JP2015219648A - Age estimation device, imaging device, age estimation method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accurately estimate the age of a subject.SOLUTION: In an age estimation device 100, a face detection part 110 detects a face area of a subject in a subject image including the subject. A featured value generation part 120 generates a featured value indicating features of the face of the subject from the face area detected by the face detection part 110. A facial expression detection part 130 detects a degree of facial expression of the subject from the face area detected by the face detection part 110. An age estimation part 140 estimates the age of the subject on the basis of the featured value generated by the featured value generation part 120 and the degree of facial expression detected by the facial expression detection part 130.

Description

  The present invention relates to an age estimation device, an imaging device, an age estimation method, and a program.

  There is an age estimation technique for estimating the age of a subject by analyzing the face image of the subject. In many age estimation techniques, a feature amount serving as an index for estimating age is generated from a face image, and the relationship between the feature amount and age is calculated from a large number of pre-collected learning images using an SVR (Support Vector). Learn using techniques such as Regression (regression support vector). Based on the learning result, the age of the subject to be age estimated is estimated.

  For example, Non-Patent Document 1 discloses a technique for obtaining a human age by obtaining an index for estimating an age from biologically generated features such as wrinkles in a face image.

  Further, as individual techniques used when generating a feature amount from a face image, Patent Documents 1 to 3 disclose a technique for detecting a face area of a subject from an image photographed by photographing means. Non-Patent Document 2 discloses a technique for extracting the position of a specific part such as the center position of the left and right eyes and the center position of the mouth from the face image.

Japanese Patent Laid-Open No. 2002-333651 JP 2005-234686 A JP 2008-300098 A

G. Guo, G .; Mu, Y. et al. Fu, and T.M. S. Huang, “Human Age Estimating Using Bio-inspired Features”, IEEE Conference on Computer Vision and Pattern Recognition, 112-119, June, 2009. T. T. et al. F. Coutes, and C.I. J. et al. Taylor, “Statistical Models of Appearance for Computer Vision”, Imaging Science and Biomedical Engineering, University of Manchester, Manchester 4

  In learning using SVR for age estimation, it is necessary to prepare face images (training data) including various age groups, races, genders, and the like and their ages (class labels). At this time, if face images having various expressions are mixed in the learning data, there is a problem that the accuracy of age estimation is lowered.

  For example, when the age estimation target person's facial expression includes a smile or anger, the age estimation accuracy tends to decrease, such as estimating the age of the target person high. As a factor of the decrease in estimation accuracy, especially due to age-related features that express wrinkles and smoothness of skin due to aging, it is attributed to wrinkles and aging caused by facial expressions such as smiles and anger It is possible that the distinction from moths has not been determined.

  In order to avoid such a decrease in age estimation accuracy, conventionally, the SVR is learned using an expressionless face image, and the age is often estimated only for the expressionless face image. However, it has been required to accurately estimate the age even for subjects having various expressions such as smiles and anger.

  The present invention has been made to solve the above-described problems, and an object thereof is to provide an age estimation device, an age estimation method, and a program suitable for accurately estimating the age of a subject.

In order to achieve the above object, an age estimation apparatus according to the present invention includes:
A face detection unit that detects a face area of the subject in a subject image including the subject;
A feature amount generation unit that generates a feature amount indicating a feature of the face of the subject from the face area detected by the face detection unit;
A facial expression detection unit that detects the degree of facial expression of the subject from the face area detected by the face detection unit;
An age estimation unit that estimates the age of the subject based on the feature amount generated by the feature amount generation unit and the degree of facial expression detected by the facial expression detection unit;
It is characterized by providing.

  According to the present invention, the age of a subject can be accurately estimated.

It is a block diagram which shows the structure of the imaging device which concerns on Embodiment 1 of this invention. It is a block diagram which shows the function structure of the imaging device and age estimation apparatus which concern on Embodiment 1 of this invention. (A), (b), (c) is a figure which shows the example of the learning image containing the to-be-photographed object. (A) is a figure showing an example of detection of a face area. (B) is a figure showing an example of extraction of a position of a specific part. (C) is a figure which shows the example of a normalized image. It is a flowchart which shows the flow of the learning process which the age estimation apparatus which concerns on Embodiment 1 of this invention performs. It is a flowchart which shows the flow of the age estimation process which the age estimation apparatus which concerns on Embodiment 1 of this invention performs. It is a block diagram which shows the function structure of the imaging device and age estimation apparatus which concern on Embodiment 2 of this invention. It is a flowchart which shows the flow of the age estimation process which the age estimation apparatus which concerns on Embodiment 2 of this invention performs.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or corresponding parts are denoted by the same reference numerals.

(Embodiment 1)
The imaging apparatus according to Embodiment 1 of the present invention is configured as shown in FIG. The imaging device 1 includes an imaging unit 10, a data processing unit 20, and a user interface unit 30.

  The imaging unit 10 includes an optical lens 11 and an image sensor 12. The imaging unit 10 generates image data of the subject by imaging the subject.

  The optical lens 11 includes a lens that collects light emitted from a subject and a peripheral circuit for adjusting imaging setting parameters such as focus, exposure, and white balance.

  The image sensor 12 is configured by, for example, a CCD (Charge Coupled Device), a CMOS (Complementary Metal Oxide Semiconductor), or the like. The image sensor 12 acquires an optical image of a subject formed by the optical lens 11 condensing light, and converts the voltage information of the acquired optical image into digital information by an analog / digital converter (not shown). Convert. Then, the obtained digital image data is stored in the memory 21.

