JP2011070471A - Objective variable calculation device, objective variable calculation method, program, and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an objective variable calculation device capable of highly precisely calculating the objective variables of evaluation data acquired in an evaluation environment. <P>SOLUTION: An objective variable calculation device is provided with: an evaluation data acquisition means 111; and an objective variable calculation means 121. A preliminarily obtained database 140 is configured to store teacher data 141 with a label created by labeling a prescribed explanation variable x with an objective variable y, test data 142 acquired in an evaluation environment, and a model 143 with significance weighting created by weighting the teacher data 141 with a label and the test data 142 by using a significance function ω(x) as the rate of the probability density function of the teacher data 141 with a label to the probability density function of the test data. Evaluation data are acquired in the evaluation environment by the evaluation data acquisition means 111. The objective variable y of the evaluation data is calculated by using the model 143 with the significance weighting by the objective variable calculation means 121. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an objective variable calculation device, an objective variable calculation method, a program, and a recording medium.

  The supervised learning method for estimating the rules lurking behind the data is used in various fields such as face recognition, robot control, and gene data analysis. In a standard supervised learning method such as a support vector machine, it is assumed that the teacher data used for training the learning machine and the test data used for testing the learning machine follow the same generation rule. However, this assumption often does not hold in actual problems in recent years, and in such a situation, a good learning result is not always obtained by a normal supervised learning method.

  For example, when estimating the age from the face image, it is possible to estimate the age with high accuracy under conditions restricted to some extent such as in a laboratory (for example, see Patent Document 1). However, for example, when estimating the age from facial images for the purpose of analyzing customer segments in stores, it is difficult to obtain the same accuracy as a laboratory due to differences in camera type, performance and installation position, lighting conditions, etc. is there.

JP 2009-93490 A

  Therefore, an object of the present invention is to provide an objective variable calculation device, an objective variable calculation method, a program, and a recording medium that can calculate an objective variable of evaluation data acquired in an evaluation environment with high accuracy.

In order to achieve the above object, the objective variable calculation device of the present invention has an evaluation data acquisition means and an objective variable calculation means,
In the database acquired in advance,
Labeled teacher data in which a predetermined explanatory variable x is labeled with an objective variable y, and
Test data acquired in the evaluation environment,
Importance function ω which is a ratio (p _te (x) / p _tr (x)) between the probability density function p _tr (x) of the labeled teacher data and the probability density function p _te (x) of the test data (X) is used to store the weighted model created by weighting the labeled teacher data and the test data,
Evaluation data is acquired in the evaluation environment by the evaluation data acquisition means,
The objective variable calculation means calculates the objective variable y of the evaluation data using the importance weighted model.

The objective variable calculation method of the present invention is:
In an evaluation environment, an evaluation data acquisition process for acquiring evaluation data;
An objective variable calculating step of calculating an objective variable y of the evaluation data using an importance weighted model,
The importance-weighted model includes a probability density function p _tr (x) of labeled teacher data in which a predetermined explanatory variable x is labeled with an objective variable y, and a probability density function p of test data acquired in the evaluation environment. using _te (x) the ratio of the _{(p te (x) / p} tr (x)) is a importance function ω (x), is obtained in advance by weighting the labeled training data and the test data It is characterized by that.

  A program according to the present invention causes a computer to execute the object variable calculation method according to the present invention.

  The recording medium of the present invention records the program of the present invention.

  ADVANTAGE OF THE INVENTION According to this invention, the objective variable calculation apparatus, the objective variable calculation method, program, and recording medium which can calculate the objective variable of the evaluation data acquired in evaluation environment with high precision can be provided.

