JP4743823B2 - Image processing apparatus, imaging apparatus, and image processing method - Google Patents

Description

  The present invention relates to a technique for performing determination related to a category of a subject such as a face in an input image.

  Conventionally, in the field of image recognition and voice recognition, recognition processing algorithms specialized for a specific recognition target are realized by computer software or hardware using a dedicated parallel image processing processor to detect the recognition target. What to do is known.

  In particular, several documents have been disclosed for detecting a face as a specific recognition target from an image including the face (see, for example, Patent Documents 1 to 5).

  According to one of the techniques, a face area is searched for a template using a standard face for an input image, and then a partial template is used for candidate feature points such as eyes, nostrils, and mouth. Authenticate a person. However, this technology uses a template first to match the entire face to detect the face area, so it is vulnerable to changes in the size of multiple faces and face orientations. It is necessary to prepare a plurality of standard faces corresponding to the orientation of the face and detect them using each of them. However, the template for the entire face is large in size and processing cost.

  According to another technique, an eye and mouth candidate group is obtained from a face image, and a face candidate group obtained by combining them and a previously stored face structure are collated to find a region corresponding to the eye and mouth. In this technique, the number of faces in the input image is one or a small number, the size of the face is also somewhat large, most of the areas in the input image are faces, and an image with a small background is the input image. Is assumed.

  According to another technique, a plurality of eye, nose, and mouth candidates are obtained, and a face is detected from the positional relationship between feature points prepared in advance.

  According to another technique, when examining the degree of coincidence between the shape data of each part of the face and the input image, the shape data is changed, and the search area of each face part is the previously determined part. It is determined based on the positional relationship. In this technique, the shape data of the iris, mouth, nose, etc. are stored, and first two irises are obtained, and then when the mouth, nose, etc. are obtained, the mouth, nose, etc. are determined based on the position of the iris. The search area for facial parts is limited. In other words, this algorithm does not detect the facial parts that make up the face such as the iris (eye), mouth, and nose in parallel, but first finds the iris (eye) and uses the result to It detects a facial part called the nose. In this method, it is assumed that there is only one face in the image and the iris is obtained accurately. If the detected iris is a false detection, other features such as mouth and nose The search area cannot be set correctly.

  According to another technique, an area model in which a plurality of determination element acquisition areas are set is moved in the input image, and at each point, the presence / absence of the determination element is determined in the determination element acquisition area to recognize the face. Is. In this technology, in order to correspond to faces of different sizes and rotated faces, it is necessary to prepare area models of different sizes and rotated area models. When there is no face, many unnecessary calculations are performed.

  In addition, some techniques for recognizing facial expressions in images have also been disclosed (see Non-Patent Documents 1 and 2, for example).

  One of the techniques is based on the premise that a partial face region is accurately cut out from a frame image by visual observation. In other techniques, the rough positioning of the face pattern is automated, but fine positioning by human visual inspection is necessary for positioning the feature points. In another technique (see, for example, Patent Document 6), facial expression elements are coded using muscle movements, nervous system connection relationships, and the like to determine the emotion. However, in this technology, the region of the part necessary for facial expression recognition is fixed, and there is a possibility that the region necessary for recognition may not be included due to changes in the face orientation and movement, and conversely, unnecessary regions are included. This may affect the accuracy of facial expression recognition.

  In addition, a system that recognizes a facial expression by detecting a change corresponding to an action unit of FACS (Facial Action Coding System), which is known as a method for objectively describing facial expression behavior, has been studied.

  In other techniques (see, for example, Patent Document 7), facial expression is estimated in real time, the three-dimensional face model is deformed, and the facial expression is reproduced. In this technique, a face is detected from the difference image between the input image including the face area and the background image not including the face area and the chromaticity indicating the skin color, and after the detected face area is binarized, Detect contour lines. Then, in the area within the contour line, determine the position of the eyes and mouth, determine the rotation angle of the face from the position of the eyes and mouth, correct the rotation, perform 2D discrete cosine transform, estimate the facial expression, Based on the amount of change in the spatial frequency component, the facial expression is reproduced by converting the 3D face model. However, skin color detection is easily affected by illumination fluctuations and the background. Therefore, in this technique, there is a high possibility that the subject is not detected or erroneously detected in the first skin color extraction process.

  As a technique for identifying an individual from a face image, the Eigenface (Eigenface) method by Turk et al. Is well known (see, for example, Non-Patent Documents 3 and 4). In this method, principal component analysis is performed on a set of gray value vectors of a large number of face images, orthonormal bases called eigenfaces are obtained in advance, and these bases are used to calculate the input face image. Dimensionally compressed face patterns are obtained by applying Karhunen-Loeve expansion to the gray value vector. The dimension-compressed pattern is used as a feature vector for identification.

  As one of methods for actually identifying an individual using a feature vector for identification, in the above document, the dimension pattern of the input image and the dimension-compressed face of each individual held in the above document are described. A method for obtaining a distance from a pattern and identifying a class to which the pattern indicating the closest distance belongs from a class to which the inputted face image belongs, that is, an individual is shown. However, this method basically uses some method to detect the position of the face in the image, and then obtains a face image that has been subjected to size normalization and rotation correction for that face region, The corrected face image is used as an input image.

Further, an image processing method capable of recognizing a face in real time has been disclosed as a prior art (see, for example, Patent Document 8). In this method, first, an arbitrary area is cut out from the input image, and it is determined whether or not the area is a face area. Next, when the area is a face area, matching is performed between the face image subjected to affine transformation and contrast correction and the registered face in the learning database, and the probability of being the same person is estimated. Based on the probability, the person who is most likely to be the same as the input face is output from the registered persons.
JP-A-9-251534 Japanese Patent No. 2776714 JP-A-9-44676 Japanese Patent No. 2973676 Japanese Patent Laid-Open No. 11-283036 Japanese Patent No. 2573126 Japanese Patent No. 30621181 JP 2003-271958 A As introduced in Shigeru Akamatsu, “Recognition of Facial Expressions by Humans and Computers III”, IEICE Vol.85 No.12 pp.936-941, Dec. 2002, G. Donate, TJSejnowski, et.al, “Classifying Facial Actions” IEEE Trans.PAMI, vol.21, no.10, Oct, 1999 Y.Tian, T.Kaneda, and JFCohn “Recognizing Action Units for Facial Expression Analysis” IEEE Tran.PAMI vol.23, no.2, Feb.2001 Shigeru Akamatsu, “Face Recognition by Computer-Survey-”, IEICE Vol.80 No.8 pp.2031-2046, Aug. 1997 M. Turk, A. Pentland, “Eigenfaces for recognition” J. Cognitive Neurosci., Vol.3, no.1, pp. 71-86, Mar, 1991

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a technique for easily discriminating who is a face in an image and the facial expression of the face.

  It is another object of the present invention to deal with changes in the position and orientation of a subject by a simple method in detection of a face in an image, facial expression discrimination, and individual discrimination.

  In order to achieve the object of the present invention, for example, an image processing apparatus of the present invention comprises the following arrangement.

That is, an input means for inputting an image including a face ;
Detecting a plurality of local features representing edges from the image by the input means is inputted, and detects the facial features of a combination of a plurality of local features that the detected face area to identify the areas of said pigment in said image Specific means,
The relative position of the local feature of each of the region of the face detected by the face region identifying means, the difference between the relative position of the local feature of each against a face image set in advance as a reference, the face with And a discriminating means for discriminating the category to which the data belongs .

  In order to achieve the object of the present invention, for example, an image processing apparatus of the present invention comprises the following arrangement.

That is, input means for continuously inputting frame images including a face;
A face for detecting a plurality of local features representing an edge from a frame image input by the input means, detecting a facial feature from a combination of the detected plurality of local features , and identifying a facial region in the frame image Area identification means;
In the image of the second frame, which is a frame after the first frame, corresponding to the position of the face area specified by the face area specifying means in the image of the first frame input by the input means. In the region, the facial expression is determined based on the difference between the relative position of each of the local features detected by the face region specifying unit and the relative position of each of the local features with respect to a face image set in advance as a reference. And a discriminating means.

  In order to achieve the object of the present invention, for example, an image processing apparatus of the present invention comprises the following arrangement.

That is, an input means for inputting an image including a face;
Face region specification for detecting a plurality of local features representing an edge from an image input by the input means, detecting a facial feature from a combination of the detected plurality of local features , and specifying a face region in the image Means,
The input means using the detection results of the local features in the face area detected by the face area specifying means and the detection results of the local features obtained in advance from the images of the faces. A first determination unit that determines who the face in the image input is, a relative position of each local feature in the face region specified by the face region specifying unit, and a reference in advance And a second discriminating means for discriminating the facial expression using the difference between the relative position of each of the local features with respect to the face image set as.

  In order to achieve the object of the present invention, for example, an image processing method of the present invention comprises the following arrangement.

That is, an input process for inputting an image including a face ;
Detecting a plurality of local features representing an edge from the image input in the input step, by detecting the facial features of a combination of a plurality of local features that the detected face area to identify the areas of said pigment in said image Specific process,
The relative position of the local feature of each of the region of the face detected by the face region specifying step, the difference between the relative position of the local feature of each against a face image set in advance as a reference, the face with characterized in that it comprises a discrimination step of discriminating a category to belong.

  In order to achieve the object of the present invention, for example, an image processing method of the present invention comprises the following arrangement.

That is, an input process for continuously inputting frame images including a face;
A face for detecting a plurality of local features representing an edge from the frame image input in the input step, detecting a facial feature from a combination of the detected plurality of local features , and identifying a facial region in the frame image Region identification process;
In an image of a second frame, which is a frame after the first frame, corresponding to the position of the face region specified in the face region specifying step in the image of the first frame input in the input step In the region, the facial expression is determined based on the difference between the relative position of each local feature detected in the face region specifying step and the relative position of each local feature with respect to a face image set in advance as a reference. And a discriminating step.

  In order to achieve the object of the present invention, for example, an image processing method of the present invention comprises the following arrangement.

That is, an input process for inputting an image including a face;
Face region specification for detecting a plurality of local features representing edges from the image input in the input step, detecting a facial feature from a combination of the detected plurality of local features , and specifying a face region in the image Process,
The input step using the detection result of each local feature in the face region detected in the face region specifying step and the detection result of each local feature obtained in advance from each face image A first discrimination step for discriminating who the face in the image inputted in step 1 is, a relative position of each local feature in the face region identified in the face region identification step, and a reference in advance And a second discrimination step of discriminating the facial expression using the difference between the relative position of each local feature with respect to the face image set as.

