JP5072102B2 - Age estimation method and age estimation device - Google Patents

Description

  The present invention relates to an age estimation method and an age estimation apparatus for estimating an age from facial features.

  As a conventional age estimation method, for example, in Patent Document 1, a plurality of feature amounts are extracted from one or a plurality of learning faces whose actual ages are known, and a plurality of feature amounts of the extracted learning faces and the actual values of the learning faces are extracted. It is disclosed that the correlation with age is obtained by multiple regression analysis (a kind of multivariate analysis). In Patent Document 1, a plurality of feature amounts are extracted from one or a plurality of learning faces whose apparent ages are known, and the correlation between the extracted feature amounts of each learning face and the apparent age of each learning face is shown. Obtaining by multiple regression analysis is also disclosed.

JP 2005-148880 A

However, although there is a rough correlation between the apparent age of the face and the actual age, as shown in the following (1) to (3), there are many contradictions as learning data, and it is difficult to estimate the age stably. Met.
(1) Difference between actual age and apparent age (old face, baby face, etc.)
(2) Individual differences and variations in apparent age (3) Differences in age-related trends by gender (growth shift, makeup, anti-aging, etc.)

  The present invention has been made in view of such circumstances, and an object thereof is to provide an age estimation method and an age estimation apparatus capable of stably estimating the age.

  The age estimation method of the present invention is extracted in the feature amount extraction step using a feature amount extraction step of extracting a feature amount from an input face image and a plurality of determinators based on different predetermined ages. From each of the plurality of determinators based on the determined feature amount, a determination step for determining whether the age of the person in the face image is equal to or greater than a predetermined age, and the plurality of determinators in the determination step And a determination result integration step of estimating the age of the person of the face image by integrating the determination results.

  According to this method, attention is paid to the fact that the facial feature amount and age are not completely correlated (there are many exceptions), and each determinator is classified using a plurality of determinators that are largely classified into a predetermined age or older. Since the age is estimated by integrating the judgment results of the above, around the prescribed age, “judgment of difference between real age and apparent age”, “individual difference of apparent age / variation”, “difference in age-related trends”, and misjudgment Although there is a high possibility of this, it is possible to obtain a high determination performance for a face away from a predetermined age, and it is possible to perform age estimation that is more stable than conventional age estimation methods.

  By integrating the determination results from the plurality of determinators in the determination step as majority addition or likelihood addition, erroneous determination can be suppressed to a low level.

  Further, each of the plurality of determinators includes a face image learning step in which learning of a face image is performed in advance according to whether or not the age of the person of the face image is equal to or greater than the predetermined age. Learning is possible with learning data, and learning becomes easy.

  The age estimation apparatus of the present invention includes a feature amount extraction unit that extracts a feature amount from an input face image, and each of the age of the person of the face image based on the feature amount extracted by the feature amount extraction unit. Determination means for determining whether or not is a predetermined age or more, and determination result integration means for estimating the age of the person in the face image by integrating the determination results from the plurality of determination means .

  According to this configuration, since the age estimation is performed by integrating the determination results of each determinator using a plurality of determinators that are largely classified to be older than or less than a predetermined age, the “real age and apparent age” are around the predetermined age. ”Difference”, “Individual differences in apparent age” and “Aging tendency by gender” are likely to be misjudged. Therefore, it is possible to estimate the age more stably than the conventional age estimation method. In addition, by using a classifier that roughly classifies, learning is possible with a small number of learning data, and learning becomes easy.

  According to the present invention, it is possible to estimate the age more stably than the conventional age estimation method. In addition, by using a classifier that roughly classifies, learning is possible with a small number of learning data, and learning becomes easy.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments for carrying out the invention will be described in detail with reference to the drawings.

  FIG. 1 is a block diagram showing a schematic configuration of an age estimation apparatus according to an embodiment of the present invention. In FIG. 1, the age estimation apparatus 1 according to the present embodiment includes a feature amount extraction unit 2 that extracts a feature amount (luminance, edge information, etc.) from an input face image, and a feature extracted by the feature amount extraction unit 2. Based on the amount, each of the six determinators 3-1 to 3-6 and the six determinators 3-1 to determine whether or not the age of the person in the face image is equal to or higher than a predetermined age. And a determination result integration unit 4 that estimates the age of the person of the face image by integrating the determination results from 3-6. The feature quantity extraction unit 2, the determinators 3-1 to 3-6, and the determination result integration unit 4 may each have a dedicated hardware configuration, or may have a configuration in which each function is programmed and executed by a microcomputer. .

