JP2010152866A - Sex-age identification method and device based on sound and image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sex-age identification method and device based on sound and image. <P>SOLUTION: The present invention relates to an identification device and method capable of accurately computing sex and age by performing sound recognition and face recognition in combination, considering a mutual relevancy between sex information and age information. The sex-age identification method includes: a step of collecting image information and sound information; a sound information-used sex and age identification step of extracting at least one characteristic value for the collected sound information and identifying sex and age by use of the extracted characteristic value; a face information-used sex and age identification step of extracting at least one characteristic value for the collected image information and identifying sex and age using the extracted characteristic value; and a step of finally determining sex and age by performing combination operation of the sex and age identified using the sound information and the sex and age identified using the face information. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a method and apparatus for identifying gender and age of a specific person from input video information and audio information, and more particularly, speech recognition in consideration of the reciprocal relationship between gender information and age information. In addition, the present invention relates to an identification apparatus and method that can accurately calculate gender and age by combining face recognition.

  As a conventional technique for gender and age identification of a user, there are a method using personal identification means such as an electronic resident card, a method using face recognition, a method using voice recognition, and the like.

  One of the methods using personal identification means is an age recognition method using an electronic resident ID (Korea Published Patent No. 1999-0008679), in which each individual always carries an individual identification means such as an electronic resident ID. There is an inconvenience of having to. In addition, personal identification means such as electronic resident cards are prone to loss, damage, forgery and the like.

  In the case of a face recognition method used as a conventional gender-age identification technique, the gender and age are determined based only on the face image information, so that it is difficult to reflect the characteristics of each individual and the recognition accuracy is low. In addition, in the case of a recognition method using voice recognition, since gender and age are determined only by voice information, there is a problem that the recognition accuracy is inferior when voice characteristics are similar, such as a woman and a child. is there.

  In addition, conventional identification methods based on face recognition or voice recognition cannot identify age and gender, reflecting the specificity that the distribution of features differs by gender, or the specificity that the distribution of gender features varies by age. Therefore, the calculation accuracy is low and the calculation amount is large.

Korean Published Patent No. 1999-0008679

  The present invention has been made in view of the above-mentioned problems, and its purpose is to improve the accuracy of recognition by utilizing the correlation between gender information and age information and combining voice recognition and face recognition. An object is to provide a gender-age identification method and apparatus capable of performing the same.

  In order to achieve the above object, a gender-age identification method according to an aspect of the present invention includes a step of collecting video information and audio information, and extracting one or more feature values from the collected audio information, Gender and age identification step using audio information for identifying gender and age using the extracted feature value, and one or more feature values are extracted for the collected video information, and the extracted Gender and age identification step using face information for identifying gender and age using feature values; gender and age identified using voice information; and gender and age identified using face information. Performing a combinational operation to finally determine gender and age.

  According to another aspect of the present invention, there is provided a gender-age identification apparatus, an input unit that collects video information and audio information, and a feature value extracted from the collected audio information, and the extracted feature value is used to extract the feature value. An audio processing unit for identifying gender and age from audio information, and a video processing unit for extracting feature values from the collected video information and identifying gender and age from the video information using the extracted feature values And a final identification unit that finally determines the sex and age of the specific person by performing a calculation by combining the gender and age identified by the video processing unit and the gender and age identified by the audio processing unit. Including.

  According to the present invention, since voice recognition and face recognition are performed in combination, the recognition accuracy is improved as compared with the conventional method using only voice recognition or the method using only face recognition.

  In addition, the present invention reflects the interrelationship between sex information and age information, for example, age identification reflects the specificity of the distribution of characteristics depending on the sex, or the specificity of the distribution of characteristics of the sex depending on the age. Since the recognition is performed, there is an effect that a higher accuracy can be ensured as compared with the conventional recognition method.

  Furthermore, the present invention provides a method for grouping speech information primarily on the basis of features that can be easily distinguished for each input information with respect to information input in feature extraction, and for each group distinguished by the criteria. By using the method of extracting feature values reflecting the features, it is possible to ensure the accuracy of the identification, and it is possible to eliminate the duplication of operations and perform the quick identification.

It is a block diagram which shows one Embodiment of the sex-age identification apparatus which concerns on this invention. It is a detailed block diagram of the audio | voice processing part which concerns on FIG. FIG. 2 is a detailed configuration diagram of a video processing unit according to FIG. 1. 3 is a flowchart of a gender-age identification method according to the present invention. It is a detailed flowchart of the audio | voice similarity identification step of FIG. 5 is a detailed flowchart of a video similarity identification step in FIG. 4.