FIG. 1A is a flowchart showing an example of the objective variable calculation method of the present invention, and FIG. 1B is a block diagram showing an example of the objective variable calculation apparatus of the present invention. FIG. 2A is a flowchart showing another example of the objective variable calculation method of the present invention, and FIG. 2B is a block diagram showing another example of the objective variable calculation apparatus of the present invention. FIG. 3 is a block diagram showing a configuration of an example of an objective variable calculation system using the objective variable calculation apparatus of the present invention. FIG. 4 is a graph showing a calculation result of the estimated age based on the face image in one embodiment of the present invention. FIG. 5 is a graph showing a calculation result of an estimated age based on a face image according to another embodiment of the present invention. FIG. 6 is a graph showing a calculation result of an estimated age based on a face image in still another embodiment of the present invention. FIG. 7 is a graph showing a calculation result of an estimated age based on a face image in still another embodiment of the present invention. FIG. 8 is a graph showing a calculation result of an estimated age based on a face image in still another embodiment of the present invention. FIG. 9 is a graph showing a calculation result of an estimated age based on a face image in still another embodiment of the present invention.

  Next, an objective variable calculation device, an objective variable calculation method, a program, and a storage medium according to the present invention will be described with examples. However, the present invention is not limited to the following examples.

(Embodiment 1)
FIG. 1 shows a flowchart of the objective variable calculation method in the present embodiment (FIG. 1A) and a block diagram of the objective variable calculation apparatus (FIG. 1B). As shown in FIG. 1B, the objective variable calculation apparatus of this embodiment includes a database 140, an evaluation data acquisition unit 111, an objective variable calculation unit 121, and an output unit 131 that are acquired in advance as main components. In the database 140, a predetermined explanatory variable x is labeled with the objective variable y, labeled teacher data 141, test data 142 acquired in the evaluation environment, and the probability density function p _tr (of the labeled teacher data 141). x) and a probability density function p _te (x) of the test data 142 using the importance function ω (x) which is a ratio (p _te (x) / p _tr (x)), the labeled teacher An importance weighted model 143 created by weighting the data 141 and the test data 142 is stored. The objective variable calculation means 121 is connected to the importance weighted model 143. Examples of the evaluation data acquiring unit 111 include a CCD (Charge Coupled Device) camera, a CMOS (Complementary Metal Oxide Semiconductor) camera, and an image scanner. Examples of the objective variable calculation means 121 include calculation means such as a central processing unit (CPU). Examples of the output means 131 include a monitor that outputs video (for example, various image display devices such as a liquid crystal display (LCD) and a cathode ray tube (CRT) display), a printer that outputs by printing, a speaker that outputs by sound, and the like. can give. The output means 131 is an arbitrary component and may not be included in the objective variable calculation apparatus of the present invention, but is preferably included. Examples of the database 140 include data stored in data storage means such as a random access memory (RAM), a read-only memory (ROM), a hard disk (HD), an optical disk, and a floppy (registered trademark) disk (FD). can give. The data storage unit 140 may be a device built-in type or an external type such as an external storage device.

  In the present invention, the evaluation data is not particularly limited, and examples thereof include a face image.

  In the present invention, the objective variable y is not particularly limited as long as it is a variable related to the evaluation data, and may be any variable. For example, when the evaluation data is a face image, examples of the objective variable y include estimated age, sex, race, and the like.

  Hereinafter, the objective variable calculation device and the objective variable calculation method of this embodiment will be described in more detail by taking as an example a case where the evaluation data is a face image and the objective variable y is an estimated age.

  The objective variable calculation method of this example is implemented as follows using the objective variable calculation apparatus of FIG.

First, the database 140 is acquired prior to the execution of the objective variable calculation method. Specifically, first, the labeled teacher data 141 obtained by labeling a predetermined explanatory variable x with the objective variable y is acquired. In this example, examples of the predetermined explanatory variable x include variables such as a luminance value of each pixel of the image, a skin color, and the number of wrinkles. Considering the calculation accuracy, it is preferable that the number of the labeled teacher data 141 is larger. Next, test data 142 is acquired in the evaluation environment. Next, the ratio (p _te (x) / p _tr (x)) between the probability density function p _tr (x) of the labeled teacher data 141 and the probability density function p _te (x) of the test data 142 An importance weighted model 143 created by weighting the labeled teacher data 141 and the test data 142 is obtained using a certain importance function ω (x). As will be described later, in the objective variable calculation apparatus and objective variable calculation method of this embodiment, the objective variable y (estimated age) of the evaluation data (face image) is calculated using the importance weighted model. For this reason, according to the objective variable calculation apparatus and objective variable calculation method of this embodiment, even when the number of test data 142 is smaller than the number of labeled teacher data 141, the evaluation data ( The objective variable y (estimated age) of (face image) can be calculated.

