candidate area is other than the corresponding face component. A means for calculating a likelihood of a positional relationship with respect to another facial component and determining whether or not the face candidate is the true face based on the likelihood of the positional relationship. Item 4. The photographing apparatus according to any one of Items 1 to 3.   When determining whether the face candidate is the true face based on the position, the determination unit normalizes each face component in the area of the face candidate, and the normalized each 6. The photographing apparatus according to claim 1, wherein the photographing apparatus is means for determining whether or not the face candidate is the true face based on a position of a face component. Acquire images continuously by continuous shooting,
A detection frame of a predetermined size is moved on the image, a feature amount is calculated from the image in the detection frame for each moved position, and a matching degree between the feature amount and a predetermined face feature amount is calculated. , Detecting the image at the position of the detection frame as a face candidate when the matching degree is equal to or greater than a predetermined threshold value,
Detecting at least one face component candidate included in the face candidate for each face component;
In determining whether or not the face candidate is a true face based on at least one of the number and position of the face component candidates detected for each face component,
When the face candidate is detected from a predetermined shooting image acquired at the time of predetermined shooting, the predetermined threshold is set to a first value and acquired by shooting after the predetermined shooting. When detecting the face candidate from the captured image, the face candidate having the lowest matching degree among the face candidates detected from the predetermined photographing image and determined to be the true face A shooting method, wherein a second value capable of detecting a candidate is set to the predetermined threshold value. The procedure to acquire images continuously by continuous shooting,
A detection frame of a predetermined size is moved on the image, a feature amount is calculated from the image in the detection frame for each moved position, and a matching degree between the feature amount and a predetermined face feature amount is calculated. , A procedure for detecting an image at the position of the detection frame as a face candidate when the matching degree is equal to or greater than a predetermined threshold;
Detecting at least one face component candidate included in the face candidate for each face component;
Determining whether the face candidate is a true face based on at least one of the number and position of the face component candidates detected for each face component; and
When the face candidate is detected from a predetermined shooting image acquired at the time of predetermined shooting, the predetermined threshold is set to a first value and acquired by shooting after the predetermined shooting. When detecting the face candidate from the captured image, the face candidate having the lowest matching degree among the face candidates detected from the predetermined photographing image and determined to be the true face A program for causing a computer to execute an imaging method, comprising: setting a second value capable of detecting a candidate to the predetermined threshold value.

Images