  In order to achieve the object of the present invention, for example, an imaging apparatus of the present invention comprises the following arrangement.

That is, the image processing apparatus includes the image processing device, and when the determined facial expression is a preset facial expression, the image processing apparatus includes an imaging unit that captures an image input to the input unit.

  According to the configuration of the present invention, it is possible to easily determine the category of the subject in the image, for example, when the subject is a face, who the person is, and the facial expression.

  In addition, it is possible to cope with changes in the position and orientation of the subject by a simple method in detecting a face in an image, determining an expression, and determining an individual.

  Hereinafter, the present invention will be described in detail according to preferred embodiments with reference to the accompanying drawings.

[First Embodiment]
FIG. 1 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment. The image processing apparatus according to the present embodiment detects a face from an image and discriminates its facial expression, and includes an imaging unit 100, a control unit 101, a face detection unit 102, an intermediate detection result holding unit 103, and a facial expression discrimination unit. 104, an image holding unit 105, a display unit 106, and a recording unit 107. Hereinafter, each part will be described.

  The imaging unit 100 captures an image based on a control signal from the control unit 101 and outputs the captured image (captured image) to the face detection unit 102, the image holding unit 105, the display unit 106, or the recording unit 107. .

  The control unit 101 performs processing for controlling the entire image processing apparatus according to the present embodiment, and includes an imaging unit 100, a face detection unit 102, an intermediate detection result holding unit 103, a facial expression determination unit 104, and an image holding unit. 105, connected to the display unit 106 and the recording unit 107, and controls each unit so that each unit operates at an appropriate timing.

  The face detection unit 102 performs processing for detecting a face region (a region of a face image included in the photographed image) in the photographed image from the imaging unit 101. This processing means that the number of face regions in the photographed image, the coordinate position of the face region in the photographed image, the size of the face region, the amount of rotation in the photographed image of the face region (for example, if the face region is a rectangle, this rectangle is photographed). In other words, a process for obtaining a rotation amount indicating how much the image is inclined in which direction. Note that these pieces of information (the number of face areas in the captured image, the coordinate position of the face area in the captured image, the size of the face area, and the amount of rotation in the captured image of the face area) are collectively referred to as “face area information”. Call it. Therefore, the face area in the captured image can be specified by obtaining the face area information.

  These detection results are output to the facial expression determination unit 104. In addition, an intermediate detection result (to be described later) obtained during the detection process is output to the intermediate detection result holding unit 103.

  The intermediate detection result holding unit 103 holds the intermediate feature detection result output from the face detection unit 102.

  The facial expression determination unit 104 receives the face area information data output from the face detection unit 102 and the intermediate feature detection result data output from the intermediate detection result holding unit 103. Based on these data, all or a part of the captured image (in some cases, only the image of the face area) is read from the image holding unit 105, and the facial expression in the read image is determined by the processing described later. Perform the process.

  The image holding unit 105 temporarily holds the captured image output from the imaging unit 100, and based on the control signal of the control unit 101, all or part of the stored captured image is transferred to the facial expression determination unit 104 or And output to the display unit 106 and the recording unit 107.

  The display unit 106 includes, for example, a CRT or a liquid crystal screen, and displays all or a part of the captured image output from the image holding unit 105 or a captured image captured by the image capturing unit 100.

  The recording unit 107 is configured by a device that records information on a storage medium such as a hard disk drive, a DVD-RAM, or a compact flash (registered trademark), and an image held in the image holding unit 105 or captured by the imaging unit 100. Record the captured image.

  Next, main processing for discriminating facial expressions in a captured image, which is executed by the operations of the above-described units, will be described with reference to FIG. 2 showing a flowchart of the processing.

  First, the imaging unit 100 captures an image based on a control signal from the control unit 101 (step S201). The captured image data is displayed on the display unit 106, output to the image holding unit 105, and further input to the face detection unit 102.

  Next, the face detection unit 102 performs processing for detecting a face area in the captured image using the input captured image (step S202). This face area detection process will be described in more detail.

  FIG. 7 is a diagram illustrating a series of processes for detecting a local feature in a captured image and specifying a face region. In the process shown in the figure, first, the primary feature which is the most primitive local feature is detected. As shown in the figure, the primary features include a vertical feature 701, a horizontal feature 702, a right-upward diagonal feature 703, and a right-down diagonal feature 704. Here, the “feature” represents a vertical edge segment when the vertical feature 701 is taken as an example.

  Since a technique for detecting a segment in each direction in a captured image is well known, this technique is used to detect a segment in each direction from a captured image and detect only a vertical feature from the captured image, and a lateral feature from the captured image. An image in which only a diagonal feature that has been detected from the photographed image is detected, and an image in which only the oblique feature from the photograph that has been descended from the right is detected. From this, the size (number of vertical and horizontal pixels) of the four images (primary feature images) is the same as that of the photographed image, so that the pixels correspond to the feature image and the photographed image on a one-to-one basis. In the feature image, the pixel value of the detected feature portion is different from the pixel value of the other portion. For example, the pixel value of the feature portion is 1, and the pixel values of the other portions are 0. To do. Therefore, if there is a pixel having a pixel value of 1 in the feature image, the corresponding pixel in the captured image can be regarded as a pixel constituting the primary feature.

  By generating the primary feature image group as described above, the primary feature in the captured image can be detected.

  Next, a secondary feature group obtained by combining any of the detected primary feature groups is detected from the captured image. As shown in the figure, the secondary feature group has a right empty V-character feature 710, a left empty V-character feature 711, a horizontal parallel line feature 712, and a vertical parallel line feature 713. The right empty V-shaped feature 710 is a combination of a right-up diagonal feature 703 and a right-down diagonal feature 704, which are primary features, and the left empty V-character feature 711 is a primary feature, a right-down diagonal feature 704 and a right up diagonal feature 703. The horizontal parallel line feature 712 is a feature combining a horizontal feature 702 that is a primary feature, and the vertical parallel line feature 713 is a feature combining a vertical feature 701 that is a primary feature.

  Similar to the generation of the primary feature image, an image in which only the right empty V-shaped feature 710 is detected from the captured image, an image in which only the left empty V-shaped feature 711 is detected from the captured image, and only the horizontal parallel line feature 712 is detected from the captured image. An image in which only the vertical parallel line feature 713 is detected from the captured image and the captured image is generated. From this, the size (number of vertical and horizontal pixels) of the four images (secondary feature images) is the same as that of the photographed image, so that the pixels correspond to the feature image and the photographed image on a one-to-one basis. In the feature image, the pixel value of the detected feature portion is different from the pixel value of the other portion. For example, the pixel value of the feature portion is 1, and the pixel values of the other portions are 0. To do. Therefore, if there is a pixel having a pixel value of 1 in the feature image, the corresponding pixel in the captured image can be regarded as a pixel constituting a secondary feature.

  By generating the secondary feature image group as described above, the secondary feature in the captured image can be detected.

  Next, a tertiary feature group combining any of the detected secondary feature groups is detected from the captured image. The tertiary feature group has features such as an eye feature 720 and a mouth feature 721 as shown in FIG. The eye feature 720 is a feature that combines a right empty V-shaped feature 710, a left empty V-shaped feature 711, a horizontal parallel line feature 712, and a vertical parallel line feature 713, which are secondary features, and a mouth feature 721 is a secondary feature. This is a combination of a right empty V-character feature 710, a left empty V-character feature 711, and a horizontal parallel line feature 712.

  Similar to the generation of the primary feature image, an image in which only the eye feature 720 is detected and an image in which only the mouth feature 721 is detected from the photographed image are generated. From this, the size (number of vertical and horizontal pixels) of the two images (tertiary feature images) is the same as that of the photographed image, so that the pixels correspond to each other one by one in the feature image and the photographed image. In the feature image, the pixel value of the detected feature portion is different from the pixel value of the other portion. For example, the pixel value of the feature portion is 1, and the pixel values of the other portions are 0. To do. Therefore, if there is a pixel having a pixel value of 1 in the feature image, the corresponding pixel in the captured image can be regarded as a pixel constituting the tertiary feature.

  By generating the tertiary feature image group as described above, the tertiary feature in the captured image can be detected.

  Next, a quaternary feature obtained by combining the detected tertiary feature groups is detected from the captured image. The quaternary feature is the facial feature itself in the figure. The facial feature is a feature that combines an eye feature 72 and a mouth feature 721 that are tertiary features.

  Similar to the generation of the primary feature image, an image (quaternary feature image) in which the facial feature is detected is generated. From this, the size (number of vertical and horizontal pixels) of the quaternary feature image is the same as that of the captured image, and therefore, the feature image and the captured image have a one-to-one correspondence with the pixels. In the feature image, the pixel value of the detected feature portion is different from the pixel value of the other portion. For example, the pixel value of the feature portion is 1, and the pixel values of the other portions are 0. To do. Therefore, if there is a pixel having a pixel value of 1 in the feature image, the corresponding pixel in the captured image can be regarded as a pixel constituting a quaternary feature. Therefore, by referring to this quaternary feature image, for example, the position of the face region can be obtained from the barycentric position of the pixel whose pixel value is 1.

  When this face area is a rectangle, in order to obtain information indicating how much the rectangle is tilted with respect to the captured image, the rotation is performed by calculating the inclination of the rectangle with respect to the captured image. The amount can be determined.

  As described above, the face area information can be obtained. The obtained face area information is output to the facial expression determination unit 104 as described above.

  Each feature image (primary feature image, secondary feature image, tertiary feature image, and quaternary feature image in this embodiment) is output to the intermediate detection result holding unit 103 as the intermediate detection result.

  Thus, by detecting the quaternary feature in the photographed image, the face area in the photographed image can be obtained. In addition, by performing the face area detection process described above on the entire captured image, each face area can be detected even if the captured image includes a plurality of face areas.

  Note that the face area detection processing can also be realized using a neural network that performs image recognition by parallel hierarchical processing, which is described in M. Matsugu, K. Mori, et.al, “ Convolutional Spiking Neural Network Model for Robust Face Detection ”, 2002, Internatinal Conference On Neural Information Processing (ICONIP02).