The importance-weighted model 143 is preferably created using a least-squares method in which the weighting is increased as the importance function ω (x) has a larger value (higher importance). The importance weighted model 143 can be created, for example, by solving an optimization problem expressed by the following equation (1).

The importance function ω (x) can be obtained by, for example, modeling by a KLIEP (Kullback-Leibler Importance Procedure Procedure) algorithm. The KLIEP algorithm uses the probability density function p _te (x) of the actual test data 142 and the probability density function of the test data estimated using the modeled importance function (double underlined in the following equation (2)). It is characterized in that a model of the importance function is obtained so as to minimize the difference from (double underlined part of the following formula (3)).

First, the importance function ω (x) is modeled by a linear combination of the following kernel functions k _γ .

At this time, the probability density function of the test data is estimated by the following equation (3) using the modeled importance function (double underlined portion of the equation (2)).

The degree of difference between the two probability density functions represented by p _te (x) and the double underlined part of the above equation (3) is the amount of Kullback-Leibler information developed by the following equation (4) (the following equation (4) ) In the double underlined part (solid line)).

に注意すれば、上に凸な最大化問題として定式化することができる。

ここで、前記式（５）の拘束条件は、重要度関数が非負であること（ω（ｘ）≧０（ｆｏｒ ａｌｌ ｘ ∈ Ｄ、Ｄはデータの分布域）を保証するためのものである。また、前記式（６）の拘束条件は、前記式（３）の二重下線部の全積分値が１、すなわち、確率密度関数の特徴の１つを保証するためのものであり、下記式（７）の展開に従う。

As can be seen from the equation (4), in order to obtain the double underline portion of the equation (3) as a good approximation of p _te (x), the double underline portion (dashed line) of the equation (4) is obtained. Maximize. here,

Can be formulated as an upwardly convex maximization problem.

Here, the constraint condition of the equation (5) is to guarantee that the importance function is non-negative (ω (x) ≧ 0 (for all x ∈ D, D is a data distribution range)). The constraint condition of the equation (6) is to guarantee that the total integral value of the double underlined portion of the equation (3) is 1, that is, one of the features of the probability density function. Follow the expansion of equation (7).

Next, taking the case where l pieces of labeled training data is obtained in advance as an example, y ^* a y ^{* =} f (x) KRLS estimating the Regression analysis in explaining.

In this example, y ^* = f (x) is modeled by a linear combination of positive definite kernels (8) below.

  When the number l of the teacher data with labels is large, a subset thereof may be used instead of all the teacher data with labels.

Next, the following equation (10) of the model (9) is learned so that the following equation (11) is minimized.

At this time, the following equation (12) which is the optimal solution of the equation (11) is given by the following equation (13). By such a method, the importance weighted model can be created. However, the method of creating the importance weighted model is merely an example. In the present invention, the method for creating the importance weighted model is not particularly limited, and any method may be used.

  Next, as shown in FIGS. 1A and 1B, the evaluation data acquisition unit 111 acquires evaluation data (face image) in the evaluation environment (step S11).

  Next, as shown in FIGS. 1A and 1B, the objective variable calculation unit 121 uses the importance weighted model 143 and uses the objective variable y (estimated age) of the evaluation data (face image). ) Is calculated (step S21).

  Next, as shown in FIGS. 1A and 1B, the output means 131 outputs the objective variable y (estimated age) of the evaluation data (face image) (step S31). The output step S31 is an optional step and may not be included in the objective variable calculation method of the present invention, but is preferably included.

  As described above, in the objective variable calculation apparatus and objective variable calculation method of the present embodiment, prior to the implementation, the importance weighted model is acquired in advance, and the evaluation data (the objective variable of the face image is used). Therefore, according to the objective variable calculation device and the objective variable calculation method of this embodiment, the evaluation data (face image) can be obtained with high accuracy regardless of the evaluation environment. Objective variable y (estimated age) can be calculated.