  The data processing unit 20 includes a memory 21, a video output unit 22, a storage unit 23, a CPU (Central Processing Unit) 24, and an age estimation device 100.

  The memory 21 is composed of, for example, a RAM (Random Access Memory) and stores image data generated by the imaging unit 10.

  The video output unit 22 reads the image data stored in the memory 21, generates RGB (R (red), G (green), B (blue)) signals corresponding to the read image data, and displays the display unit 31. Output to.

  The storage unit 23 is configured by a nonvolatile memory such as a ROM (Read-Only Memory) or a flash memory, for example. The storage unit 23 stores various control programs executed by the CPU 24 and data obtained by various processes of the imaging device 1.

  The CPU 24 controls the overall operation of the imaging apparatus 1 by executing various arithmetic processes according to a program stored in the storage unit 23 while using the memory 21 as a temporary storage area.

  The user interface unit 30 includes a display unit 31, an operation unit 32, an external interface 33, and an external storage medium 34.

  The display part 31 is comprised by a liquid crystal display device, an organic EL (Electro Luminescence) display, etc., for example. Based on the RGB signal supplied from the video output unit 22, the display unit 31 displays various image data such as an image captured by the imaging device 1 and an image showing an operation menu of the imaging device 1.

  The operation unit 32 receives an operation instruction from the user. The operation unit 32 includes various operation buttons such as a power switch, a shutter button, and a button for selecting various functions of the imaging device 1. When receiving an operation instruction from the user, the operation unit 32 supplies the received instruction information to the CPU 24 of the data processing unit 20.

  In addition, the display part 31 and the operation part 32 may be comprised by what is called a touch panel arrange | positioned mutually superimposed.

  The external interface 33 is an interface for exchanging data with an external device of the imaging apparatus 1. For example, the external interface 33 converts image data obtained by imaging by the imaging device 1 into USB (Universal Serial Bus) standard data, and transmits / receives data to / from an external device via a USB cable. To do.

  The external storage medium 34 is a storage medium for exchanging data with devices external to the imaging apparatus 1.

  The age estimation device 100 estimates the age of the subject to be age estimated, for example, obtained by imaging by the imaging unit 10. The age estimation device 100 is controlled by the CPU 24 and functions like each unit shown in FIG. Specifically, the age estimation apparatus 100 functions as a face detection unit 110, a feature amount generation unit 120, a facial expression detection unit 130, an age estimation unit 140, a classification unit 150, and a learning unit 160.

With these functions, the age estimation apparatus 100 executes the following processes (1) and (2). Hereinafter, the processes (1) and (2) will be described in detail with reference to the drawings.
(1) Process for learning age identification criterion using learning subject (2) Process for estimating age of subject of age estimation based on learning result

  First, (1) a process of learning age identification criteria using a learning subject will be described.

  This learning process is executed in a dedicated learning mode different from the normal imaging process executed by the imaging apparatus 1. This learning process may be executed at the product adjustment stage such as when the imaging device 1 is manufactured or shipped from the factory, or may be executed under a general user.

  In the learning process, the face detection unit 110 detects a face area of the learning subject in a plurality of learning images each including the learning subject.

  As learning images, the imaging apparatus 1 includes learning images 200a, 200b, 200c including learning subjects 210a, 210b, 210c having various facial expressions as shown in FIGS. 3A to 3C, for example. The image is captured by the image capturing unit 10 or acquired from an external device via the external interface 33 or the external storage medium 34, and is collected in advance and stored in the storage unit 23 or the like.

  When the learning process is started, the face detection unit 110 acquires learning images from the storage unit 23, sequentially selects each of the acquired learning images, and detects a face region of the learning subject.

  FIG. 4A shows an example in which the face area 220 of the learning subject 210b is detected when the face detection unit 110 selects the learning image 200b. The face detection unit 110 detects a face region 220 corresponding to the face of the learning subject 210b from the luminance distribution of the learning image 200b.

  For example, the face detection unit 110 identifies a skin color pixel based on the brightness, saturation, and hue of each pixel, and detects the identified skin color pixel region as the face region 220. Alternatively, the face detection unit 110 extracts a shape corresponding to a facial component such as the mouth and eyes of the learning subject 210b from the learning image 200b, and detects the face region 220 on the basis of the position of the structural component. To do. As a specific method for detecting such a face area, for example, the techniques disclosed in Patent Documents 1 to 3 described above can be used.

  The feature amount generation unit 120 generates a feature amount indicating the facial feature of the learning subject 210b from the face region 220 detected by the face detection unit 110. Specifically, the feature quantity generation unit 120 functions as a part extraction unit 121 and a normalization unit 122.

  The part extraction unit 121 extracts the position of the specific part in the face region 220 detected by the face detection unit 110. For example, the part extracting unit 121 detects the shape of a part constituting the face such as both eyes and mouth by a method such as edge detection, and extracts the center position of the detected part. Specifically, as illustrated in FIG. 4B, the part extracting unit 121 extracts three points, that is, the center positions of the right eye 221 and the left eye 222 and the center position of the mouth 223. As a specific method for extracting the position of such a specific part, for example, the technique disclosed in Non-Patent Document 2 described above can be used.