  The processing contents of the neural network will be described with reference to FIG. FIG. 8 is a diagram illustrating a configuration of a neural network for performing image recognition.

  This neural network hierarchically handles information related to recognition (detection) of objects or geometric features in a local region in input data, and its basic structure is a so-called Convolutional network structure (LeCun, Y. and Bengio, Y., 1995, “Convolutional Networks for Images Speech, and Time Series” in Handbook of Brain Theory and Neural Networks (M. Arbib, Ed.), MIT Press, pp. 255-258). In the final layer (uppermost layer), the presence / absence of the subject to be detected and the position information on the input data if it exists are obtained. If this neural network is applied to the present embodiment, from this final layer, the presence / absence of a face area in the photographed image and, if there is a face area, position information on the photographed image of this face area. can get.

  In the figure, a data input layer 801 is a layer for inputting image data. The first feature detection layer (1, 0) is a local low-order feature of the image pattern input from the data input layer 801 (geometric features such as a specific direction component and a specific spatial frequency component, as well as a color component feature). Multiple features at multiple scale levels or resolutions at the same location in a local region (or a local region centered around each point of a predetermined sampling point across the entire screen) Detect only the number of categories.

  The feature integration layer (2,0) has a predetermined receptive field structure (hereinafter, the receptive field means the coupling range with the output element of the immediately preceding layer, and the receptive field structure means the distribution of the coupled load). Then, the outputs of a plurality of neuron elements in the same receptive field from the feature detection layer (1, 0) are integrated (calculation such as sub-sampling by local averaging, maximum output detection, etc.). This integration process has a role of allowing positional deviation and deformation by spatially blurring the output from the feature detection layer (1, 0). Each receptive field of neurons in the feature integration layer has a common structure among neurons in the same layer.

  Each feature detection layer (1,1), (1,2),..., (1, M) and each feature integration layer (2,1), (2,2),. , M), like the above-described layers, the former ((1, 1),...) Detects a plurality of different features in each feature detection module, and the latter ((2, 1),. Integration of detection results regarding multiple features from the previous feature detection layer is performed. However, the former feature detection layer is coupled (wired) to receive the cell element output of the former feature integration layer belonging to the same channel. Sub-sampling, which is a process performed in the feature integration layer, is to average the output from a local region (local receptive field of the feature integration layer neuron) from the feature detection cell population of the same feature category. .

  In order to detect each feature shown in FIG. 7 using the neural network shown in FIG. 8, the receptive field structure used for detection of each feature detection layer is to detect the feature. Thus, each feature can be detected. In addition, when the receptive field structure used for face detection in the face detection layer of the final layer is prepared for each size and each rotation amount, and the result of face feature detection is obtained Depending on which receptive field structure is used for detection, face data such as the size and orientation of the face can be obtained.

  Returning to FIG. 2, next, the control unit 101 refers to the result of the face area detection process performed by the face detection unit 102 in step S202, and determines whether or not a face area exists in the captured image (step S203). ). As this determination method, for example, it is determined whether or not a quaternary feature image is obtained. If it is obtained, it is determined that a face region exists in the captured image. In addition, among the neurons in the (face) feature detection layer, it is determined whether there is a neuron with an output value equal to or higher than a reference value, and a face (region) exists at the position indicated by the neuron with the reference value or higher. You may do that. In that case, if there are no neurons above the reference value, it is assumed that there is no face.

  If the face area does not exist in the captured image as a result of the determination process in step S203, the face detection unit 102 notifies the control unit 101 to that effect, so the process returns to step S201. To take a new image.

  On the other hand, if a face area exists, the face detection unit 102 notifies the control unit 101 to that effect, so the process proceeds to step S204, and the control unit 101 performs facial expression determination on the captured image held in the image holding unit 105. And the feature image held in the intermediate detection result holding unit 103 is output to the facial expression discrimination unit 104. The facial expression discrimination unit 104 uses the input feature image and face area information to capture a captured image. Processing for determining facial expressions included in the inner face region is performed (step S204).

  Note that the image output from the image holding unit 105 to the facial expression determination unit 104 is the entire captured image in the present embodiment, but is not limited to this. For example, the control unit 101 uses the face area information to display the captured image. May be specified, and an image of only this face area may be output to the facial expression determination unit 104.

  Next, the facial expression determination process performed by the facial expression determination unit 104 will be described in more detail. As described above, the action unit (AU) used in the FACS (Facial Action Coding System), which is a general expression description method, is detected in order to determine the facial expression, and the facial expression is determined according to the detected AU type. Can be done. AU includes raising the outside of the eyebrows and pulling the lips sideways. Since all human facial expressions can be described by combinations of AUs, all facial expressions can be discriminated in principle if all AUs can be detected. However, there are 44 AUs, and it is not easy to detect all of them.

  Therefore, in this embodiment, as shown in FIG. 9, the eyebrow end points (B1 to B4), the eye end points (E1 to E4), and the mouth end points (M1 and M2) are used for facial expression discrimination, and these feature points. The facial expression is discriminated by determining the change in the relative position of the. Some AUs can be described by changes in these feature points, and basic facial expressions can be distinguished. The change of each feature point in each facial expression is held in the facial expression discrimination unit 104 as facial expression discrimination data, and is used for the facial expression discrimination process of the facial expression discrimination unit 104.

  FIG. 9 is a diagram showing each feature point.

  Each feature point for facial expression detection shown in FIG. 9 is an end portion of eyes, eyebrows, and the like, and the shape of this end portion is roughly a right empty V character and a left empty V character. This corresponds to the right empty V-character feature 710 and the left empty V-character feature 711 of the secondary feature shown in FIG.

  In addition, detection of feature points used for facial expression discrimination is obtained at an intermediate stage of face detection processing in the face detection unit 102. The intermediate processing result of the face detection process is held in the intermediate feature result holding unit 103.

  However, the right empty V-shaped feature 710 and the left empty V-shaped feature 711 exist at various positions such as the background in addition to the face. Therefore, the face area in the secondary feature image is specified using the face area information obtained by the face detection unit 102, and the end points of the right empty V-shaped feature 710 and the left empty V-shaped feature 711 in this area, that is, the eyebrows Detect endpoints, eye endpoints, and mouth endpoints.

  Therefore, as shown in FIG. 9, the search range (RE1, RE2) of the eyebrow / eye end points and the search range (RM) of the mouth end points are set in the face region. Then, referring to the pixel values within the set search range, the positions of the pixels at both ends in the horizontal direction in the figure are detected from among the pixel groups constituting the right empty V-shaped feature 710 and the left empty V-shaped feature 711. The detected position is used as the position of the feature point. The relative position of the search range (RE1, RE2, RM) with respect to the center position of the face area is set in advance.

  For example, in the pixel group constituting the right empty V-shaped feature 710 in the search range RE1, the positions of the pixels at the end in the horizontal direction in the figure are B1 and E1, and this is the position of one end of either the eyebrow or the eye. To do. Further, the position in the vertical direction of each of the positions B1 and E1 is referred to, and the position located on the upper side is the position of one end of the eyebrow. In the figure, B1 is at a position higher than E1, and therefore B1 is set to the position of one end of the eyebrow.

  In this way, the position of one end of the eyes and eyebrows can be obtained. Similarly, by performing the same processing for the left empty V-shaped feature 711 in the search range RE1, the positions of the eyebrows and the other end positions B2 and E2 of the eyes can be obtained.

  By the processing described above, the positions of the eyes, the eyebrows, and both ends of the mouth, that is, the positions of the feature points can be obtained. Since the feature image and the captured image have the same size and correspond to each pixel on a one-to-one basis, the position of each feature point in the feature image can be used as it is in the captured image.

  In the present embodiment, the secondary feature is used for the process of obtaining the position of each feature point. However, the present invention is not limited to this, and any one or a combination of the primary feature, the tertiary feature, or the like may be used.

  For example, in addition to the right empty V-character feature 710 and the left empty V-character feature 711, the eye feature 720 and the mouth feature 721 that are the tertiary features shown in FIG. A feature 703, a right-down diagonal feature 704 can also be used.

  Processing for obtaining feature points using primary features and tertiary features will be described with reference to FIG. FIG. 10 is a diagram for explaining processing for obtaining feature points using primary features and tertiary features in the face area shown in FIG. 9.

  As shown in FIG. 10, an eye search range (RE3, RE4) and a mouth search range (RM2) are set, and the eye feature 720 and the mouth feature 721 are obtained by referring to the pixel values in the set search range. A range in which the pixel group to be configured is arranged is obtained. Then, the eyebrows / eye endpoint search ranges (RE5, RE6) and the mouth endpoint search range (RM3) are set to cover this range.

  Then, in each search range (RE5, RE6, RM3), a continuous line segment consisting of a vertical feature 701, a horizontal feature 702, a right-up diagonal feature 703, and a right-down diagonal feature 704 is traced, and as a result, The positions of both ends can be obtained in the horizontal direction, and the ends of the eyes, eyebrows and mouth can be obtained. Since the primary feature is basically edge extraction, for each detection result, an end point can be detected by thinning a region above a certain threshold value and tracking the result.

  Next, facial expression determination processing performed using each obtained feature point will be described. In order to eliminate individual differences in facial expression discrimination, face detection processing is first performed on the face image when there is no expression, and the detection result of each local feature is obtained. Then, using these detection results, the relative position of each feature point shown in FIG. 9 or FIG. 10 is obtained, and the data is held in the facial expression determination unit 104 as a reference relative position. Then, the facial expression discrimination unit 104 refers to the reference relative position and the relative position of each of the obtained feature points, and performs a process of obtaining how much each feature point has changed from the reference, that is, “deviation”. . In addition, since the size of the face in the photographed image is generally different from the size of the face photographed in advance without expression, based on the relative position of each obtained feature point, for example, the distance between both eyes, Normalize the position of feature points.

  Then, a score depending on the change is obtained for each feature point, and the facial expression is discriminated based on the distribution of the score. For example, the expression of joy is characterized by (1) lowering of the buttocks, (2) lifting of the cheek muscles, and (3) lifting of the edge of the mouth. Change in “distance to”, “width of mouth width”, and “length of eye width”. And the score distribution obtained from these changes is a score distribution peculiar to a joyful expression.