(Embodiment 2)
FIG. 2 is a flowchart (example (A)) and block diagram (example) in which the importance weighted model is acquired in advance for each evaluation environment when the evaluation environment changes. (B)). In this figure, the same parts as those in FIG. As shown in FIG. 5B, in the objective variable calculation apparatus of the present embodiment, the test data 142 and the importance-weighted model 143 are acquired in advance, three each in front, upward, and downward. Except for this point, the configuration is the same as that of the objective variable calculation apparatus shown in FIG.

  Hereinafter, as in the first embodiment, the objective variable calculation device and the objective variable calculation method of the present embodiment will be described by taking as an example the case where the evaluation data is a face image and the objective variable y is an estimated age.

  The objective variable calculation method of this example is implemented as follows using the objective variable calculation apparatus of FIG.

First, the database 140 is acquired prior to the execution of the objective variable calculation method. Specifically, first, the labeled teacher data 141 is acquired in the same manner as in the first embodiment. Next, an evaluation environment in which a face image facing the front is acquired (front evaluation environment), an evaluation environment in which an upward face image is acquired (upward evaluation environment), and an evaluation environment in which a downward face image is acquired (downward evaluation) The three test data 142 are acquired in each of the environments. Next, a ratio (p _te (x) / p _tr (p) of the probability density function p _tr (x) of the labeled teacher data 141 and the probability density function p _te (x) of each of the three test data 142. Using the three importance functions ω (x) that are x)), three importance weighted models 143 created by weighting each of the labeled teacher data 141 and the three test data 142 are obtained. To do.

  Next, as shown in FIGS. 2A and 2B, in the same manner as in the first embodiment, the evaluation data acquisition unit 111 performs evaluation data (face image) in any one of the three evaluation environments. ) Is acquired (step S11).

  Next, as shown in FIGS. 2A and 2B, after estimating the angle (orientation) of the evaluation data (face image), the objective variable calculation unit 121 performs the same as in the first embodiment. Of the three importance-weighted models 143, an objective variable y (estimated age) of the evaluation data (face image) is used using an importance-weighted model corresponding to the evaluation environment from which the evaluation data (face image) was acquired. ) Is calculated (step S21).

  Next, as shown in FIGS. 2A and 2B, the output means 131 outputs the objective variable y (estimated age) of the evaluation data (face image) (step S31). The output step S31 is an optional step and may not be included in the objective variable calculation method of the present invention, but is preferably included.

  As described above, in the objective variable calculation device and the objective variable calculation method of the present embodiment, the importance weighted model is acquired in advance for each evaluation environment. Therefore, according to the objective variable calculation device and objective variable calculation method of the present embodiment, even when the evaluation environment changes suddenly, such as when evaluation data (face images) with different angles (orientations) are acquired. The objective variable y (estimated age) of the evaluation data (face image) can be calculated with high accuracy.

(Embodiment 3)
The program of this embodiment is a program that can execute the above-described objective variable calculation method on a computer. Or the program of this embodiment may be recorded on a recording medium, for example. The recording medium is not particularly limited, and examples thereof include a read only memory (ROM), a hard disk (HD), an optical disk, a floppy (registered trademark) disk (FD), and the like.

(Embodiment 4)
FIG. 3 shows an exemplary configuration of an objective variable calculation system using the objective variable calculation apparatus of the present invention. As illustrated, the objective variable calculation system includes evaluation data acquisition units 111a, 111b, and 111c, output units 131a, 131b, and 131c, communication interfaces 150a, 150b, and 150c, and a server 170. The evaluation data acquisition unit 111a and the output unit 131a are connected to a communication interface 150a. The evaluation data acquisition unit 111a, the output unit 131a, and the communication interface 150a are installed in the place X. The evaluation data acquisition unit 111b and the output unit 131b are connected to the communication interface 150b. The evaluation data acquisition unit 111b, the output unit 131b, and the communication interface 150b are installed at the place Y. The evaluation data acquisition unit 111c and the output unit 131c are connected to a communication interface 150c. The evaluation data acquisition unit 111c, the output unit 131c, and the communication interface 150c are installed in the place Z. The communication interfaces 150a, 150b, and 150c are connected to the server 170 via the line network 160.