  The normalization unit 122 normalizes the learning image 200b to an optimal image for generating a feature amount based on the positions of the right eye 221, the left eye 222, and the mouth 223 extracted by the part extraction unit 121. Specifically, as shown in FIG. 4C, the normalization unit 122 causes the face region 220 to occupy most of the center of the image by scaling, rotation, translation, etc. of the learning image 200b. Then, a normalized image 230 having a size of height (vertical width) H and width (horizontal width) W is generated.

  More specifically, the normalization unit 122 performs learning so that the positions of the three points of the right eye 221, the left eye 222, and the mouth 223 extracted by the part extraction unit 121 are constant positions in the normalized image 230. An affine matrix is calculated from the coordinates of these three points in the image 200b and the coordinates of these three points after normalization. Specifically, as illustrated in FIG. 4C, the normalization unit 122 has both eyes (the right eye 221 and the left eye 222) positioned at a predetermined height ev from the lower end of the normalized image 230, and the mouth 223. Is located at a predetermined height mv from the lower end of the normalized image 230, and the affine matrix is calculated such that the distance between the eyes is equal to the predetermined distance eh. Then, the normalization unit 122 deforms the learning image 200b by affine transformation according to the calculated affine matrix, and generates a normalized image 230.

  The feature amount generation unit 120 generates a feature amount indicating the feature of the face of the learning subject 210b from the face region 220 in the learning image 200b (that is, the normalized image 230) after normalization by the normalization unit 122. This feature amount is an amount that serves as an index for estimating the age from the face. Specifically, as in the case of Non-Patent Document 1 described above, the degree of features that generally change with age, such as various wrinkles and eye bags that appear on a person's face, is used as the feature amount. it can.

  For example, the feature quantity generation unit 120 detects feature edges such as wrinkles by detecting local edges included in the face region 220 using a technique such as a Gabor filter or an LBP (Local Binary Pattern) histogram. Generate. The Gabor filter is a filter for extracting which direction line or edge is included in the image. LBP is a method in which 0 or 1 bits are created and coded based on the magnitude relationship between a central pixel value and its peripheral pixel values. By using the LPB histogram, it is possible to detect edge shape information and the like included in the image.

  Using such a method, the feature amount generation unit 120 divides the face region 220 into a plurality of partial regions, for example, in units of parts constituting the face, and generates feature amounts in units of the divided partial regions. Since the image is divided into a plurality of partial areas, the feature amount is calculated as a value (space variant) depending on the position in the image. Therefore, in order to generate feature quantities with high accuracy, it is necessary to normalize the learning images 200a to 200c by the normalizing unit 122 described above.

  The facial expression detection unit 130 detects the degree of facial expression of the learning subject from the face region 220 detected by the face detection unit 110. Specifically, the facial expression detection unit 130 detects the degree of smile of the learning subject as the degree of facial expression.

  As a specific method for detecting the degree of smile, for example, the technique disclosed in Patent Document 2 or Patent Document 3 described above can be used. Patent Document 2 discloses a technique for generating an index representing a facial expression on an image such as a smile using a parameter indicating the degree of opening of the mouth and a parameter indicating the appearance of teeth. Patent Document 3 discloses a technique for calculating a smile degree by acquiring a difference between reference face information for each component of a face of a subject and other face information and summing the acquired differences. Yes.

  The facial expression detection unit 130 detects the smile level of the learning subject as the smile level k using such a method. For example, the facial expression detection unit 130 detects a low smile level of k = 0 for the learning subject 210a shown in FIG. 3A, and k = 0 for the learning subject 210b shown in FIG. A medium smile level of .5 is detected, and a high smile level of k = 1 is detected for the learning subject 210c shown in FIG.

  The classifying unit 150 classifies a plurality of learning images collected in advance into a smiling group and a non-smiling group according to the smile k of the learning subject detected by the facial expression detection unit 130. Specifically, the classification unit 150 classifies the learning image into a smiling group and a non-smiling group according to whether or not the smile degree k exceeds a predetermined threshold th. The threshold th is a value for determining whether the facial expression of the learning subject included in the learning image is a smile or a non-smile, and is set to a value between 0 and 1.

  The learning unit 160 identifies the age of the smile group and the non-smile group based on the feature amount based on the relationship between the age of the learning subject and the feature amount in the learning images classified into the groups. Learning identification criteria for

  More specifically, the learning unit 160 learns the first identification criterion using a technique such as SVR (support vector regression) using only the learning images classified as the smile group as samples. As a result, a smile age classifier 141 is generated. In addition, the learning unit 160 learns the second identification criterion by a method such as SVR using only the learning images classified as the non-smiling group as samples. As a result, the non-smile age classifier 142 is generated.

  Using the smile age classifier 141 and the non-smile age classifier 142 generated in this way, the age estimation device 100 executes an age estimation process of a subject to be described later.

  The flow of the learning process as described above will be described with reference to the flowchart shown in FIG. This learning process is started when a learning start instruction is received via the operation unit 32 or the like, for example.

  When the learning process is started, the face detection unit 110 selects any one of the plurality of learning images collected in advance (step S1), and determines the face area of the learning subject in the selected learning image. Detect (step S2). That is, for example, as illustrated in FIG. 4A, the face detection unit 110 detects the face area 220 of the learning subject 210b in the learning image 200b.