  This unique score distribution is the same for other facial expressions. Therefore, the distribution shape is modeled parametrically by mixed Gaussian approximation, and similarity determination between the obtained score distribution and the score distribution provided for each facial expression is obtained by determining the magnitude of the distance in the parameter space. . Then, the facial expression indicated by the score distribution having a higher degree of similarity to the obtained score distribution (score distribution with a smaller distance) is used as the facial expression as the determination result.

  A method of performing threshold processing on the score total is also applicable. This threshold processing is more effective for accurately discriminating a non-expression scene similar to an expression scene (for example, a face pronounced “i” in conversation) and an expression scene. Note that either the discrimination of the score distribution shape or the threshold processing of the sum may be performed. Thus, by performing facial expression determination from the threshold distribution of the score distribution and the sum of the scores, the facial expression scene can be accurately recognized and the detection rate can be improved.

  Since the facial expression can be determined by the above processing, the facial expression determination unit 104 outputs a code corresponding to the determined facial expression (an individual code for each facial expression). This code may be a number, for example, and the expression method is not particularly limited.

  Next, the facial expression determination unit 104 determines whether or not the determined facial expression is a specific facial expression (for example, a smile) set in advance, and notifies the control unit 101 of the determination result (step S205).

  Here, if the facial expression determined by the processing up to step S204 is the same as the specific facial expression set in advance, for example, in the case of the present embodiment, the “code indicating the facial expression” output by the facial expression determination unit 104 And the code indicating the specific facial expression set in advance, the control unit 101 records the captured image held by the image holding unit 105 in the recording unit 107. When the recording unit 107 is a DVD-RAM or a compact flash (registered trademark), the control unit 101 controls the recording unit 107 to take a captured image on a storage medium such as a DVD-RAM or a compact flash (registered trademark). Is recorded (step S206). The image to be recorded may be a face area image, that is, a face image having a specific expression.

  On the other hand, when the facial expression determined by the processing up to step S204 is not the same as the specific facial expression set in advance, for example, in the case of the present embodiment, the “expression indicating facial expression” output by the facial expression determination unit 104 When the preset code indicating a specific facial expression does not match, the control unit 101 controls the imaging unit 100 to capture a new image.

  In addition, when the determined facial expression is a specific facial expression, for example, in step S206, the control unit 101 controls the imaging unit 100 to capture the next image, and the captured image is recorded in the recording unit. 107 may be held. Further, the control unit 101 may control the display unit 106 to display the captured image on the display unit 106.

  In general, the facial expression does not change abruptly but has a certain degree of continuity. Therefore, when the processing in step S202 and step S204 is completed in a relatively short time, it continues with an image showing a specific facial expression. The images often show similar expressions. Therefore, in order to make the face area detected in step S202 clearer, the control unit 101 sets shooting parameters of the shooting unit 100 (exposure parameters such as exposure correction, autofocus, color correction, etc.) and re-shoots. It is also possible to operate so as to perform, display and record.

  FIG. 11 is a diagram showing a basic configuration of the image processing apparatus according to the present embodiment.

  A CPU 1001 controls the entire apparatus using programs and data stored in the RAM 1002 and the ROM 1003, and executes a series of processes related to the facial expression determination. The CPU 1001 corresponds to the control unit 101 in FIG.

  A RAM 1002 has an area for temporarily storing programs and data loaded from the external storage device 1007 and the storage medium drive 1008, image data input from the imaging unit 100 via the I / F 1009, and the like. An area necessary for the CPU 1001 to execute various processes is also provided. In FIG. 1, the intermediate detection result holding unit 103 and the image holding unit 105 correspond to the RAM 1002.

  Reference numeral 1003 denotes a ROM which stores, for example, a boat program and setting data for the entire apparatus.

  Reference numerals 1004 and 1005 denote a keyboard and a mouse, respectively, which are used to input various instructions to the CPU 1001.

  A display device 1006 includes a CRT, a liquid crystal screen, and the like, and can display various types of information including images and characters. In FIG. 1, this corresponds to the display unit 106.

  Reference numeral 1007 denotes an external storage device that functions as a large-capacity information storage device such as a hard disk drive device. A program executed by the OS (operating system) or the CPU 1001 to execute a series of processes related to the facial expression determination. And so on. This program is read into the RAM 1002 according to an instruction from the CPU 1001 and executed by the CPU 1001. Note that this program includes programs corresponding to the face detection unit 102 and the expression determination unit 104 when the face detection unit 102 and the expression determination unit 104 shown in FIG.

  Reference numeral 1008 denotes a storage medium drive device 1008 that reads a program or data recorded on a storage medium such as a CD-ROM or DVD-ROM and outputs it to the RAM 1002 or the external storage device 1007. Note that a program executed by the CPU 1001 for executing the series of processes related to the facial expression determination may be recorded on the storage medium, and the storage medium drive device 1008 may read the data to the RAM 1002 according to an instruction from the CPU 1001. .

  Reference numeral 1009 denotes an I / F for connecting the image capturing unit 100 shown in FIG. 1 and the present apparatus, and image data captured by the image capturing unit 100 is output to the RAM 1002 via the I / F 1009.

  Reference numeral 1010 denotes a bus connecting the above-described units.

  Next, a case where shooting is performed when the subject has a specific facial expression by mounting the image processing apparatus according to the present embodiment on the imaging apparatus will be described with reference to FIG. FIG. 12 is a diagram illustrating a configuration of an example in which the image processing apparatus according to the present embodiment is used in an imaging apparatus.

  An imaging device 5101 in FIG. 12 includes an imaging optical system 5102 including a photographing lens and a zoom photographing drive control mechanism, a CCD or CMOS image sensor 5103, an imaging parameter measurement unit 5104, a video signal processing circuit 5105, a storage unit 5106, A control signal generation unit 5107 that generates control signals for controlling imaging operations, imaging conditions, and the like, a display display 5108 that also serves as a viewfinder such as an EVF, a strobe light emitting unit 5109, a recording medium 5110, and the like are provided. A processing device 5111 is provided as a facial expression detection device.

  The imaging apparatus 5101 uses, for example, an image processing apparatus 5111 to detect a human face image (detection of an existing position / size / rotation angle) and a facial expression from a captured video. When the position information and facial expression information of the person is input from the image processing apparatus 5111 to the control signal generation unit 5107, the control signal generation unit 5107 is based on the output from the imaging parameter measurement unit 5104. A control signal for optimally capturing an image is generated. Specifically, for example, a face image of a person is obtained in the front of a predetermined size or larger in the center of the shooting area, and the time of shooting can be taken as the shooting time point.

  By using the above-described image processing apparatus in the imaging apparatus in this way, it is possible to perform face detection, facial expression detection, and shooting with good timing based on the detection. In the above description, the imaging apparatus 5101 provided with the above-described processing apparatus as the image processing apparatus 5111 has been described. Naturally, the above-described algorithm is implemented as a program and is installed in the imaging apparatus 5101 as processing means that is operated by the CPU. It is also possible.

  Further, the image processing apparatus applicable to the imaging apparatus is not limited to the one according to the present embodiment, and the image processing apparatus according to the embodiment described below may be applied.

  As described above, the image processing apparatus according to the present embodiment uses local features such as a primary feature, a secondary feature, etc., so that not only can a facial region in a captured image be identified, It is possible to more easily perform facial expression determination processing without performing eye detection processing.

  In addition, even if the position and orientation of the face in the captured image vary, the above local features can be obtained, and as a result, facial expression determination processing can be performed, so the position of the face in the captured image Robust facial expressions can be determined for the direction and direction.

  Further, according to the present embodiment, only a specific facial expression can be photographed in the process of photographing many times.

  In this embodiment, the image for detecting the face area is a captured image. However, the image is not limited to this. The image may be saved in advance or downloaded. There may be.

[Second Embodiment]
In the present embodiment, the face detection area detection process (step S202) and the facial expression determination process (step S204) in the first embodiment are performed in parallel. Thereby, the whole process can be performed at higher speed.

  FIG. 3 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment. In the configuration according to the present embodiment, the configuration of the intermediate detection result holding unit 303 and the configuration of the image holding unit 305 are substantially different from those according to the first embodiment.

  The intermediate detection result holding unit 303 further includes an intermediate detection result holding unit A313 and an intermediate detection result holding unit B314. Similarly, the image holding unit 305 includes an image holding unit A315 and an image holding unit B316.

  Next, the operation of the configuration shown in FIG. 3 will be described using the timing chart of FIG.

  In the timing chart of FIG. 4, “A” indicates that the operation is in the A mode, and “B” indicates that the operation is in the B mode. The A mode of “image shooting” indicates that the captured image is held in the image holding unit A315 when the image is held in the image holding unit 305, and the B mode is held in the image holding unit B316. Hereinafter, the A mode and the B mode of image capturing are alternately switched, and the image capturing unit 300 captures images accordingly, so that the image capturing unit 300 continuously captures images. Note that the control unit 101 gives the shooting timing.

  Further, the “face detection” A mode refers to holding in the intermediate detection result holding unit A313 when holding the intermediate processing result in the intermediate detection result holding unit 303 in the face area detection processing of the face detection unit 302. , B mode indicates holding in the intermediate detection result holding unit B314.

  Further, the A mode of “expression discrimination” refers to the image held in the image holding unit A 315, the intermediate processing result held in the intermediate detection result holding unit A 313, and the face detection unit 302 in the facial expression discrimination process of the facial expression discrimination unit 304. The B mode is an image held in the image holding unit B316, an intermediate feature detection result held in the intermediate detection result holding unit B314, and a face detection unit. This indicates that facial expression is discriminated using face area information 302.

  Next, the operation of the image processing apparatus according to this embodiment will be described.

  First, the captured image is held in the image holding unit A315 of the image holding unit 305 by shooting an image in the A mode of image shooting. In addition, an image is displayed on the display unit 306, and further, an image is input to the face detection unit 302. Next, the face detection unit 302 performs processing for generating face area information by performing processing similar to that of the first embodiment on the input image. If a face is detected in the image, face area information data is input to the facial expression determination unit 304. The intermediate feature detection result obtained during the face detection process is held in the intermediate detection result holding unit A313 of the intermediate detection result holding unit 303.