  This objective variable calculation system has objective variable calculation means on the server 170 side, and a database is stored in the server 170. For example, the evaluation data acquired using the evaluation data acquisition unit 111a at the place X is transmitted to the server 170, and the objective variable of the evaluation data is calculated on the server 170 side. Further, the calculated objective variable of the evaluation data is output by the output means 131a.

  According to the objective variable calculation system of the present embodiment, the evaluation data acquisition means and the output means are installed at the site, and the server or the like is installed at another location, so that the objective variable of the evaluation data can be calculated online. Therefore, the installation of the apparatus does not take a place, and maintenance is easy. Further, even when the installation locations are separated, centralized management and remote operation can be performed at one location. The objective variable calculation system of this embodiment may correspond to the above-described second embodiment.

  First, the standard lighting environment (standard lighting environment), the overall dark lighting environment (dark lighting environment), the lighting environment that is exposed to strong light from the side of the head (strong lighting environment from the side), Video of the upper body of a person who is seated and facing in front (camera direction) in the four types of evaluation environment, where the light is shining from directly above (strong lighting environment from above), and the face from the camera is about 15 degrees I took a picture at an angle looking down. Next, the captured moving image file is read into the head detection program, the positions of both eyes, nose and mouth are estimated, and the face image normalized to 64 × 64 pixel size based on the position information of both eyes Extracted. The number of images per evaluation environment was about 10,000, and the number of persons was 514 (male: female = 1: 1).

Next, feature extraction of the face image (64 × 64 pixels) was performed using neutral network feature extraction. The data after feature extraction is 100-dimensional, and regression analysis was performed using KRLS. As the kernel function, a Gaussian kernel of the following formula (14) was used.

  For the importance weight parameter γ, KRLS parameters σ (kernel width), and λ (regularization term parameter), the optimal model is obtained by a 6-fold (weighted) cross-confirmation method while changing the values to grid formula It was determined. The ratio of the number of labeled teacher data and the number of test data at this time was 5: 1 for both the number of images and the number of persons.

  The graph of FIG. 4 shows the result of the woman when the evaluation environment is the dark lighting environment. Reference Example 1-1 is a bar graph showing a WMSE (Weighted Mean Square Error) when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. It is. Reference Example 1-2 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the dark lighting environment is evaluated with evaluation data acquired in the dark lighting environment. From comparison between the reference example 1-1 and the reference example 1-2, it was found that it is more difficult to create an age estimation model in a dark illumination environment than in a standard illumination environment. Comparative Example 1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard illumination environment is evaluated with evaluation data acquired in the dark illumination environment. From Comparative Example 1, it was found that when the age estimation model obtained using the labeled teacher data in the standard illumination environment is used for the evaluation data acquired in the dark illumination environment, the accuracy of the age estimation is greatly reduced. Example 1 uses the importance function that is a ratio of the probability density function of the labeled teacher data in the standard illumination environment and the probability density function of the test data acquired in the dark illumination environment, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the said dark illumination environment is calculated using the weighting model created by weighting the test data. In Example 1 in which the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 1.

  The graph of FIG. 5 shows the male result when the evaluation environment is the dark illumination environment. Reference Example 2-1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. Reference Example 2-2 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the dark lighting environment is evaluated with evaluation data acquired in the dark lighting environment. From comparison between the reference example 2-1 and the reference example 2-2, it was found that it is more difficult to create an age estimation model in a dark illumination environment than in a standard illumination environment. Comparative Example 2 is a bar graph showing WMSE when an age estimation model obtained using the labeled teacher data in the standard lighting environment is evaluated with the evaluation data acquired in the dark lighting environment. From Comparative Example 2, it was found that when the age estimation model obtained using the labeled teacher data in the standard illumination environment is used for the evaluation data acquired in the dark illumination environment, the accuracy of the age estimation is greatly reduced. Example 2 uses the importance function that is the ratio of the probability density function of the labeled teacher data in the standard illumination environment and the probability density function of the test data acquired in the dark illumination environment, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the said dark illumination environment is calculated using the weighting model created by weighting the test data. In Example 2 where the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 2.