  When the face detection unit 110 detects the face region 220, the feature amount generation unit 120 normalizes the learning image (step S3), and the feature amount indicating the facial feature of the learning subject 210b from the detected face region 220. Is generated (step S4). That is, the part extracting unit 121 extracts the position of a specific part included in the face region 220 as shown in FIG. 4B, for example, and the normalizing unit 122 is shown as shown in FIG. 4C, for example. The normalized image 230 is generated based on the extracted position. Then, the feature amount generation unit 120 generates a feature amount from the face region 220 in the normalized image 230 by the method described above.

  When the feature amount generation unit 120 generates a feature amount, the facial expression detection unit 130 detects the degree of smile (smile level k) as the degree of facial expression of the learning subject from the face area 220 detected by the face detection unit 110. (Step S5).

  When the facial expression detection unit 130 detects the smile level k, the classification unit 150 determines whether or not the detected smile level k exceeds a threshold th set to a value between 0 and 1, thereby smiling. A determination is made (step S6). When the smile k exceeds the threshold th (step S6; YES), the classification unit 150 classifies the selected learning image into a smile group (step S7). On the other hand, when the smile level k is equal to or less than the threshold th (step S6; NO), the classification unit 150 classifies the selected learning image into a non-smile group (step S8).

  When the selected learning image is classified into either the smile group or the non-smile group, the classification unit 150 determines whether or not the classification has been completed for all the learning images (step S9).

  If the classification has not ended (step S9; NO), the face detection unit 110 selects another learning image from the learning images stored in the storage unit 23 (step S10). Then, the learning process returns to step S2, and the processes from steps S2 to S9 are repeated to classify the newly selected learning image into either a smiling group or a non-smiling group. Such a classification process is repeated for all learning images prepared in advance.

  When the classification by the classification unit 150 is finally completed for all learning images (step S9; YES), the learning unit 160 generates the smile age classifier 141 using the learning images classified into the smile group as a sample (step S9). S11) and the non-smiling age classifier 142 is generated using the learning images classified into non-groups as samples (step S12). Thus, the learning process shown in the flowchart of FIG. 5 ends.

  Secondly, (2) a process for estimating the age of the subject to be estimated based on the learning result will be described.

  In the age estimation process, the imaging device 1 captures a subject image including a subject to be age estimated by the imaging unit 10 or an image obtained by capturing an object in advance by an external device, for example, the external interface 33. Or via the external storage medium 34.

  The face detection unit 110 detects the face area of the subject in the acquired subject image. The feature value generation unit 120 generates a feature value indicating the feature of the face of the subject from the face area of the subject detected by the face detection unit 110. The expression detection unit 130 detects the degree of smile (smile degree k) as the degree of expression of the subject from the face area of the subject detected by the face detection unit 110.

  The processing that each of the face detection unit 110, the feature amount generation unit 120, and the facial expression detection unit 130 executes on the subject image including the subject has been described with reference to FIGS. This is the same as the processing executed for the learning image 200b including the learning subject 210b. This is because it is preferable to use the same algorithm at the time of learning and at the time of age estimation in order to estimate the age of the subject with high accuracy. Therefore, detailed description is omitted here.

  The age estimation unit 140 estimates the age of the subject based on the feature amount generated by the feature amount generation unit 120 and the smile level k detected by the facial expression detection unit 130. Specifically, the age estimation unit 140 functions as a smile age classifier 141, a non-smile age classifier 142, and a selection unit 143.

  The smile age classifier 141 and the non-smile age classifier 142 each identify the age of the subject from the feature amount generated by the feature amount generation unit 120 based on different identification criteria learned by the learning unit 160. Specifically, the smile age classifier 141 functions as a first age discrimination unit, and acquires the smile age SAge based on the first identification criterion learned from the learning images classified into the smile group. . The non-smiling age classifier 142 functions as a second age identifying unit, and acquires the non-smiling age Nage based on the second identification criterion learned from the learning images classified into the non-smiling group.

  Smile age Sage is the estimated age of the subject when it is assumed that the facial expression of the subject is a smile. The non-smile age Nage is an estimated age of the subject when it is assumed that the facial expression of the subject is not a smile. When estimating age using facial wrinkles as features, it is difficult to distinguish between wrinkles due to aging and wrinkles due to smiles. The age Sage is lower than the non-smile age Nage obtained by using only the learning subject that is not smiling as a sample.

  The selection unit 143 selects one of the smile age classifier 141 and the non-smile age classifier 142 according to the smile level k detected by the facial expression detection unit 130. Then, the age estimation unit 140 estimates the age identified by the age identification unit selected by the selection unit 143, that is, either the smile age SAge or the non-smile age Nage as the age of the subject.

  More specifically, the selection unit 143 selects the smile age classifier 141 as a classifier for identifying the age of the subject when the smile level k exceeds a predetermined threshold th, and the smile level k is a predetermined level. When it is equal to or less than the threshold th, the non-smile age classifier 142 is selected. The threshold th is a value for determining whether the subject's facial expression is a smile or a non-smile, and is set to the same value as during learning.

  As described above, when the selection unit 143 selects the age discriminator according to the degree of smile of the subject, the age estimation unit 140 estimates the smile age Sage as the age of the subject when the subject's facial expression is smiling, When the facial expression of the subject is not a smile, the non-smile age Nage is estimated as the age of the subject.

  In other words, if the facial feature amount is the same, the age estimating unit 140 sets the lower age of the subject as the degree of smile of the subject is larger than the smile of the subject whose facial expression is smile. Estimated as age. With such an estimation method, the age of the subject can be estimated appropriately according to the degree of smile, especially because it is difficult to distinguish between wrinkles due to aging and wrinkles due to smiles. It is possible to prevent a decrease in the accuracy of age estimation, such as.