  Next, B-mode image capturing, B-mode face detection processing, and A-mode facial expression determination processing are performed in parallel. In B-mode image capturing, the captured image is stored in the image storage unit B316 of the image storage unit 305. In addition, an image is displayed on the display unit 306, and further, an image is input to the face detection unit 302. Then, the face detection unit 302 performs processing for generating face area information by performing the same processing as in the first embodiment on the input image, and the intermediate processing result is stored in the intermediate processing result holding unit B314. Hold.

  In addition, the A-mode facial expression discrimination process is performed in parallel with the B-mode image capture and the B-mode face area detection process. In the A mode facial expression determination process, the facial expression determination unit 304 detects the facial feature information from the face detection unit 302 and the intermediate feature detection held in the intermediate detection result storage unit A313 for the image input from the image holding unit A315. Using the result, the facial expression discrimination unit 304 discriminates the facial expression. If the facial expression discriminated by the facial expression discriminating unit 304 is a desired facial expression, the image stored in the image holding unit A315 is recorded, and the process ends.

  If the facial expression discriminated by the facial expression discriminating unit 304 is different from the desired facial expression, A-mode image capturing, A-mode face area detection processing, and B-mode facial expression discrimination processing are performed in parallel. In the A mode image shooting, the captured image is held in the image holding unit A315 of the image holding unit 305. In addition, an image is displayed on the display unit 306, and the image is further input to the face detection processing unit 302. Subsequently, the face detection unit 302 performs a process of detecting a face area on the input image. Further, in the B-mode facial expression discrimination process performed in parallel, the facial expression discrimination unit 304 applies the face area information from the face detection unit 302 and the intermediate detection result holding unit B 314 to the image input from the image holding unit B 316. The facial expression determination unit 304 detects the facial expression using the intermediate detection result held in the above.

  Thereafter, the same processing is repeated until the facial expression determined by the facial expression determination unit 304 is determined to be a specific facial expression. When the facial expression is determined to be a desired expression, the image in the image holding unit A315 is recorded if the facial expression determination process is in the A mode, and the image in the image holding unit B316 is recorded in the B mode.

  Note that the mode switching in each process is performed by the control unit 101, and the timing is switched when the control unit 101 detects the end of the face detection processing performed by the face detection unit 102.

  As described above, the image holding unit 305 includes the image holding unit A315 and the image holding unit B316, and the intermediate detection result holding unit 303 includes the intermediate detection result holding unit A313 and the intermediate detection result holding unit B314. In addition, the face area detection process and the facial expression determination process can be performed in parallel, and as a result, the shooting rate of the image for determining the facial expression can be increased.

[Third Embodiment]
The image processing apparatus according to this embodiment performs the face area detection processing performed by the face detection unit 102 in the first and second embodiments and the facial expression determination processing performed by the facial expression determination unit 104 in parallel, thereby The purpose is to improve performance.

  In the second embodiment, using the fact that the image shooting and face area detection processing takes more time than the expression determination processing, the expression determination processing, the next image shooting, and the face in the next image are processed. Region detection processing was performed in parallel. On the other hand, in this embodiment, in the face detection process, the process of detecting the quaternary feature quantity shown in FIG. 7 of the first embodiment is compared with the detection of the tertiary feature quantity from the primary feature quantity. Taking advantage of the processing time, the detection result of the previous image is used for the face area information, and the detection result of the current image is used for the detection result of the feature points used for detection of facial expressions such as eyes and mouth. Thereby, parallel processing of face area detection processing and facial expression discrimination processing is realized.

  FIG. 5 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment.

  The imaging unit 500 captures time-series images or moving images and outputs image data of each frame to the face detection unit 502, the image holding unit 505, the display unit 506, and the recording unit 507. In the configuration according to the present embodiment, the face detection unit 502 and the facial expression determination unit 504 are substantially different from those according to the first embodiment.

  The face detection unit 502 performs the same process as the face area detection process according to the first embodiment, but outputs an end signal to the facial expression determination unit 504 when the process ends.

  The facial expression determination unit 504 further includes a configuration including a previous image detection result holding unit 514.

  Next, processing performed by each unit illustrated in FIG. 5 will be described with reference to a timing chart illustrated in FIG.

  When the image of the first frame is taken by the imaging unit 500, the data of this image is input to the face detection unit 502. The face detection unit 502 generates face area information by performing the same processing as that of the first embodiment on the input image, and outputs it to the facial expression determination unit 504. The face area information input to the facial expression determination unit 504 is held in the previous image detection result holding unit 514. Further, the intermediate feature detection result obtained in the middle is input to the intermediate detection result holding unit 503 and held.

  Subsequently, when the image of the next frame is taken by the imaging unit 500, the data of this image is input to the image holding unit 505. Further, the captured image is displayed on the display unit 506, and the image is further input to the face detection unit 502. Then, the face detection unit 502 generates face area information by performing the same processing as in the first embodiment. When the face area detection process is completed, the face detection unit 502 inputs the detection result of the intermediate feature to the intermediate detection result holding unit 503, and a signal indicating that a series of processes to be performed by the facial expression determination unit 504 is completed. Is output.

  If the facial expression of the discrimination result of the facial expression discrimination unit 504 is not a desired facial expression, the face area information obtained by the face detection unit 502 is held in the previous image detection result holding unit 514 of the facial expression discrimination unit 504.

  When the facial expression determination unit 504 receives the end signal from the face detection unit 502, the facial expression information 601 for the previous image (one or more previous frames) held in the previous image detection result holding unit 514 Using the current image (current frame image) held in the image holding unit 505 and the intermediate feature detection result 602 of the current image held in the intermediate detection result holding unit 503, expression discrimination processing for the current image I do.

  That is, for an area in the original image that corresponds in position to the area specified by the face area information in the image of one or more previous frames, facial expression discrimination processing is performed using the intermediate detection result obtained from this area. Do.

  If the difference between the time when the previous image was taken and the time when the current image was taken is short, the position of the face area in each image does not change significantly. Therefore, as described above, the face area information obtained from the previous image is used, and the search areas shown in FIGS. 9 and 10 are set wider so that the positions of the face areas of the previous image and the current image are set. The influence of such a shift can be suppressed, and facial expression discrimination processing can be performed.

  If the facial expression discriminated by the facial expression discriminating unit 504 is a desired facial expression, the image stored in the image holding unit 505 is recorded and the process ends. If the facial expression discriminated by the facial expression discriminating unit 504 is different from the desired facial expression, the next image is shot, the face detecting unit 502 performs face detection processing, and the facial expression discriminating unit 504 captures the captured image and the previous image detection result. Expression determination processing is performed using the face detection result for the previous image held in the holding unit 514 and the intermediate processing result held in the intermediate detection result holding unit 503.

  Thereafter, the same processing is repeated until the facial expression determined by the facial expression determination unit 504 becomes a desired facial expression. When it is determined that the expression is desired, the image stored in the image holding unit 505 is recorded and the process ends.

  In this way, facial expression discrimination processing is performed using the face area information for the previous image held in the previous image detection result holding unit 514 and the intermediate feature detection processing result held in the intermediate detection result holding unit 503. Thus, the face area detection process and the facial expression determination process can be performed in parallel, and as a result, the shooting rate of the image for determining the facial expression can be increased.

[Fourth Embodiment]
Although the technique for discriminating the facial expression has been described in the above embodiment, this embodiment will explain a technique for discriminating who the face belongs to, that is, a technique for discriminating an individual with respect to the face.

  FIG. 13 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment. The image processing apparatus according to this embodiment includes an imaging unit 1300, a control unit 1301, a face detection unit 1302, an intermediate detection result holding unit 1303, an individual determination unit 1304, an image holding unit 1305, a display unit 1306, and a recording unit 1307. Hereinafter, each part will be described.

  The imaging unit 1300 captures an image based on a control signal from the control unit 1301 and outputs the captured image (captured image) to the face detection unit 1302, the image holding unit 1305, the display unit 1306, or the recording unit 1307. .

  The control unit 1301 performs processing for controlling the entire image processing apparatus according to the present embodiment, and includes an imaging unit 1300, a face detection unit 1302, an intermediate detection result holding unit 1303, an individual determination unit 1304, and an image holding unit. 1305, connected to a display unit 1306 and a recording unit 1307, and controls each unit so that each unit operates at an appropriate timing.

  The face detection unit 1302 performs processing for detecting a face region (a region of a face image included in the photographed image) in the photographed image from the imaging unit 1301. This process is performed to determine the presence or absence of a face area in the photographed image. If a face area exists, the number of face areas, the coordinate position of the face area in the photographed image, the size of the face area, In other words, the processing is to obtain the amount of rotation (for example, when the face area is a rectangle, the amount of rotation indicating how much the rectangle is inclined in which direction in the captured image). Note that these pieces of information (the number of face areas in the captured image, the coordinate position of the face area in the captured image, the size of the face area, and the amount of rotation in the captured image of the face area) are collectively referred to as “face area information”. Call it. Therefore, the face area in the captured image can be specified by obtaining the face area information.

  These detection results are output to the facial expression discrimination unit 1304. Further, an intermediate detection result described later obtained during the detection process is output to the intermediate detection result holding unit 1303.

  The intermediate detection result holding unit 1303 holds the intermediate feature detection result output from the face detection unit 1302.

  The individual determination unit 1304 receives the face area information data output from the face detection unit 1302 and the intermediate feature detection result data output from the intermediate detection result holding unit 1303. Then, based on these data, a process for determining who this face belongs to is performed. Details of this discrimination processing will be described later.

  The image holding unit 1305 temporarily holds the captured image output from the imaging unit 1300, and displays all or part of the stored captured image based on a control signal from the control unit 1301, a display unit 1306, and a recording unit. To 1307.

  The display unit 1306 is configured by, for example, a CRT, a liquid crystal screen, and the like, and displays all or a part of the captured image output from the image holding unit 1305 or the captured image captured by the imaging unit 1300.

  The recording unit 1307 is configured by a device that records information on a storage medium such as a hard disk drive, a DVD-RAM, or a compact flash (registered trademark), and an image stored in the image holding unit 1305 or an image captured by the imaging unit 1300. Record the captured image.

  Next, main processing for determining who the face in the photographed image is, which is executed by the operation of each unit described above, will be described with reference to FIG. 14 showing a flowchart of the processing.