  The face image input to the feature amount extraction unit 2 is an image captured by a camera (not shown) or an image captured by a camera and recorded in a recording device. The determinator 3-1 determines whether the age of the person in the face image is 10 years or older, and the determinator 3-2 determines whether the age of the person in the face image is 20 years or older. To do. Further, the determinator 3-3 determines whether the age of the person in the face image is 30 years or older, and the determinator 3-4 determines whether the age of the person in the face image is 40 years or older. Determine. Further, the determinator 3-5 determines whether the age of the person in the face image is 50 years or older, and the determinator 3-6 determines whether the age of the person in the face image is 60 years or older. Determine. The determination results of the respective determinators 3-1 to 3-6 are output as binary values of the above / less than or in likelihood.

  The characteristics of each of the determiners 3-1 to 3-6 can be correctly determined with high probability for an input that is far from a predetermined age, and the determination performance is high for an input near a predetermined age. It has the property of deteriorating. For example, the determinator 3-6 that determines the age of 60 can determine that the face image of a young person is less than 60 years old with a high probability, but may determine that the face image in the late 50s is 60 years or older. . When the determination result is represented by the likelihood, the likelihood increases with increasing distance from the predetermined age as shown in FIG. 2, and the likelihood decreases near the predetermined age.

  Further, as illustrated in FIG. 3, each of the determinators 3-1 to 3-6 performs learning of a face image in advance depending on whether or not the age of the person in the face image is equal to or higher than a predetermined age. For example, the 10-year-old determinator 3-1 learns by dividing into 10 years old or older and less than 10 years old. That is, only the youngest face image is learned as less than 10 years old. Moreover, the 60-year-old determination machine 3-6 learns by dividing into 60 years old or older and less than 60 years old. That is, only the oldest face image is learned as being over 60 years old. By using the classifiers 3-1 to 3-6 that roughly classify, learning is possible with a small number of learning data, and learning becomes easy.

  Returning to FIG. 1, the determination result integration unit 4 integrates the determination results from the respective determinators 3-1 to 3-6 and estimates the age of the person in the face image. The determination result integration unit 4 performs majority addition when the determination results of the respective determinators 3-1 to 3-6 take a binary value of above / less than, and performs likelihood addition when taking the likelihood. FIG. 4 shows the output of each determinator 3-1 to 3-6 when a 35-year-old face image is input and the output of each determinator 3-1 to 3-6 takes a binary value greater than or less than It is a figure which shows the output result (b) of a result (a) and the determination result integration part 4. FIG. As shown in the output result (a), each determination result of the determinators 3-1 to 3-3 for determining the age of 10 to 30 is “over”, and the determinator 3 for determining the age of 40 to 60 Each determination result of −4 to 3-6 is “less than”. Therefore, as shown in the output result (b), the output of the 10-year-old determinator 3-1 becomes “O” in all of “10's” to “60's or more”, and the 20-year-old determinator 3-2 The output is “◯” in all of “20's” to “60's and above”, and the output of the 30-year-old judging machine 3-3 is “○” in all of “30's” to “60's and above”. The output of the age-determining machine 3-4 is “O” in all of “under 10s” to “30s”, and the output of the age-determining machine 3-5 is from “under 10s” to “40s” All are “◯”, and the output of the 60-year-old determinator 3-6 is “◯” in all of “under 10s” to “50s”. As a result, in the majority decision “6”, the age of the input image is estimated to be 30s.

  On the other hand, FIG. 5 shows a case where a 35-year-old face image is input and the output results of the respective determinators 3-1 to 3-6 when the outputs of the respective determinators 3-1 to 3-6 are estimated ( It is a figure which shows the output result (b) of the determination result integration part 4 a). As shown in the output result (a), the determination result of the determinator 3-1 for determining the age of 10 is “5” for less than, “95” for the above, and the determination result of the determinator 3-2 for determining the age of 20 is Less than "10", more than "90", the determination result of the determiner 3-3 determining 30 years old is less than "40", more than "60", the determination of the determiner 3-4 determining 40 years old The result is “70”, less than “30”, and the determination unit 3-5 for determining 50 years old. The determination result is “90”, less than “10”, and the determination unit 3-6 for determining 60 years old. The determination result of “95” is less than “5” and “5” is more than that. Therefore, as shown in the output result (b), the output of the 10-year-old determinator 3-1 is “5” for teenagers, “95” for teenagers, “95” for 20s, and “95” for 30s. 95 ", 40s is" 95 ", 50s is" 95 ", 60s and above are" 95 ", and the output of the 20-year-old decision machine 3-2 is" 10 "for those under 10s, 10 ", 20" is "90", 30 is "90", 40 is "90", 50 is "90", 60 and above is "90" Is "40" for teenagers, "40" for teenagers, "40" for 20s, "60" for 30s, "60" for 40s, "60" for 50s, "60" for those over 60s The output of the 40-year-old determinator 3-4 is "70" for teenagers, "70" for teenagers, "70" for 20s, "70" for 30s, "3" for 40s ”,“ 30 ”in the 50s,“ 30 ”in the 60s and above, and the output of the decision machine 3-5 at the age of 50 is“ 90 ”for the teenagers and“ 90 ”for the 10s and“ 90 ”for the 20s. ”,“ 90 ”for the 30s,“ 90 ”for the 40s,“ 10 ”for the 50s,“ 10 ”for the 60s and above. "The teenagers are" 95 ", the 20s are" 95 ", the 30s are" 95 ", the 40s are" 95 ", the 50s are" 95 ", and the 60s and above are" 5 ". As a result, with the likelihood value “500”, the age of the input image is estimated to be in the thirties.