  The output unit 40 outputs the estimation result of the age of the subject by the age estimation unit 140. For example, the display unit 31, the external interface 33, or the external storage medium 34 functions as the output unit 40. The output unit 40 displays the age estimation result on the display unit 31 under the control of the video output unit 22 or provides the age estimation result to an external device via the external interface 33 or the external storage medium 34. .

  The flow of the age estimation process as described above will be described with reference to the flowchart (age estimation process 1) shown in FIG. This age estimation process is started when an instruction to estimate the age of the subject is received from the user operating the imaging device 1 via the operation unit 32 or the like, for example.

  When the age estimation process starts, the face detection unit 110 acquires a subject image including the subject of age estimation target from the memory 21, the storage unit 23, or the like (step S101), and detects the face area of the subject in the acquired subject image. (Step S102).

  When the face detection unit 110 detects the face area 220, the feature amount generation unit 120 normalizes the subject image (step S103), and generates a feature amount indicating the feature of the subject's face from the normalized subject image. (Step S104).

  When the feature amount generation unit 120 generates the feature amount, the facial expression detection unit 130 detects the degree of smile of the subject (smile level k) from the face area detected by the face detection unit 110 (step S105).

  That is, in the age estimation process, the age estimation apparatus 100 performs processing similar to the process executed on the learning image in the learning process (each process of face detection, feature amount generation, and smile level detection). Run on the image.

  When the facial expression detection unit 130 detects the smile level k, the selection unit 143 performs smile determination by determining whether or not the detected smile level k exceeds a predetermined threshold th (step S106).

  When the smile level k exceeds the threshold th (step S106; YES), the selection unit 143 selects the smile age classifier 141. The smile age classifier 141 acquires the smile age Sage from the feature amount generated by the feature amount generation unit 120 (step S107). The age estimation unit 140 estimates the acquired smile age Sage as the age of the subject.

  On the other hand, when the smile level k is less than or equal to the threshold th (step S106; NO), the selection unit 143 selects the non-smile age classifier 142. The non-smile age classifier 142 acquires the non-smile age Nage from the feature amount generated by the feature amount generation unit 120 (step S108). The age estimation unit 140 estimates the acquired non-smile age Nage as the age of the subject.

  When the estimation result of the age of the subject by the age estimation unit 140 is obtained, the output unit 40 outputs the estimation result of the age of the subject by the age estimation unit 140 to the outside by displaying it on the display unit 31 or the like (step S109). ). Thus, the age estimation process shown in the flowchart of FIG. 6 ends.

  As described above, the age estimation device 100 and the imaging device 1 according to the first embodiment use the learning age classifier 141 and the non-smile age using the learning images that are classified in advance into the smile group and the non-smile group. The discriminator 142 is generated. Then, one of the smile age classifier 141 and the non-smile age classifier 142 is selected according to the smile level of the subject whose age is to be estimated, and the age of the subject is estimated by the selected classifier.

  As a result, it is possible to avoid a decrease in age estimation accuracy caused by whether the subject's facial expression is smiling or non-smiling, such as estimating the age of the smiling subject high, and it has either a smiling or non-smiling facial expression It is possible to accurately estimate the age of the subject.

  The above-described age estimation apparatus 100 includes two age identification units, a smile age classifier 141 and a non-smile age classifier 142. However, the age estimation apparatus 100 may include three or more age identification units generated according to the smile level of the learning subject.

  For example, when the age estimating apparatus 100 includes three age identifying units, in the learning stage, the classifying unit 150 selects a plurality of learning images collected in advance according to the smile level of the learning subject detected by the facial expression detecting unit 130. And classify them into three groups with low, medium and high smile levels. Then, for each of these three groups, the learning unit 160 identifies the age based on the feature amount based on the relationship between the age of the learning subject and the feature amount in the learning images classified into the respective groups. Learning identification criteria for In the age estimation stage, the selection unit 143 selects an age identification unit that matches the smile level of the subject to be age estimated from the three age identification units, and uses it for age estimation.

  In this way, by increasing the number of age discriminating units, it is possible to select an optimal age discriminator according to the facial expression of the subject, so that the accuracy of age estimation can be improved.

(Embodiment 2)
Next, Embodiment 2 of the present invention will be described.

  FIG. 7 shows functional configurations of the imaging apparatus and the age estimation apparatus according to the second embodiment. The imaging device 1a and the age estimation device 100a according to the second embodiment include an age calculation unit 144 in the age estimation unit 140a instead of the selection unit 143 in the age estimation unit 140 according to the first embodiment. Other configurations are the same as the configurations included in the imaging device 1 and the age estimation device 100 according to the first embodiment.

  The age estimation apparatus 100a according to the second embodiment uses the functions of the classification unit 150, the learning unit 160, and the like, prior to subject age estimation processing, (1) a process of learning age identification criteria using a learning subject. Run. Since this learning process is the same as that of Embodiment 1, description is abbreviate | omitted here.

  In the second embodiment, (2) the process of estimating the age of the subject of age estimation based on the learning result will be described with reference to the flowchart (age estimation process 2) shown in FIG.

  The processes of steps S201 to S205 in this age estimation process are the same as the processes of steps S101 to S105 (FIG. 6) in the age estimation process of the first embodiment.