  First, the imaging unit 1300 captures an image based on a control signal from the control unit 1301 (step S1401). The captured image data is displayed on the display unit 1306, output to the image holding unit 1305, and further input to the face detection unit 1302.

  Next, the face detection unit 1302 performs processing for detecting a face region in the captured image using the input captured image (step S1402). Since this face area detection process is performed in the same manner as in the first embodiment, a description thereof is omitted, but as an intermediate process result in the face detection process, it is useful for personal identification such as eyes / mouth, eyes, and end points of the mouth. That a special feature is detected is a major feature of the face detection processing method according to the present embodiment.

  Next, the control unit 1301 refers to the result of the face area detection process performed by the face detection unit 1302 in step S1402, and determines whether or not a face area exists in the captured image (step S1403). As the determination method, for example, among the neurons in the (face) feature detection layer, it is determined whether there is a neuron with an output value equal to or higher than a reference value, and the face (region) is located at the position indicated by the neuron with the reference value or higher. Suppose that exists. In addition, if there are no neurons above the reference value, it is assumed that no face exists.

  As a result of the determination processing in step S1403, when the face area does not exist in the captured image, the face detection unit 1302 notifies the control unit 1301 to that effect, so the process returns to step S1401, and the control unit 1301 performs the imaging unit 1300. To take a new image.

  On the other hand, if a face area exists, the face detection unit 1302 notifies the control unit 1301 of that fact, so the process proceeds to step S1404, and the control unit 1301 holds intermediate detection result information obtained by the face detection unit 1302 as an intermediate detection result. The face area information from the face detection unit 1302 is input to the individual determination unit 1304.

  Here, as described above, the number of faces can be obtained by the number of neurons equal to or greater than a reference value. Note that face detection by a neural network is robust to face size fluctuations and rotation fluctuations, and therefore one neuron exceeding the reference value is not necessarily one for one face in an image. In general, there will be multiple. Therefore, the number of faces in the image is obtained by collecting the neurons exceeding the reference value based on the distance between the neurons exceeding the reference value. In addition, the average position and the gravity center position of the plurality of neurons collected in this way are set as the face position.

  The amount of rotation and the size of the face are obtained as follows. As described above, an eye or mouth detection result is obtained as an intermediate processing result when detecting a facial feature. That is, as shown in FIG. 10 shown in the first embodiment, the eye search range (RE3, RE4) and the mouth search range (RM2) are set using the face detection result, and the eye feature detection result and the mouth feature are set. Eye features and mouth features can be detected within the range of the detection result. Specifically, in these range of neurons in the eye detection layer and in the mouth detection layer, the average position or centroid position of a plurality of neurons exceeding the reference value is set as the eye (left and right eyes) and mouth position. . Then, the face size and the amount of rotation are obtained from the positional relationship of these three points. When obtaining the face size and rotation amount, it is also possible to obtain only the position of both eyes from the eye feature detection result, that is, it is possible to obtain the face size and rotation amount from only the position of both eyes without using the mouth feature. .

  The individual determination unit 1304 then uses the face area information and the intermediate detection result information held in the intermediate detection result holding unit 1303 to determine who the face included in the face area in the captured image belongs to. Processing is performed (step S1404).

  Here, a determination process (personal determination process) performed by the individual determination unit 1304 will be described. In this description, first, a feature vector used in this determination process will be described, and then a classifier that performs identification using the feature vector will be described.

  As described in the background art, the individual discrimination process is generally performed independently of the face detection process for detecting the position and size of the face in the image. That is, normally, the process for obtaining the feature vector used for the individual discrimination process and the face detection process are independent of each other. On the other hand, in the present embodiment, the feature vector used for the individual discrimination process is obtained from the intermediate processing result of the face detection process, so that the number of feature quantities to be obtained in the process of performing the individual discrimination process can be smaller than in the past. Therefore, the whole process becomes simpler.

  FIG. 15 is a diagram illustrating feature vectors used in the individual discrimination process. FIG. 15A is a diagram showing a feature vector 1301 used for the individual discrimination process, FIG. 15B is a diagram showing a result of detecting a right empty V-shaped feature of a secondary feature, and FIG. 15C is a left empty V FIG. 15D is a diagram showing a character feature detection result, and FIG. 15D is a diagram showing a captured image including a face area.

  Here, the dotted lines in FIGS. 15B and 15C indicate the edges of the eyes in the face. This edge is not an actual feature vector, but is shown for easy understanding of the relationship between the V-shaped feature detection result and the eyes. In FIG. 15B, reference numerals 1502a to 1502d denote the firing distribution areas of the neurons in each feature in the right empty V-shaped feature detection result of the secondary feature, with black indicating a large value and white indicating a small value. ing. Similarly, in FIG. 15C, reference numerals 1503a to 1503d indicate the firing distribution areas of neurons in each feature in the left empty V-shaped feature detection result of the secondary feature, with black indicating a large value and white indicating a small value. Show.

  In general, if the average shape feature of the detection target, the output value of the neuron becomes a large value, and if there is a change such as rotation or movement, the output value becomes a small value. The distribution of the output values of the neurons shown in b) and (c) is weaker from the coordinates where the detection target exists toward the periphery.

  As schematically shown in FIG. 15, the feature vector 1501 used for the individual discrimination process is one of the intermediate detection results held in the intermediate detection result holding unit 1303, and the right empty V character of the secondary feature. It is created from the feature detection result and the left empty V-shaped feature detection result. In addition, this feature vector uses an area 1504 including both eyes instead of the entire face area 1505 shown in FIG. More specifically, in the region including both eyes, the output values of the right empty V-shaped feature detection layer neurons and the left empty V-shaped feature detection layer neurons are considered as arrays, and the output values of the same coordinates , And a feature vector is created by selecting a large value.

  In the Eigenface method described in the background art, the entire face area is decomposed with a basis called an eigenface, and the coefficient is used as a feature vector used for individual discrimination. That is, in the Eigenface method, individual discrimination is performed using the characteristics of the entire face area. However, it is also possible to perform individual discrimination without using the entire face area as long as the characteristics show different tendencies among individuals. In the area including both eyes shown in FIG. 15, the right empty V-shaped feature detection result and the left empty V-shaped feature detection result include information such as the size of each eye, the distance between both eyes, and the eyebrow-eye distance. It is included, and it is possible to identify the individual from this information.

  In addition, the Eigenface method has a drawback that it is vulnerable to fluctuations in lighting conditions. However, the right empty V-shaped feature detection result and the left empty V-shaped feature detection result shown in FIG. 15 are robust to lighting conditions and size / rotation fluctuations. It is obtained by using the receptive field learned to detect the face, and is not easily affected by illumination conditions, size / rotation fluctuations, and is suitable for creating a feature vector for individual discrimination.

  Further, as described above, it is a very simple process to generate a feature vector for performing individual discrimination from the right empty V-character feature detection result and the left empty V-character feature detection result. Thus, it is very useful to generate a feature vector for individual discrimination using an intermediate processing result obtained in the middle of face detection processing.

  In the present embodiment, the classifier for performing individual discrimination using the obtained feature vector is not particularly limited, but there is a nearest neighbor classifier as an example. The nearest neighbor discriminator is a method of storing a training vector indicating each individual as a prototype and identifying an object by a class to which the prototype closest to the input feature vector belongs. In other words, the feature vector of each individual is obtained in advance by the method described above, retained, and the distance between the feature vector obtained from the input image and the retained feature vector is obtained, and the feature with the closest distance is obtained. The person showing the vector is used as the identification result.

  As another classifier, Support Vector Machine (hereinafter referred to as SVM) proposed by Vapnik et al. May be used. This SVM learns linear threshold element parameters from training data on the basis of margin maximization.

  In addition, it is a classifier with excellent discrimination performance by combining non-linear transformations called kernel tricks (Vapnik, “Statistical Learning Theory”, John Wiley & Sons (1998)). That is, a parameter for discrimination is obtained from training data indicating each individual, and the individual is discriminated from the parameter and the feature vector obtained from the input image. However, since SVM basically constitutes a discriminator that identifies two classes, when discriminating a plurality of persons, the discriminating is performed by combining a plurality of SVMs.

  As described above, the face detection processing performed in step S1402 uses a neural network that performs image recognition by parallel hierarchical processing. The receptive field used when detecting each feature is acquired by learning using a large amount of face images and non-face images. In other words, in a neural network that performs face detection processing, information that is common to a large number of face images but not common to non-face images is extracted from the input image, and that information is used to distinguish between faces and non-faces. You can think of it.

  On the other hand, a discriminator that performs individual discrimination is designed to discriminate a difference between feature vectors created for each individual from a face image. In other words, for each individual, multiple facial images with slightly different facial expressions and orientations are prepared, and if these facial images are used as training data, clusters are formed for each individual, and each cluster is separated using SVM. The surface to be obtained can be obtained with high accuracy.

  In addition, if it is a nearest neighbor classifier, there is a rationale that if a sufficient number of prototypes are given, an error probability less than twice the Bayes error probability can be achieved, and it is possible to identify differences between individuals It is.

  FIG. 16 is a table showing data used in learning in each of the three classifiers as a table. In other words, the table in FIG. 8 is used to perform training for making the face detection classifier detect human faces (including Mr. A and Mr. B), and to identify Mr. A to the Mr. A classifier. The data used when performing training and the data used when training for identifying Mr. B in the Mr. B classifier are shown, and during training for face detection using the face detection classifier Then, feature vectors obtained from facial images of all persons (A, B, and others) used as samples are used as correct answer data, and background images (non-face images) that are not facial images are used as incorrect answer data. Use.

  On the other hand, at the time of training for identification of Mr. A using the Mr. A classifier, the feature vector obtained from the face image of Mr. A is used as correct answer data, and a person other than Mr. A ("Mr. B", " The feature vector obtained from the face image of “others”) is used as incorrect answer data. The background image is not used during training.

  Similarly, at the time of training for identification of Mr. B using Mr. B's classifier, the feature vector obtained from Mr. B's face image is used as correct data, and a person other than Mr. B ("Mr. A", A feature vector obtained from the face image of “others”) is used as incorrect answer data. The background image is not used during training.