  The graph of FIG. 6 shows the result of the woman when the evaluation environment is a strong lighting environment from the side. Reference Example 3-1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. Reference Example 3-2 is a bar graph showing WMSE when the age estimation model obtained using the labeled teacher data in the strong lighting environment from the side is evaluated with the evaluation data acquired in the strong lighting environment from the side It is. Comparison between the reference example 3-1 and the reference example 3-2 shows that it is more difficult to create an age estimation model in a strong lighting environment from the side than in a standard lighting environment. Comparative Example 3 is a bar graph showing the WMSE when the age estimation model obtained using the labeled teacher data in the standard lighting environment is evaluated with the evaluation data acquired in the strong lighting environment from the side. From Comparative Example 3, it can be seen that if the age estimation model obtained using the labeled teacher data in the standard lighting environment is used for the evaluation data obtained in the strong lighting environment from the side, the accuracy of the age estimation is greatly reduced. It was. Example 3 uses the importance function that is the ratio of the probability density function of the labeled teacher data in the standard lighting environment and the probability density function of the test data acquired in the strong lighting environment from the side, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the strong illumination environment from the side is calculated using the weighted model created by weighting the attached teacher data and the test data. In Example 3 in which the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 3.

  The graph of FIG. 7 shows the result of the male when the evaluation environment is a strong lighting environment from the side. Reference Example 4-1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. Reference Example 4-2 is a bar graph showing WMSE when the age estimation model obtained using the labeled teacher data in the strong lighting environment from the side is evaluated with the evaluation data acquired in the strong lighting environment from the side. It is. Comparison between the reference example 4-1 and the reference example 4-2 reveals that it is more difficult to create an age estimation model in a strong lighting environment from the side than in a standard lighting environment. Comparative Example 4 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the strong lighting environment from the side. From Comparative Example 4, it can be seen that if the age estimation model obtained using the labeled teacher data in the standard lighting environment is used for the evaluation data acquired in the strong lighting environment from the side, the accuracy of the age estimation is greatly reduced. It was. Example 4 uses the importance function that is a ratio of the probability density function of the labeled teacher data in the standard lighting environment and the probability density function of the test data acquired in the strong lighting environment from the side, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the strong illumination environment from the side is calculated using the weighted model created by weighting the attached teacher data and the test data. In Example 4 where the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 4.

  The graph of FIG. 8 shows the result of the woman when the evaluation environment is a strong lighting environment from above. Reference Example 5-1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. Reference Example 5-2 is a bar graph showing WMSE when the age estimation model obtained using the labeled teacher data in the strong illumination environment from above is evaluated with the evaluation data acquired in the strong illumination environment from above It is. Comparison between the reference example 5-1 and the reference example 5-2 indicates that it is more difficult to create an age estimation model in the strong illumination environment from above than in the standard illumination environment. Comparative Example 5 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the strong lighting environment from above. From Comparative Example 5, it can be seen that if the age estimation model obtained using the labeled teacher data in the standard lighting environment is used for the evaluation data acquired in the strong lighting environment from the side, the accuracy of the age estimation is greatly reduced. It was. Example 5 uses the importance function which is a ratio of the probability density function of the labeled teacher data in the standard lighting environment and the probability density function of the test data acquired in the strong lighting environment from above, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the strong illumination environment from above is calculated using the weighted model created by weighting the attached teacher data and the test data. In Example 5 in which the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 5.