  That is, when the age estimation process is started, the face detection unit 110 acquires a subject image including the subject of age estimation target from the memory 21, the storage unit 23, etc. (step S201), and the face area of the subject in the acquired subject image Is detected (step S202).

  When the face detection unit 110 detects the face region 220, the feature amount generation unit 120 normalizes the subject image (step S203), and generates a feature amount indicating the feature of the subject's face from the normalized subject image. (Step S204).

  When the feature amount generation unit 120 generates the feature amount, the facial expression detection unit 130 detects the degree of smile (smile level k) of the subject from the face area detected by the face detection unit 110 (step S205).

  When the facial expression detection unit 130 detects the smile degree k, the smile age classifier 141 acquires the smile age Sage from the feature amount generated by the feature amount generation unit 120 (step S206). Subsequently, the non-smile age classifier 142 acquires the non-smile age Nage from the feature amount generated by the feature amount generation unit 120 (step S207).

  The age calculation unit 144 calculates a weighted linear sum of the smile age Sage and the non-smile age Nage based on the smile level k of the subject detected by the facial expression detection unit 130, thereby obtaining the smile age Sage and the non-smile age Nage. Is calculated as the estimated age (step S208).

Specifically, based on the following formula (1), the age calculating unit 144 calculates an age closer to the smile age SAge as the smile degree k is larger (an age closer to the non-smile age Nage as the smile degree k is smaller). Calculated as the estimated age Age.
Age = k × Sage + (1−k) × Nage (1)

  Normally, the smile age Sage has a lower value than the non-smile age Nage, so the age calculation unit 144 calculates the lower age as the smile degree k is larger as the estimated age of the subject. The age estimation unit 140a estimates the age calculated by the age calculation unit 144 using such a relational expression as the age of the subject.

  When the estimation result of the subject's age by the age estimation unit 140a is obtained, the output unit 40 outputs the estimation result of the subject's age by the age estimation unit 140a to the outside by displaying it on the display unit 31 (step S209). ). Thus, the age estimation process shown in the flowchart of FIG. 8 ends.

  As described above, the imaging device 1a and the age estimation device 100a according to the second embodiment have the smile age classifier 141 and the non-smile age generated using the learning images classified into the smile group and the non-smile group. Using both the classifier 142, the smile age Sage and the non-smile age Nage of the subject of age estimation are acquired. Then, an estimated age of the subject is calculated by taking a weighted linear sum of the smile age SAge and the non-smile age Nage according to the smile level of the subject.

  As a result, the age of the subject can be accurately estimated in consideration of the influence of the degree of smile on the subject on the age estimation.

  The above equation (1) is an example. The age calculation unit 144 is not limited to the linear sum, and for example, when another relational expression that can estimate the age more accurately from the smile k is known, the estimated age of the subject is determined according to such other relational expression. May be calculated,

(Modification)
Although the embodiments of the present invention have been described above, these embodiments are merely examples, and the scope of application of the present invention is not limited thereto. That is, the embodiments of the present invention can be applied in various ways, and all the embodiments are included in the scope of the present invention.

  For example, in the above embodiment, the facial expression detection unit 130 detects the degree of smile as the degree of facial expression of the subject. However, the age estimation device and the imaging device according to the present invention may detect the degree of other facial expressions such as anger instead of or in addition to a smile. A learning image classified according to the anger level of the learning subject is used as a sample in advance to generate a plurality of classifiers by anger level, and the age of the subject is determined using these classifiers. It may be estimated.

  As described above, by using the classifier according to the degree of various facial expressions such as smile and anger, the age can be accurately estimated even for subjects having various facial expressions.

  In the above embodiment, the age estimation devices 100 and 100a are mounted inside the imaging devices 1 and 1a. However, the age estimation apparatus according to the present invention may be an apparatus independent of the imaging apparatuses 1 and 1a. For example, a general information processing apparatus such as a personal computer can function as an age estimation apparatus. In this case, the age estimation device acquires a subject image obtained by an imaging device independent from the own device via data communication means with the outside, and executes the above-described age estimation processing on the acquired subject image. To do.

  In the above embodiment, the age estimation devices 100 and 100a have the function of learning age identification criteria (classification unit 150 and learning unit 160) using a learning subject. However, the age estimation apparatus according to the present invention may not have a learning function. For example, the age estimation device according to the present invention acquires a learning result learned in advance in another device independent of the device itself through data communication means with the outside, and has been described above based on the acquired learning result. It can also comprise so that an age estimation process may be performed.

  The age of the subject estimated by the age estimation devices 100 and 100a according to the first embodiment may be a real age or an apparent age. When the age to be estimated is the real age, the learning unit 160 learns the identification criterion described above using the real age of the learning subject. When the age to be estimated is the apparent age, the learning unit 160 learns the identification criterion described above using the apparent age of the learning subject. Regardless of which age is estimated, conventionally, when the degree of facial expression such as smile or anger is large, the age of the subject tends to be estimated high. However, the age estimation device according to the present invention can accurately estimate the actual age or the apparent age of the subject.

  It should be noted that not only can an age estimation apparatus provided with a configuration for realizing the functions according to the present invention be provided in advance, an existing personal computer, an information terminal device, or the like can be provided as an age estimation apparatus according to the present invention by applying a program. It can also function. That is, by applying a program for realizing each functional configuration by the age estimation devices 100 and 100a exemplified in the above embodiment so that a CPU or the like that controls an existing personal computer or information terminal device can be executed, It can function as an age estimation apparatus according to the present invention. The age estimation method according to the present invention can be implemented using an age estimation device.