  Accordingly, the secondary feature detection result used when detecting the eye that is the tertiary feature and the secondary feature detection result used for the personal discrimination described above are partly in common, but as described above, A classifier for detecting features (neural network) and a classifier for individual discrimination are different in the type of classifier (that is, the neural network shown in the first embodiment, the SVM, the nearest neighbor classifier, and the like). In addition to the difference in the data set used for training, the information used for identification, which is extracted from the same detection results, will differ as a result. The eyes can be detected and the latter can identify individuals.

  When creating a feature vector, if the face size and orientation obtained by the face detection unit 1302 are not within a predetermined range, the intermediate processing result held in the intermediate detection result holding unit 1303 Thus, rotation correction and size normalization can be performed. Since the discriminator for individual discrimination is designed to discriminate fine differences between individuals, the accuracy tends to be improved by unifying the size and rotation. The rotation correction and the size normalization can be performed when the intermediate processing result held in the intermediate detection result holding unit 1303 is read from the intermediate detection result holding unit 1303 so as to be input to the individual determination unit 1304.

  Since the individual identification of the face has been completed by the above processing, the individual identification unit 1304 matches the code corresponding to the determined individual (individual code for each individual) with the code corresponding to a certain individual set in advance. It is determined whether or not (step S1405). This code may be a number, for example, and the expression method is not particularly limited. This determination result is notified to the control unit 1301.

  Here, if the individual determined by the processing up to step S1404 is the same as the specific individual set in advance, for example, in the case of the present embodiment, the “code indicating the individual” output by the individual determination unit 1304 If the predetermined code indicating a specific individual matches, the control unit 1301 records the captured image held by the image holding unit 1305 in the recording unit 1307. Further, when the recording unit 1307 is a DVD-RAM or a compact flash (registered trademark), the control unit 1301 controls the recording unit 1307 to take a captured image on a storage medium such as a DVD-RAM or a compact flash (registered trademark). Is recorded (step S1406). The image to be recorded may be a face area image.

  On the other hand, if the individual determined by the processing up to step S1404 is not the same as the specific individual set in advance, for example, in the case of the present embodiment, the “code indicating the individual” output by the individual determination unit 1304 When the preset code indicating a specific individual does not match, the control unit 1301 controls the imaging unit 1300 to capture a new image.

  In addition, when the determined individual has a specific facial expression, for example, in step S1406, the control unit 1301 controls the imaging unit 1300 to capture the next image, and the captured image is recorded in the recording unit. You may make it hold | maintain to 1307. FIG. Further, the control unit 1301 may control the display unit 1306 to display the captured image on the display unit 1306.

  In addition, the control unit 1301 sets shooting parameters (exposure parameters such as exposure correction, autofocus, color correction, and the like of the imaging system) so that the face area detected in step S202 is clearly imaged. It is also possible to operate to perform re-photographing, display and recording.

  As described above, when a face in an image is detected based on an algorithm that detects a final detection target from hierarchically detected local features, exposure correction, autofocus, and In addition to being able to perform processing such as color correction, it is possible to identify individuals using the eye candidate detection result and the mouth candidate detection result, which are intermediate feature detection results obtained during the face detection process. It becomes possible without performing a new detection process for detecting eyes and mouth, and there is an effect that an individual can be detected and photographed while suppressing an increase in processing cost. In addition, personal recognition that is resistant to fluctuations in the position and size of the face is possible.

  Further, the image processing apparatus according to the present embodiment may be a computer having the configuration shown in FIG. Further, the present invention may be applied to the image processing apparatus 5111 in the imaging apparatus shown in FIG. 12, and in this case, it is possible to perform imaging according to the individual determination result.

[Fifth Embodiment]
The image processing apparatus according to the present embodiment is the face area detection process described in the above embodiment, the facial expression determination process described in the first to third embodiments, and the fourth embodiment described in the fourth embodiment. Perform personal identification processing.

  FIG. 17 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment. Basically, the configuration of the image processing apparatus according to the first embodiment is added with the configuration of the image processing apparatus according to the fourth embodiment and an integration unit 1708. Each unit excluding the integration unit 1708 performs the same operation as the part having the same name in the above embodiment. That is, an image from the imaging unit 1700 is output to the face detection unit 1702, the image holding unit 1705, the recording unit 1707, and the display unit 1706. The face detection unit 1702 performs face area detection processing similar to that in the above embodiment, and outputs the detection processing result to the facial expression detection unit 1704 and the individual determination unit 1714 as in the above embodiment. Further, the intermediate detection result obtained during the process is output to the intermediate detection result holding unit 1703. The facial expression detection unit 1704 performs the same processing as the facial expression detection unit 104 in the first embodiment. The individual determination unit 1714 performs the same processing as the individual determination unit 1304 in the fourth embodiment.

  The integration unit 1708 receives data of processing results of the face detection unit 1702, the expression detection unit 1704, and the individual determination unit 1714, and using this, the face detected by the face detection unit 1702 is a face of a specific individual. If it is a face of a specific individual, and if it is a specific individual face, a determination process is performed to determine whether or not this specific face is a specific expression. That is, it is determined whether a specific individual has a specific facial expression.

  Next, FIG. 18 which shows the flowchart of the process about the main process for discriminating who is the face in the photographed image and what kind of expression is executed by the operation of each part described above. It explains using.

  The processing in each step from step S1801 to step S1803 is the same as the processing in each step from step S1401 to step S1403 in FIG. That is, in the processing from step S1801 to step S1803, the control unit 1701 and the face detection unit 1702 determine whether or not a face area exists in the image from the imaging unit 1700.

  If it exists, the process proceeds to step S1804, and the facial expression detection unit 1704 determines the facial expression in the detected face area by performing the same process as the process in step S204 of FIG.

  In step S1805, the individual determination unit 1714 performs individual determination of the face in the detected face area by performing the same processing as in step S1404 in FIG.

  Note that the processes in steps S1804 and S1805 are performed for each face detected in step S1802.

  In step S 1806, the integration unit 1708 outputs “a code corresponding to the determined facial expression” output from the facial expression detection unit 1704 and “a code corresponding to the determined individual” output from the individual determination unit 1714. , Manage each face.

  FIG. 19 is a diagram showing a configuration example of the managed data. As described above, both the facial expression detection unit 1704 and the individual determination unit 1714 perform facial expression determination and individual determination for each face detected by the face detection unit 1702. Therefore, the integration unit 1708 associates a unique ID for each face (numerals 1, 2, and so on in the figure) with “a code corresponding to the determined facial expression” and “a code corresponding to the determined individual”. Manage. For example, the code “smile” as the “code according to the determined facial expression” and the code “A” as the “code according to the determined individual” are for the face whose ID is 1, so these codes Are associated with one ID and managed. The same applies to the case where the ID is 2. In this way, the integration unit 1708 generates and holds table data (for example, having the configuration shown in FIG. 19) for managing each code.

  After that, the integration unit 1708 determines in step S1806 by referring to this table data whether or not a specific individual has a specific facial expression. For example, if it is determined whether or not Mr. A is laughing using the table data in FIG. 19, since Mr. A is smiling in the table data in FIG. 19, it is determined that Mr. A is laughing.

  As a result of the determination, if a specific individual has a specific facial expression, the integration unit 1708 notifies the control unit 1701 of the fact, so the process proceeds to step S1807, and step S1406 in FIG. Do the same process.

  In the present embodiment, the face detection process and the facial expression determination process are continuously performed, but the method described in the second and third embodiments may be used. In that case, the entire processing time can be shortened.

  As described above, according to the present embodiment, by detecting a face from an image, specifying an individual, and specifying the facial expression, taking a photograph of a desired facial expression of a desired individual from among a large number of people, For example, it becomes possible to take a picture of a smiling moment of a child from a plurality of children.

  That is, if the image processing apparatus according to the present embodiment is applied to the image processing apparatus of the imaging apparatus described in the first embodiment, both the individual determination process and the facial expression determination process can be performed. It is possible to perform shooting when an individual has a specific expression. Furthermore, it can be used as an interface between humans and machines by recognizing specific individuals and facial expressions.

[Sixth Embodiment]
In this embodiment, the facial expression determination process and the individual determination process described in the fifth embodiment are performed sequentially. Thereby, a specific facial expression of a specific individual can be determined with higher accuracy.

  FIG. 20 is a diagram illustrating a functional configuration of the image processing apparatus according to the present embodiment. The configuration shown in the figure is substantially the same as the configuration of the image processing apparatus according to the fifth embodiment shown in FIG. 18, but the individual discrimination unit 2014 and the facial expression discrimination unit 2004 are connected. The difference is that a facial expression discrimination data holding unit 2008 is used instead of the integration unit 1708.

  FIG. 21 is a flowchart of main processing performed by the image processing apparatus according to the present embodiment. Hereinafter, processing performed by the image processing apparatus according to the present embodiment will be described with reference to FIG.

  The processing in each step from step S2101 to step S2103 is the same as the processing in each step from step S1801 to step S1803 in FIG.

  In step S2104, the individual determination unit 2014 performs the same process as in step S1804, thereby performing the individual determination process. Note that the processing in step S2104 is processing performed for each face detected in step S1802. In step S2105, the individual determination unit 2014 determines whether the face determined in step S2104 is a specific face. For example, as described in the fifth embodiment, this is done by referring to management information (in this case, a table associating each face-specific ID with a code indicating an individual).

  If the code indicating a specific face matches the code indicating the determined face, that is, if the face determined in step S2104 is a specific face, the individual determination unit 2014 detects that fact. The facial expression determination processing 2004 performs facial expression determination processing in the same manner as in the first embodiment. In this embodiment, the facial expression detection section 2004 performs facial expression determination processing. Is used, the “expression discrimination data corresponding to each individual” held in the expression discrimination data holding unit 2008 is used.

  FIG. 22 is a diagram showing a configuration example of the facial expression discrimination data. As shown in the figure, parameters for facial expression discrimination are prepared in advance for each individual. Here, the parameters are “the distance between the end of the eye and the end of the mouth”, “the length of the width of the mouth”, and “the length of the width of the eye” described in the first embodiment, For example, “shadow” and “shadow under eyes”. Basically, as described in the first embodiment, individual-independent facial expression recognition can be performed based on differences from reference data created from expressionless image data, but individual-dependent changes are detected. By doing so, more accurate facial expression discrimination can be performed.