  As described above, according to the age estimation device 1 of the present embodiment, the feature amount extraction unit 2 that extracts the feature amount from the input face image, and the feature amount extracted by the feature amount extraction unit 2, respectively. However, each of the six determinators 3-1 to 3-6 and the six determinators 3-1 to 3-6 that determine whether or not the age of the person in the face image is equal to or greater than a predetermined age. Since the determination result integration unit 4 for estimating the age of the person in the face image is integrated by integrating the determination results, the “difference between the actual age and the apparent age” and “individual differences in the apparent age” Although there is a high possibility of misjudgment due to “fluctuation” and “difference in aging according to gender”, the judgment performance for faces away from a given age can be improved, and it is more stable than conventional age estimation methods. Age estimation is possible. In addition, by using the classifiers 3-1 to 3-6 that roughly classify, learning is possible with a small number of learning data, and learning becomes easy.

  In addition, in this Embodiment, with respect to the difference in the aged tendency by gender (growth gap, makeup, etc.), each of the determiners 3-1 to 3-6 is learned for each gender, and at the time of age estimation You may make it switch to the determination machine which matched sex using the sex determination process.

  Moreover, in this Embodiment, you may use a statistical apparent age for learning of each determination machine 3-1 to 3-6. For the statistical apparent age, the apparent ages of a plurality of people with respect to a specific face are aggregated. Faces with large variations in apparent age are removed from the learning data, and the rest is used as learning data. The apparent age averages the apparent ages of multiple people.

  In the present embodiment, six determinators 3-1 to 3-6 are used. However, the number of determinators is not limited, and may be smaller or larger. In addition, although the age difference between the judging machines is 10-year-old intervals, this interval is not limited, and may be, for example, 5-year-old intervals.

  The present invention has an effect that the age can be stably estimated, and can be applied to all devices for estimating the age.

The block diagram which shows schematic structure of the age estimation apparatus which concerns on one embodiment of this invention The figure which shows the characteristic at the time of giving a likelihood determination function in each determination machine of FIG. The figure for demonstrating an example of the learning method in each determination machine of FIG. When a 35-year-old face image is input to the age estimation apparatus of FIG. 1, the output result (a) of each determinator and the output of the determination result integrating unit when the output of each determinator takes a binary value that is greater or less than The figure which shows a result (b) When a 35-year-old face image is input to the age estimation apparatus in FIG. 1, the output result (a) of each determinator and the output result (b) of the determination result integrating unit when the output of each determinator takes the likelihood Figure showing

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Age estimation apparatus 2 Feature-value extraction part 3-1 to 3-6 Judgment machine 4 Judgment result integration part

Claims (5)

A feature amount extraction step of extracting a feature amount from the input face image;
Using a plurality of determinators based on different predetermined ages, each of the plurality of determinators has a predetermined age of a person in the face image based on the feature amount extracted in the feature amount extraction step. A determination step of determining whether or not the age of
A determination result integration step of estimating the age of the person of the face image by aggregating output values of determination results from the plurality of determination devices in the determination step;
An age estimation method comprising: The age estimation method according to claim 1 , wherein the aggregation of output values of determination results from the plurality of determinators in the determination step is majority addition. The age estimation method according to claim 1 , wherein the aggregation of output values of determination results from the plurality of determinators in the determination step is likelihood addition. The face image learning step in which each of the plurality of determinators includes a face image learning step in which learning of a face image is performed in advance according to whether or not the age of the person in the face image is equal to or greater than the predetermined age. 4. The age estimation method according to any one of 3. Feature quantity extraction means for extracting feature quantities from the input face image;
A determination unit that determines whether or not the age of the person of the face image is equal to or higher than a predetermined age based on the feature amount extracted by the feature amount extraction unit;
A determination result integration unit that estimates the age of the person of the face image by aggregating output values of determination results from the plurality of determination units;
Age estimation device with

Images