  Moreover, the application method of such a program is arbitrary. For example, the program can be stored and applied to a computer-readable recording medium (CD-ROM (Compact Disc Read-Only Memory), DVD (Digital Versatile Disc), MO (Magneto Optical disc), etc.), the Internet, etc. It is also possible to apply the program by storing it in a storage on the network and downloading it.

  As mentioned above, although preferable embodiment of this invention was described, this invention is not limited to the specific embodiment which concerns, This invention includes the invention described in the claim, and its equivalent range It is. Hereinafter, the invention described in the scope of claims of the present application will be appended.

(Appendix 1)
A face detection unit that detects a face area of the subject in a subject image including the subject;
A feature amount generation unit that generates a feature amount indicating a feature of the face of the subject from the face area detected by the face detection unit;
A facial expression detection unit that detects the degree of facial expression of the subject from the face area detected by the face detection unit;
An age estimation unit that estimates the age of the subject based on the feature amount generated by the feature amount generation unit and the degree of facial expression detected by the facial expression detection unit;
An age estimation device comprising:

(Appendix 2)
The feature quantity generation unit
A part extracting unit that extracts a position of a specific part in the face area detected by the face detecting unit;
A normalization unit that normalizes the subject image based on the position extracted by the part extraction unit;
Including
Generating the feature amount from the face area in the subject image after normalization by the normalization unit;
The age estimation apparatus according to supplementary note 1, wherein:

(Appendix 3)
The age estimating unit
A plurality of age discriminating units each identifying the age of the subject from the feature amount based on different discrimination criteria;
A selection unit that selects one of the plurality of age identification units according to the degree of the facial expression detected by the facial expression detection unit;
Including
The age identified by the age identifying unit selected by the selecting unit is estimated as the age of the subject.
The age estimation apparatus according to Supplementary Note 1 or 2, characterized by:

(Appendix 4)
The face detection unit detects a face area of the learning subject in a plurality of learning images each including a learning subject;
The feature amount generation unit generates a feature amount indicating a feature of the face of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images.
The facial expression detection unit detects the degree of facial expression of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images;
A classifying unit that classifies the plurality of learning images into a plurality of groups according to the degree of facial expression of the learning subject detected by the facial expression detection unit;
For each of the plurality of groups, a learning unit that learns the identification criterion based on the relationship between the age of the learning subject and the feature amount in the learning images classified into the respective groups;
Further comprising
The age estimation apparatus according to Supplementary Note 3, wherein

(Appendix 5)
The age estimating unit
A first age identification unit for identifying the age of the subject from the feature amount based on a first identification criterion;
A second age identifying unit for identifying the age of the subject from the feature amount based on a second identification criterion;
An age calculating unit that calculates an age between the age identified by the first age identifying unit and the age identified by the second age identifying unit based on the degree of facial expression detected by the facial expression detecting unit When,
Including
Estimating the age calculated by the age calculation unit as the age of the subject;
The age estimation apparatus according to Supplementary Note 1 or 2, characterized by:

(Appendix 6)
The face detection unit detects a face area of the learning subject in a plurality of learning images each including a learning subject;
The feature amount generation unit generates a feature amount indicating a feature of the face of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images.
The facial expression detection unit detects the degree of facial expression of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images;
Among the plurality of learning images, learning images in which the degree of facial expression of the learning subject detected by the facial expression detection unit exceeds a predetermined threshold are classified into a first group, and the facial expression detection unit detects the learning image. A classifying unit for classifying learning images in which the degree of expression of the learning subject is equal to or less than the predetermined threshold value into a second group;
Based on the relationship between the age of the learning subject and the feature amount in the learning image classified into the first group, the first identification criterion is learned, and the learning image classified into the second group A learning unit that learns the second identification criterion based on the relationship between the age of the learning subject and the feature amount in the image;
Further comprising
The age calculation unit calculates an age closer to the age identified by the first age identification unit as the estimated age of the subject as the degree of expression of the subject detected by the facial expression detection unit increases.
The age estimation apparatus according to appendix 5, characterized in that:

(Appendix 7)
The facial expression detection unit detects the degree of smile of the subject as the degree of facial expression;
The age estimation device according to any one of appendices 1 to 6, characterized in that:

(Appendix 8)
The age estimation unit estimates a lower age as the age of the subject as the degree of the facial expression detected by the facial expression detection unit is larger,
The age estimation device according to any one of appendices 1 to 7, characterized in that:

(Appendix 9)
The age estimation device according to any one of appendices 1 to 8,
An imaging unit that acquires the subject image by imaging the subject;
An output unit for outputting an estimation result of the age of the subject by the age estimation unit;
An imaging apparatus comprising:

(Appendix 10)
Detecting a face area of the subject in a subject image including the subject;
Generating a feature amount indicating a feature of the face of the subject from the face region;
Detecting the degree of expression of the subject from the face area;
Estimating the age of the subject based on the feature amount and the degree of facial expression;
The age estimation method characterized by including.