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

  FIG. 1 is a block diagram showing an embodiment of a gender-age identification apparatus according to the present invention.

  As shown in FIG. 1, the gender-age identification apparatus according to the present invention includes an input unit 10, an age-gender calculation unit 20, and an output unit 30.

  The input unit 10 collects video information and audio information of a specific person.

  Such an input unit 10 can be configured to include video information acquisition means such as a camera that can acquire video information and acoustic information acquisition means such as a speaker (microphone) that can acquire acoustic information.

  Further, the input unit 10 separately extracts only the voice information of the specific person from the acoustic information acquired by the face extraction means and the acoustic information acquisition means that separately extracts only the face information of the specific person from the video information acquired by the video information acquisition means. A voice extraction means that can be extracted can be included. In this case, each feature extraction unit of the age-gender calculation unit 20 does not need to separately extract face information and audio information from the video information and the sound information every time, so that quick calculation is possible.

  Such face extraction means and voice extraction means can be implemented using conventional face detection technology. For example, for face extraction, a knowledge-based method (Knowledge-baSed MethodS), a feature-based method (Feature-baSed MethodS), a template matching method (Template-matching MethodS), a method based on appearance (Appearance-baSedM) The face extraction means can be implemented using a thermal infrared (Infra Red) method, a three-dimensional face recognition method, a multimodal method, or the like.

  The age-gender computing unit 20 includes an audio processing unit 100 that identifies age and gender based on audio information, a video processing unit 200 that identifies age and gender based on video information, an audio processing unit 100, and a video processing unit. And a final discriminating unit 300 that determines the age and sex by combining the 200 calculation results.

  The output unit 30 outputs the age and sex transmitted from the age-sex calculating unit 20.

  Hereinafter, the age-sex calculator 20 will be described in detail with reference to FIGS. 2 and 3.

  FIG. 2 is a detailed configuration diagram of the audio processing unit 100 according to FIG.

  As shown in FIG. 2, the speech processing unit 100 includes a speech feature extraction unit 110 that extracts feature values from speech information and a speech calculation unit 120 that identifies gender and age from the extracted feature values.

  More specifically, the speech feature extraction unit 110 extracts one or more feature values or feature vectors (hereinafter referred to as “feature values”) from the speech information. The speech feature extraction unit 110 uses a linear predictive coefficient method, a cepstrum method, a mel frequency cepstrum coefficient method, a filter bank energy method such as a filter bank energy method. These can be combined to extract feature values.

  The voice feature extraction unit 110 extracts a plurality of feature values from the same voice information by applying a plurality of the above-described feature value identification methods, or uses a plurality of samples using a single feature value identification method. Individual feature values can be identified. When feature values are obtained for M speech samples by N feature identification methods, the feature values can be expressed in a matrix form of (N * M).

  In the embodiment of the present invention, in order to accurately and quickly extract features from speech, a gender feature extraction unit 111, an age-specific feature extraction unit-M112, an age-specific feature extraction unit-FC113, an age-specific feature extraction unit-F114, and A speech feature extraction unit 110 is configured including a gender feature extraction unit-C115.

  The gender feature extraction unit 111 extracts a difference between male and female with respect to the input voice information, that is, extracts a feature value reflecting the gender feature, and sets the voice information as a male group (M) based on the extracted feature value. Or categorize into women and children groups (FC).

  The age-specific feature extraction unit-M112 extracts feature values for the speech information classified into the male group (M) by the gender feature extraction unit 111. Since the voice information input in this case is voice information determined to be male voice information, a feature value can be extracted by reflecting male age characteristics.

  The age-specific feature extraction unit-FC 113 may extract feature values reflecting the age-specific features of women and children from the voice information classified into the women and children group (FC) by the gender feature extraction unit 111. it can. Thereafter, the input voice information is again divided into a female group (F) and a child group (C). Here, the child group (C) is a group for people before the voice change period, in which it is difficult to distinguish the characteristics of men and women.

  The age-specific feature extraction unit-F114 can extract the feature value by reflecting the age-specific feature of the woman in the voice information classified into the female group (F) by the age-specific feature extraction unit-FC113.

  The gender feature extraction unit-C115 extracts a feature value reflecting the gender feature of the child from the audio information classified into the child group (C) by the age-specific feature extraction unit-FC113.