  For example, if a particular person has a smile with a large mouth, a shadow on his cheeks, or a shadow under his eyes, the distinctive change described above is used in facial expression discrimination for that person. By using it, facial expression discrimination with higher accuracy becomes possible.

  Accordingly, the facial expression detection unit 2004 receives a code indicating the face determined by the individual determination unit 2004, and reads parameters for facial expression determination according to the code from the facial expression determination data holding unit 2008. For example, when the facial expression discrimination data has a configuration as shown in FIG. 22, the individual discrimination unit 2004 determines that a face in the image belongs to Mr. A, and outputs a code indicating Mr. A to the facial expression detection unit 2004. For example, the facial expression detection unit 2004 obtains parameters corresponding to Mr. A from the facial expression discrimination data holding unit 2008 (parameters indicating the variation rate of the distance between the mouth and mouth> 1.1, the cheek region edge density 3.0). Reading and using this, facial expression discrimination processing is performed.

  As a result, the facial expression detection unit 2004 determines whether or not the variation rate of the distance between the eyes and the cheek region edge density obtained by performing the processing described in the first embodiment are within the range indicated by this parameter. By checking, facial expression discrimination can be performed with higher accuracy.

  Returning to FIG. 21, next, the facial expression detection unit 2004 determines whether or not the facial expression determined in step S2106 is a specific facial expression set in advance. This is done by determining whether or not the code indicating the facial expression determined in step S2106 matches the preset code indicating the specific facial expression.

  If they match, the process proceeds to step S2108, and notification to that effect is sent to the control unit 1701, and the same process as step S1406 in FIG. 14 is performed.

  In this way, facial expression recognition tailored to the individual is performed after each individual is specified, so that the accuracy of facial expression recognition is improved. In addition, by detecting a face from an image, specifying an individual, and specifying the facial expression, a photograph of the desired facial expression of the desired individual from among a large number of people can be taken, for example, from among a plurality of children It will be possible to shoot a moment of smiling children. Furthermore, it can be used as an interface between humans and machines by recognizing specific individuals and facial expressions.

  In the above embodiment, the “certain specific person” and the “certain facial expression” can be appropriately set by the user via a predetermined operation unit. Therefore, if set appropriately, naturally, codes indicating them are also changed accordingly.

  Moreover, in the said embodiment, although the to-be-photographed object was made into the face, not only this but a vehicle, a building, etc. may be sufficient.

[Other Embodiments]
An object of the present invention is to supply a recording medium (or storage medium) that records software program codes for realizing the functions of the above-described embodiments to a system or apparatus, and the computer of the system or apparatus (or CPU or MPU). Needless to say, this can also be achieved by reading and executing the program code stored in the recording medium. In this case, the program code itself read from the recording medium realizes the functions of the above-described embodiment, and the recording medium on which the program code is recorded constitutes the present invention.

  Further, by executing the program code read by the computer, not only the functions of the above-described embodiments are realized, but also an operating system (OS) running on the computer based on the instruction of the program code. It goes without saying that a case where the function of the above-described embodiment is realized by performing part or all of the actual processing and the processing is included.

  Furthermore, after the program code read from the recording medium is written into a memory provided in a function expansion card inserted into the computer or a function expansion unit connected to the computer, the function is based on the instruction of the program code. It goes without saying that the CPU or the like provided in the expansion card or the function expansion unit performs part or all of the actual processing and the functions of the above-described embodiments are realized by the processing.

  When the present invention is applied to the recording medium, program code corresponding to the flowchart described above is stored in the recording medium.

It is a figure which shows the function structure of the image processing apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart of the main process for discriminating the facial expression in a picked-up image. It is a figure which shows the function structure of the image processing apparatus which concerns on the 2nd Embodiment of this invention. 4 is a timing chart showing the operation of the configuration shown in FIG. 3. It is a figure which shows the function structure of the image processing apparatus which concerns on the 3rd Embodiment of this invention. 6 is a timing chart showing the operation of the configuration shown in FIG. 5. It is a figure which shows a series of processes for detecting the local feature in a picked-up image, and specifying a face area. It is a figure which shows the structure of the neural network for performing image recognition. It is a figure which shows each feature point. It is a figure for demonstrating the process which calculates | requires the feature point using a primary feature and a tertiary feature in the face area | region shown in FIG. 1 is a diagram illustrating a basic configuration of an image processing apparatus according to a first embodiment of the present invention. It is a figure which shows the structure of the example which used the image processing apparatus which concerns on the 1st Embodiment of this invention for an imaging device. It is a figure which shows the function structure of the image processing apparatus which concerns on the 4th Embodiment of this invention. It is a flowchart of the main process for discriminating who the face in a picked-up image belongs. It is a figure explaining the feature vector used by individual discrimination processing. It is the figure which showed the data used at the time of learning in each of three classifiers as a table | surface. It is a figure which shows the function structure of the image processing apparatus which concerns on the 5th Embodiment of this invention. It is a flowchart of the main process for discriminating who the face in the photographed image belongs and what kind of expression. It is a figure which shows the structural example of the data which the integrated part 1708 manages. It is a figure which shows the function structure of the image processing apparatus which concerns on the 6th Embodiment of this invention. It is a flowchart of the main process which the image processing apparatus which concerns on the 6th Embodiment of this invention performs. It is a figure which shows the structural example of facial expression discrimination | determination data.

Claims (15)

An input means for inputting an image including a face;
A face region that detects a plurality of local features representing edges from an image input by the input means, detects a facial feature from a combination of the detected plurality of local features, and identifies the region of the face in the image Specific means,
The difference between the relative position of each local feature in the face region detected by the face region specifying means and the relative position of each local feature with respect to a face image set in advance as a reference is used. An image processing apparatus comprising: a discriminating unit that discriminates a category to which the data belongs.   The image processing apparatus according to claim 1, wherein the face area specifying unit includes a hierarchical neural network, and the determination unit uses an intermediate layer output of the hierarchical neural network as a detection result of the local feature.   The image processing apparatus according to claim 1, wherein the face category is a facial expression. The face area specifying unit detects a first local feature group in the image input by the input unit, and each second local feature has a specific positional relationship in the first local feature group. By repeatedly performing a process of obtaining a second local feature group obtained by combining a plurality of first local features of a specific type from the image, the nth local feature group (n ≧ 2) is obtained,
The discrimination means includes a relative position of any one of the local feature groups from the first local feature group to the n-th local feature group in the face area specified by the face area specifying means, The image processing apparatus according to claim 3, wherein the facial expression is discriminated using a difference from a relative position obtained in advance as a reference of the local feature.   The discriminating means has a distribution according to a difference between a relative position of each of the local features in the face region and a relative position obtained as a reference in advance for each of the local features in the face region. The image processing according to claim 3, wherein the image processing unit calculates which of the distributions corresponding to each facial expression obtained in advance and which has the highest similarity, and determines the facial expression represented by the distribution with the highest similarity. apparatus. The input means performs a process of inputting images continuously by continuously performing a process of inputting a next image when the face area specifying means completes the process of specifying a face area,
The discrimination means is a relative position of each local feature in the face area specified by the face area specifying means using the image input by the input means at the previous stage when the input means inputs an image. 4. The image according to claim 3, wherein a process of discriminating the facial expression of the face is performed based on a difference between the local position in the face region and a relative position obtained in advance as a reference. 5. Processing equipment.   The image processing apparatus according to claim 1, wherein the determination unit determines who the face is as the face category. Input means for continuously inputting frame images including a face;
A face for detecting a plurality of local features representing an edge from a frame image input by the input means, detecting a facial feature from a combination of the detected plurality of local features, and identifying a facial region in the frame image Area identification means;
In the image of the second frame, which is a frame after the first frame, corresponding to the position of the face area specified by the face area specifying means in the image of the first frame input by the input means. In the region, the facial expression is determined based on the difference between the relative position of each of the local features detected by the face region specifying unit and the relative position of each of the local features with respect to a face image set in advance as a reference. An image processing apparatus comprising: a determining unit that An input means for inputting an image including a face;
Face region specification for detecting a plurality of local features representing an edge from an image input by the input means, detecting a facial feature from a combination of the detected plurality of local features, and specifying a face region in the image Means,
The input means using the detection results of the local features in the face area detected by the face area specifying means and the detection results of the local features obtained in advance from the images of the faces. A first determination unit that determines who the face in the image input is, a relative position of each local feature in the face region specified by the face region specifying unit, and a reference in advance An image processing apparatus comprising: a second determination unit that determines a facial expression using a difference between the relative position of each local feature with respect to the face image set as. It said second determining means, according to claim 9, wherein the first, depending on the person who is determined by the determination means, characterized by using the parameters corresponding to the individual for the expression determination with respect to the face of interest Image processing apparatus. An input process for inputting an image including a face;
A face region that detects a plurality of local features representing edges from the image input in the input step, detects a facial feature from a combination of the detected plurality of local features, and identifies the region of the face in the image Specific process,
The difference between the relative position of each local feature in the face area detected in the face region specifying step and the relative position of each local feature with respect to a face image set in advance as a reference is used. And a discrimination step of discriminating a category to which the image belongs. An input process for continuously inputting frame images including a face;
A face for detecting a plurality of local features representing an edge from the frame image input in the input step, detecting a facial feature from a combination of the detected plurality of local features, and identifying a facial region in the frame image Region identification process;
In an image of a second frame, which is a frame after the first frame, corresponding to the position of the face region specified in the face region specifying step in the image of the first frame input in the input step In the region, the facial expression is determined based on the difference between the relative position of each local feature detected in the face region specifying step and the relative position of each local feature with respect to a face image set in advance as a reference. An image processing method comprising: a determining step. An input process for inputting an image including a face;
Face region specification for detecting a plurality of local features representing edges from the image input in the input step, detecting a facial feature from a combination of the detected plurality of local features, and specifying a face region in the image Process,
The input step using the detection result of each local feature in the face region detected in the face region specifying step and the detection result of each local feature obtained in advance from each face image A first discrimination step for discriminating who the face in the image inputted in step 1 is, a relative position of each local feature in the face region identified in the face region identification step, and a reference in advance And a second discrimination step of discriminating the facial expression using the difference between the relative position of each of the local features with respect to the face image set as. A program that causes a computer to execute the image processing method according to any one of claims 11 to 13 . A computer-readable storage medium storing the program according to claim 14 .

Images