  The graph of FIG. 9 shows the male result when the evaluation environment is a strong lighting environment from above. Reference Example 6-1 is a bar graph showing WMSE when an age estimation model obtained using labeled teacher data in the standard lighting environment is evaluated with evaluation data acquired in the standard lighting environment. Reference Example 6-2 is a bar graph showing WMSE when the age estimation model obtained using the labeled teacher data in the strong illumination environment from above is evaluated with the evaluation data acquired in the strong illumination environment from above It is. Comparison between the reference example 6-1 and the reference example 6-2 revealed that it is more difficult to create an age estimation model in a strong illumination environment from above than in a standard illumination environment. Comparative Example 6 is a bar graph showing WMSE when an age estimation model obtained using the labeled teacher data in the standard lighting environment is evaluated with the evaluation data acquired in the strong lighting environment from above. From Comparative Example 6, it can be seen that if the age estimation model obtained using the labeled teacher data in the standard lighting environment is used for the evaluation data acquired in the strong lighting environment from above, the accuracy of the age estimation is greatly reduced. It was. Example 6 uses the importance function, which is the ratio of the probability density function of the labeled teacher data in the standard lighting environment and the probability density function of the test data acquired in the strong lighting environment from above, It is a bar graph which shows WMSE when the estimated age of the evaluation data acquired in the strong illumination environment from above is calculated using the weighted model created by weighting the attached teacher data and the test data. In Example 6 where the estimated age of the evaluation data was calculated using the importance weighted model, the calculation accuracy was improved as compared with Comparative Example 6.

  As described above, according to the first to sixth embodiments, the estimated age can be calculated with high accuracy using the face image acquired in the evaluation environment different from the standard illumination environment from which the labeled teacher data is acquired. did it.

111 Evaluation Data Acquisition Unit 121 Objective Variable Calculation Unit 131 Output Unit 140 Database 141 Labeled Teacher Data 142 Test Data 143 Importance Weighted Model

Claims (10)

Having evaluation data acquisition means and objective variable calculation means,
In the database acquired in advance,
Labeled teacher data in which a predetermined explanatory variable x is labeled with an objective variable y, and
Test data acquired in the evaluation environment,
Importance function ω which is a ratio (p _te (x) / p _tr (x)) between the probability density function p _tr (x) of the labeled teacher data and the probability density function p _te (x) of the test data (X) is used to store the weighted model created by weighting the labeled teacher data and the test data,
Evaluation data is acquired in the evaluation environment by the evaluation data acquisition means,
An objective variable calculating apparatus, wherein the objective variable calculating means calculates an objective variable y of the evaluation data using the importance weighted model. In the case where the evaluation environment changes,
The importance weighted model is acquired in advance for each of the evaluation environments,
The objective variable y is calculated using the importance weighted model corresponding to the evaluation environment from which the evaluation data is acquired among the importance weighted models acquired in advance for each evaluation environment. The objective variable calculation apparatus according to claim 1, wherein: The objective variable calculation apparatus according to claim 1, wherein the importance-weighted model is created using a least square method. 4. The objective variable calculation device according to claim 1, wherein the evaluation data, the labeled teacher data, and the test data are facial images, and the objective variable y is an estimated age. 5. . In an evaluation environment, an evaluation data acquisition process for acquiring evaluation data;
An objective variable calculating step of calculating an objective variable y of the evaluation data using an importance weighted model,
The importance-weighted model includes a probability density function p _tr (x) of labeled teacher data in which a predetermined explanatory variable x is labeled with an objective variable y, and a probability density function p of test data acquired in the evaluation environment. using _te (x) the ratio of the _{(p te (x) / p} tr (x)) is a importance function ω (x), is obtained in advance by weighting the labeled training data and the test data Objective variable calculation method characterized by the above. In the case where the evaluation environment changes,
The importance weighted model is acquired in advance for each of the evaluation environments,
The objective variable y is calculated using the importance weighted model corresponding to the evaluation environment from which the evaluation data is acquired among the importance weighted models acquired in advance for each evaluation environment. The objective variable calculation method according to claim 5, wherein: 7. The objective variable calculation method according to claim 5, wherein the importance weighted model is created using a least square method. The objective variable calculation method according to claim 5, wherein the evaluation data, the labeled teacher data, and the test data are facial images, and the objective variable y is an estimated age. . A program for causing a computer to execute the objective variable calculation method according to any one of claims 5 to 8. A recording medium on which the program according to claim 9 is recorded.

Images