  The voice calculation unit 120 can identify the gender and age of the input voice in response to the input of the feature value extracted by the voice feature extraction unit 110 as described above.

  For this purpose, the speech computation unit 120 includes a combination computation unit that determines a representative feature value by reflecting a weighted value for the feature value extracted from the speech feature extraction unit 110, and a gender based on the determined representative feature value. And an identification unit that identifies gender and age with reference to a standard DB storing age standard feature values or audio and video standard samples.

  Further, as shown in FIG. 2, the voice calculation unit 120 is a combination calculation optimized for each of the male group (M), the female group (F), and the child group (C) grouped by the voice feature extraction unit 110. It is preferable to comprise so that a part and an identification part may be provided respectively.

  Below, it demonstrates on the basis of embodiment shown in FIG. 2 with which the combination calculating part and the identification part were each provided in this way.

  The voice calculation unit 120 receives a feature value extracted from the voice information classified into the male group (M) by the voice feature extraction unit 110, calculates a gender and age, and a female group (M121). F) a voice calculation unit F122 that calculates the gender and age in response to the input of the feature value extracted from the voice information classified into F), and the feature value extracted from the voice information classified into the child group (C) The voice calculation unit -C123 that receives the input and calculates gender and age can be used.

  More specifically, the voice calculation unit-M121 includes a combination calculation unit-M121A and an identification unit-M121B. The combination calculation unit-M121A determines a representative feature value by assigning a weight value to one or more feature values extracted from the speech information classified into the male group (M). The identification unit-M121B can identify gender and age with reference to the standard DB based on the representative feature value. Further, since the union calculation unit-M121A receives the input of the voice information divided into male groups and performs the union calculation, the feature values extracted from the gender feature extraction unit 111 and the age-specific feature extraction unit-M112 as described above Can be input.

  Similarly, the voice calculation unit-F122 is a combination calculation unit-F122A that determines a representative feature value by assigning a weight to one or more feature values extracted from the voice information divided into the female group (F). , And an identification unit -F122B that identifies gender and age with reference to the standard DB based on the representative feature value. As described above, the combination calculation unit-F122A can receive the feature values extracted from the gender feature extraction unit 111, the age feature extraction unit-FC113, and the age feature extraction unit-F114.

  The voice calculation unit -C123 includes a combination calculation unit -C123A that determines a representative feature value by assigning a weight value to one or more feature values extracted from the voice information divided into child groups (C); An identification unit -C123B that identifies gender and age with reference to the standard DB based on the representative feature value. Further, as described above, the combination calculation unit-C123A can receive the feature values extracted from the gender feature extraction unit 111, the age feature extraction unit-FC113, and the age feature extraction unit-C115.

  In order to identify such age and sex, the age and sex can be identified using algorithms such as GMM (GauSSian Mixture Model), NN (Neural Network), SVM (Support Vector Machine) and the like. However, the algorithm described above is merely an example, and it goes without saying that the age and sex can be identified from the feature values using various algorithms other than the algorithm described above.

  For example, when the GMM algorithm is used, each combination calculation unit 121A, 122A, 123A calculates N likelihood values (likelihood) corresponding to the number N of feature identification methods or the number N of a plurality of samples. Then, the representative value can be determined by such N likelihood values. In order to determine the representative value, the combination calculation units 121A, 122A, and 123A determine the average value of the N likelihood values, the maximum value, the minimum value, or the total value to obtain the representative value. Can be determined.

  In addition, the combination calculation units 121A, 122A, and 123A can identify the representative feature value by assigning a weight value in identifying the representative feature value. Such weights can be set by case or using information accumulated empirically. For example, in an environment where stuttering frequently occurs, the feature value portion corresponding to the stuttering band is set to a low weight value, and the feature value portion corresponding to the middle band of the general voice band is highly weighted. A value can be assigned. Each union operation unit 121A, 122A, 123A can determine representative feature values by assigning different weights to the above-described groups (male, female, child) and reflecting voice features. .

  In the above, audio information was divided into male groups, female groups, and child groups. However, if it is difficult to classify feature values extracted from audio information into specific groups, the extracted audio information is applied to each group in duplicate. It is preferable to do. In other words, in the case of speech information that is difficult to classify into groups, after applying the calculation to each corresponding group, the similarity between the results of each identification unit, the normal identification probability, the reliability, etc. are considered. Finally, the final identification unit 300 determines the age and sex.