(Appendix 11)
On the computer,
A function for detecting a face area of a subject in a subject image including the subject;
A function of generating a feature amount indicating a feature of the face of the subject from the face region;
A function of detecting the degree of expression of the subject from the face area;
A function of estimating the age of the subject based on the feature amount and the degree of the facial expression;
A program to realize

DESCRIPTION OF SYMBOLS 1, 1a ... Imaging device, 10 ... Imaging part, 11 ... Optical lens, 12 ... Image sensor, 20 ... Data processing part, 21 ... Memory, 22 ... Video output part, 23 ... Memory | storage part, 24 ... CPU, 30 ... User Interface unit 31 ... Display unit 32 ... Operation unit 33 ... External interface 34 ... External storage medium 40 ... Output unit 100, 100a ... Age estimation device 110 ... Face detection unit 120 ... Feature quantity generation unit DESCRIPTION OF SYMBOLS 121 ... Site extraction part, 122 ... Normalization part, 130 ... Facial expression detection part, 140,140a ... Age estimation part, 141 ... Smile age classifier, 142 ... Non-smile age classifier, 143 ... Selection part, 144 ... Age calculation 150, classification unit, 160 learning unit, 200a, 200b, 200c learning image, 210a, 210b, 210c learning object, 220 face region, 221 right eye 222 ... left, 223 ... mouth, 230 ... normalized image

Claims (11)

A face detection unit that detects a face area of the subject in a subject image including the subject;
A feature amount generation unit that generates a feature amount indicating a feature of the face of the subject from the face area detected by the face detection unit;
A facial expression detection unit that detects the degree of facial expression of the subject from the face area detected by the face detection unit;
An age estimation unit that estimates the age of the subject based on the feature amount generated by the feature amount generation unit and the degree of facial expression detected by the facial expression detection unit;
An age estimation device comprising: The feature quantity generation unit
A part extracting unit that extracts a position of a specific part in the face area detected by the face detecting unit;
A normalization unit that normalizes the subject image based on the position extracted by the part extraction unit;
Including
Generating the feature amount from the face area in the subject image after normalization by the normalization unit;
The age estimation apparatus according to claim 1. The age estimating unit
A plurality of age discriminating units each identifying the age of the subject from the feature amount based on different discrimination criteria;
A selection unit that selects one of the plurality of age identification units according to the degree of the facial expression detected by the facial expression detection unit;
Including
The age identified by the age identifying unit selected by the selecting unit is estimated as the age of the subject.
The age estimation apparatus according to claim 1 or 2, characterized in that. The face detection unit detects a face area of the learning subject in a plurality of learning images each including a learning subject;
The feature amount generation unit generates a feature amount indicating a feature of the face of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images.
The facial expression detection unit detects the degree of facial expression of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images;
A classifying unit that classifies the plurality of learning images into a plurality of groups according to the degree of facial expression of the learning subject detected by the facial expression detection unit;
For each of the plurality of groups, a learning unit that learns the identification criterion based on the relationship between the age of the learning subject and the feature amount in the learning images classified into the respective groups;
Further comprising
The age estimation apparatus according to claim 3. The age estimating unit
A first age identification unit for identifying the age of the subject from the feature amount based on a first identification criterion;
A second age identifying unit for identifying the age of the subject from the feature amount based on a second identification criterion;
An age calculating unit that calculates an age between the age identified by the first age identifying unit and the age identified by the second age identifying unit based on the degree of facial expression detected by the facial expression detecting unit When,
Including
Estimating the age calculated by the age calculation unit as the age of the subject;
The age estimation apparatus according to claim 1 or 2, characterized in that. The face detection unit detects a face area of the learning subject in a plurality of learning images each including a learning subject;
The feature amount generation unit generates a feature amount indicating a feature of the face of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images.
The facial expression detection unit detects the degree of facial expression of the learning subject from the face area detected by the face detection unit for each of the plurality of learning images;
Among the plurality of learning images, learning images in which the degree of facial expression of the learning subject detected by the facial expression detection unit exceeds a predetermined threshold are classified into a first group, and the facial expression detection unit detects the learning image. A classifying unit for classifying learning images in which the degree of expression of the learning subject is equal to or less than the predetermined threshold value into a second group;
Based on the relationship between the age of the learning subject and the feature amount in the learning image classified into the first group, the first identification criterion is learned, and the learning image classified into the second group A learning unit that learns the second identification criterion based on the relationship between the age of the learning subject and the feature amount in the image;
Further comprising
The age calculation unit calculates an age closer to the age identified by the first age identification unit as the estimated age of the subject as the degree of expression of the subject detected by the facial expression detection unit increases.
The age estimation apparatus according to claim 5. The facial expression detection unit detects the degree of smile of the subject as the degree of facial expression;
The age estimation apparatus according to any one of claims 1 to 6, wherein The age estimation unit estimates a lower age as the age of the subject as the degree of the facial expression detected by the facial expression detection unit is larger,
The age estimation apparatus according to any one of claims 1 to 7, characterized in that: The age estimation device according to any one of claims 1 to 8,
An imaging unit that acquires the subject image by imaging the subject;
An output unit for outputting an estimation result of the age of the subject by the age estimation unit;
An imaging apparatus comprising: Detecting a face area of the subject in a subject image including the subject;
Generating a feature amount indicating a feature of the face of the subject from the face region;
Detecting the degree of expression of the subject from the face area;
Estimating the age of the subject based on the feature amount and the degree of facial expression;
The age estimation method characterized by including. On the computer,
A function for detecting a face area of a subject in a subject image including the subject;
A function of generating a feature amount indicating a feature of the face of the subject from the face region;
A function of detecting the degree of expression of the subject from the face area;
A function of estimating the age of the subject based on the feature amount and the degree of the facial expression;
A program to realize

Images