  FIG. 3 is a detailed block diagram of the video processing unit shown in FIG.

  As shown in FIG. 3, the video processing unit 200 includes a video feature extraction unit 210 that extracts feature values from video information, and a video calculation unit 220 that calculates gender and age from the extracted feature values. .

  The video feature extraction unit 210 can extract feature values in response to input of video information. The video feature extraction unit 210 further includes an age-specific feature extraction unit 211, an age-specific feature extraction unit-C212, a sex-specific feature extraction unit-C213, a sex-specific feature extraction unit-A214, an age-specific feature extraction unit-M215, and an age-specific feature. A feature extraction unit-F216 may be included.

  The age-specific feature extraction unit 211 extracts feature values by reflecting the age-specific features with respect to the input face information, and the input face information based on the extracted feature values is converted into an adult group (A) or a child group. Classify into (C). For example, in the case of face information, it is easy to distinguish adults and children based on the proportionality between the size of the face and the size of the eyes, the presence or absence of wrinkles at the eyes, and the like. The age-specific feature extraction unit 211 can extract a feature value for face information input reflecting the above-mentioned age-specific features.

  For face information classified into child groups (C) by the age-specific feature extraction unit 211, an age-specific feature extraction unit -C212 extracts feature values reflecting the child's age-specific features, and a gender feature extraction unit- C213 extracts a feature value reflecting the sex characteristics of the child.

  The gender feature extraction unit-A 214 can extract a feature value by reflecting the gender feature of the adult in the face information classified into the adult group (A) by the age-specific feature extraction unit 211. Based on the extracted feature value, the input face information is divided into a male group (M) and a female group (F).

  The age-specific feature extraction unit-M215 extracts a feature value by reflecting the male age-specific feature in the face information classified into the male group (M) by the gender feature extraction unit-A214. The age-specific feature extraction unit-F216 extracts feature values reflecting the female age-specific features in the face information divided into the female group (F) by the gender feature extraction unit-A214.

  The video calculation unit 220 identifies gender and age from the video information using the feature values extracted by the video feature extraction unit 210 as described above.

  That is, the video calculation unit 220 reflects the weight value on one or more feature values extracted from the video feature extraction unit 210 to determine a representative feature value and a reference DB based on the representative feature value. It consists of an identification unit that identifies gender and age with reference.

  Further, as shown in FIG. 3, the video calculation unit 220 is optimal for each of the male group (M), the female group (F), and the child group (C) grouped by the video feature extraction unit 210 as described above. The combination operation unit and the identification unit are configured. That is, the video calculation unit 220 receives the input of the feature value extracted from the video information classified into the male group, and extracts the video calculation unit-M221 that calculates the age and sex, and the audio information classified into the female group. -F222 for calculating the age and sex by receiving the input of the feature value, and the video calculation unit for calculating the age and sex by receiving the input of the feature value extracted from the audio information divided into the child groups -C223.

  The image calculation unit-M221 receives a combination of one or more feature values extracted from the face information divided into the male group (M) and determines a representative feature value, and the representative feature value. An identification unit -M221B that identifies gender and age with reference to the standard DB based on the standard DB. Since the combination calculation unit-M221A receives input of feature values extracted from face information divided into male groups (M), the age-specific feature extraction unit 211, the gender feature extraction unit-A214, and the age-specific feature extraction unit-M215 The representative feature value can be identified by receiving the input of the feature value extracted from.

  The video calculation unit-F222 receives the input of feature values extracted from the face information divided into the female group (F) and determines the representative feature value, the combination calculation unit-F222A, the representative feature value, and the reference DB And an identification unit-F222B for identifying gender and age.

  In this case, since the combination calculation unit-F222A targets the face information divided into the female group (F), it is extracted from the age-specific feature extraction unit 211, the gender feature extraction unit-A214, and the age-specific feature extraction unit-F216. The representative feature value can be identified by receiving the inputted feature value.

  The video calculation unit-C223 receives the input of feature values extracted from the face information divided into the child group (C) and determines the representative feature value, and the combination calculation unit-C223A, the representative feature value and the reference DB And an identification unit -C223B for identifying gender and age. In this case, the combination calculation unit -C223A receives the input of the feature value extracted from the face information divided into the child group (C), so the age-specific feature extraction unit 211, the age-specific feature extraction unit -C212, and the gender feature The representative feature value can be identified by receiving the input of the feature value extracted from the extraction unit-C213.

  Each of the identification units 221B, 222B, and 223B can calculate gender and age with reference to the reference DB by receiving the representative feature values from the combination calculation units 221A, 222A, and 223A described above. Since the specific description thereof is similar to that described above with reference to the voice calculation unit 120, further detailed description thereof is omitted.

  In addition, when calculating age and gender using such a video processing unit 200, as described above, the face information is any of the male group (M), the female group (F), and the child group (C). When it is difficult to divide into one group, the face information can be applied to each group.

  Hereinafter, the final identification unit 300 will be described in detail.

  The final identification unit 300 receives gender and age input from some or all of the identification units 121B, 122B, 123B, 221B, 222B, and 223B, and performs a combination operation on the received gender and age. To identify the final gender and age.

  That is, it is possible to calculate the mutual similarity for each of the plurality of sexes and ages received, and determine the sex and age having the highest mutual similarity as the final sex and age. Alternatively, it is possible to grasp and store normal identification probabilities for a plurality of genders and ages that have been input and the reliability index every time they are identified, and use them to determine the final gender and age.

  The final identification unit 300 identifies the gender and the age using the mutual similarity with respect to the gender and the age output from the audio processing unit 100, and the gender and the age output from the video processing unit 200. After identifying the gender and age using the mutual similarity, the final gender and age can be identified and output using the two identified genders and ages.

  Alternatively, the final identification unit 300 may identify and output the final gender and age using the mutual similarity with respect to the entire sex and age identification results output from the audio processing unit 100 and the video processing unit 200. It can also be implemented.

  Hereinafter, the reference DB will be described in detail.

  The reference DB stores gender and age-specific reference feature values or audio and video reference samples, and includes a feature value extracted from face information or audio information, and a relationship model of sex and age with respect to the feature value. Is done.

  Using the feature value-gender and age correspondence stored in the reference DB, the voice calculation unit 120 or the video calculation unit 220 refers to the reference DB based on the representative feature value described above to determine the gender and age. Can be earned. For example, the identification unit can identify gender and age using the distance value between the representative feature value and the relationship model of the reference DB.

  In addition, the reference DB includes video data and audio data, and corresponding gender and age so that the gender and age can be identified directly using video or audio information when it is difficult to extract feature values smoothly. Is done.

  The video data included in the reference DB can be acquired, for example, by separating a camera and a person by a distance of 0.5 m, 1 m, and 3 m, respectively. At this time, if the separation distance is 3 m, the image is taken so that the whole body of the person is included. Such video data can be photographed to be 100 frames for 10 seconds. Use the face detector, height detector, eye detector, etc. for the images shot as described above to acquire the face, hairstyle, eyelid, eyebrow shape, etc. of each human subject, and use the detailed DB. Can be configured. The present invention can be implemented so as to identify feature values using the detailed DB configured as described above.

  In the case of speech data included in the reference DB, for example, 50 sentences prepared in advance can be obtained by uttering three times repeatedly. Such audio data can have various forms such as 16 kHz, 16 bit, and mono type.

  Since such a reference DB has a sample property, for example, data can be configured for 120 persons. At this time, the overall male / female ratio can be 1: 1, and the ratio to each age zone can also be 1: 1.

  The reference DB has learning ability, and when the calculation for gender-age is performed according to the embodiment of the present invention, the result value of the calculation (the representative characteristic value of the calculation and the final gender and age) is currently configured. It is preferable to reconstruct (update) the DB by reflecting it in the data so that the reliability can be continuously improved. Of course, it goes without saying that the result value used for DB update must be a result value for which reliability has been confirmed.

  FIG. 4 is a flowchart of the gender-age identification method according to the present invention.

  The input unit 10 collects face information and voice information of a specific person who wants to identify gender and age (S100).

  The speech processing unit 100 extracts feature values from the collected speech information by reflecting the age-specific features and the gender features, and identifies representative feature values for the extracted one or more feature values. Then, the representative feature value is inquired of the reference DB to identify the sex and age (S200).

  At the same time, the video processing unit 200 extracts the feature value by reflecting the age-specific feature and the gender feature from the face information, and identifies the representative feature value for the extracted one or more feature values. Then, the representative feature value is inquired of the reference DB to identify gender and age (S300).

  The final identification unit 300 finally identifies gender and age in consideration of the mutual similarity or probability with respect to at least one gender and age identified in steps S200 and S300 (S400).

  Hereinafter, with reference to FIG. 5, the step of identifying gender and age from the voice of FIG. 4 (S200) will be described in detail.

  In general, it is difficult to distinguish female and child's voice information because they are similar to each other. However, it is easy to distinguish between female and child's voice information and male's voice information. The feature value is preferentially reflected and the feature value is extracted to classify the male, female and child groups (S210).

  As described above, the preferential reflection of the gender feature on the voice information is based on the fact that the voice information has a large difference depending on the gender feature, so that the calculation can be performed quickly and efficiently. become.

  Based on the classification results, the input voice information is classified into male groups or female and child groups, and one or more age-specific feature values reflecting male age characteristics are extracted from the voice information classified into male groups. (S220).

  In addition, for voice information classified into female and child groups, age-specific features that reflect the age-specific features of women and children so that the voice information can be distinguished from female or child groups. A value is extracted and a woman and a child are distinguished (S230).

  After that, age-specific feature extraction that reflects the age-specific features of women is performed on the voice information that is classified into the female groups (S240).

  In addition, the child's sex and age-specific features are extracted from the voice information distinguished by the child group (S250).

  Among the extracted feature values, a representative feature value for the voice information is determined, and the sex and age of the target person are identified.

  For example, the voice calculation unit 120 determines a representative feature value for one or more feature values extracted by the voice feature extraction unit 110, and uses the reference DB to determine the gender and age based on the determined representative feature value. Can be identified. The determination of the representative feature value or the identification of the sex and the age is preferably performed for each of the male group, the female group, and the child group as described above.

  That is, whether voice calculation-M is performed on feature values of voice information classified into male groups (S225) or whether voice calculation-F is performed on feature values of voice information classified into female groups (S245). ), Voice calculation-C is performed on the feature values of the voice information divided into child groups (S255), and the sex and age can be identified.

  As described above, voice information is primarily grouped on the basis of features that can be easily distinguished (for example, voice information is a feature by gender), and features that reflect the characteristics of each group for each group distinguished by the criteria. The use of a method for extracting values is one of the major features of the present invention. By using the stepwise extraction method as described above, the present invention can ensure the accuracy of identification, and can quickly identify the age and gender of the subject by eliminating the duplication of computation. it can.

  Hereinafter, with reference to FIG. 6, the step of identifying the gender and age from the video of FIG. 4 (S300) will be described in detail.

  In the case of video information, it is generally easy to distinguish adults from children. For example, an adult and a child can be easily distinguished by using biological information such as height or using a relative ratio between the size of the face and the size of the ears, eyes, mouth, and nose.

  By using such points, in the video similarity identification step of the present invention, for the video information (video information including face information or face information; hereinafter referred to as “face information”) that is primarily input. Feature value extraction is performed in consideration of the age-specific features (S310). The face information input through the above steps can be easily divided into a child group and an adult group.

  Then, age-specific features are extracted from the face information divided into child groups in consideration of the age-specific features of the children (S320), and gender feature extraction is performed in consideration of the child's gender features (S330).

  The face information classified into the adult groups is subjected to sex feature extraction considering the sex characteristics of the adults to distinguish the face information of the adult groups into male groups or female groups (S340).

  Thereafter, one or more feature values are extracted from the face information classified into male groups using a feature extraction method that takes into account the male age-specific features (S350). For face information classified into female groups, one or more feature values are extracted in consideration of the age-specific features of women (S360).

  The video calculation unit 220 determines a representative feature value by reflecting the weight value with respect to the feature value extracted by the video feature extraction unit 210 as described above, and uses the determined representative feature value and the reference DB to determine the gender. And identify the age. Such sex and age identification based on video information is preferably performed for each of a child group, a male group, and a female group, as shown in FIG. 6 (S325, S355, S365).

  The present invention has been described in detail with reference to the accompanying drawings. However, this is merely an example, and it is obvious that various modifications and changes can be made within the scope of the technical idea of the present invention. . Therefore, the protection scope of the present invention should not be limited to the above-described embodiments, but should be determined including the scope described in the appended claims and a scope equivalent thereto.

DESCRIPTION OF SYMBOLS 10 Input part 20 Age-sex calculating part 30 Output part 100 Audio | voice processing part 200 Image | video processing part 300 Final identification part

Claims (19)

Collecting video information and audio information;
Extracting one or more feature values from the collected voice information, and using the extracted feature values to identify gender and age;
Extracting one or more feature values for the collected video information, and using the extracted feature values to identify gender and age;
Finally determining the gender and age by performing a combination operation of the gender and age identified using the voice information and the gender and age identified using the face information;
A gender-age identification method comprising: The gender and age identification step using the audio information includes:
A first gender feature extraction step of extracting a feature value reflecting the gender feature of the voice with respect to the input voice information;
A first age-specific feature extraction step for extracting feature values reflecting the male age-specific features for the voice information divided into male groups by the first gender feature extraction step;
A second age-specific feature extracting step of extracting feature values reflecting the age-specific features of women and children with respect to the voice information divided into the female and child groups by the first sex-feature extracting step;
The gender-age identification method according to claim 1, further comprising: The gender and age identification step using the audio information includes:
A third age-specific feature extraction step for extracting a feature value reflecting the female age-specific feature for the voice information classified into the female group by the second age-specific feature extracting step;
A second sex feature extraction step for extracting a feature value reflecting the sex feature of the child with respect to the voice information divided into child groups by the second age feature extraction step;
The gender-age identification method according to claim 2, further comprising:   The gender-age identification method according to claim 2 or 3, wherein the feature values are extracted from M samples by applying different N feature value identification methods. The gender and age identification step using the audio information includes:
A representative feature value determining step of determining a representative feature value by reflecting a weight value with respect to the one or more extracted feature values;
An identification step of identifying the gender and age with reference to a standard DB based on the representative feature value;
The gender-age identification method according to claim 1, further comprising:   The gender-age identification method according to claim 5, wherein the representative feature value determination step and the identification step are performed for each of a male group, a female group, and a child group.   The representative feature value determining step includes a step of determining any one of an average value, a maximum value, a minimum value, and a sum value of feature values reflecting one or more of the weight values as the representative feature value. The gender-age identification method according to claim 5, further comprising: Gender and age identification step using the video information,
A first feature extracting step of extracting a feature value reflecting the age-specific feature with respect to the collected video information;
A second feature extracting step of distinguishing between an adult and a child according to a result of the first feature extracting step, and then classifying the group into a group of men, women and children and extracting one or more feature values for each group;
The gender-age identification method according to claim 1, further comprising: Gender and age identification step using the video information,
A representative feature value determining step of determining a representative feature value by reflecting a weight value with respect to the one or more extracted feature values;
An identification step of identifying the gender and age using a reference DB based on the representative feature value,
The gender-age identification method according to claim 8, wherein the representative feature value determination step and the calculation step include being performed for each of a male group, a female group, and a child group. The final step of gender and age is
Calculating a gender and age identified using at least one of the audio information and a gender and age identified using at least one of the video information;
Determining the gender and age with the highest mutual similarity as the final gender and age;
The gender-age identification method according to claim 1, further comprising:   10. The reference DB according to claim 5, wherein the reference DB includes gender and age characteristic values, and is continuously reconstructed to reflect characteristic values whose reliability is confirmed according to gender and age. Described gender-age identification method. An input unit for collecting video information and audio information;
A voice processing unit that extracts a feature value for the collected voice information, and uses the extracted feature value to identify gender and age from the voice information;
Extracting a feature value for the collected video information, and using the extracted feature value, a video processing unit for identifying gender and age from the video information;
A final identification unit that finally determines the gender and age of the specific person by performing a combination operation of the sex and age identified by the video processing unit and the gender and age identified by the audio processing unit;
A gender-age identification device comprising: The voice processing unit
A voice feature extraction unit that extracts a feature value by reflecting a gender feature or an age feature of the voice with respect to the collected voice information;
A representative feature value extracted from the feature value extracted from the voice feature extraction unit, and a voice calculation unit that identifies the gender and age using the representative feature value;
The gender-age identification apparatus according to claim 12, comprising:   The speech feature extraction unit may determine whether or not the collected speech is a male speech before extracting one or more feature values for each group of men, women, and children. Item 14. The gender-age identification device according to Item 13.   The gender-age identification apparatus according to claim 14, wherein the voice calculation unit determines a representative characteristic value for each of the male, female, and child groups, and identifies the gender and the age using the representative characteristic value. The video processing unit
A video feature extraction unit that extracts a feature value by reflecting a gender feature or an age feature of the video with respect to the collected video information;
A video feature that extracts representative feature values from the feature values extracted from the video feature extractor, and identifies the sex and age using the representative feature values;
The gender-age identification apparatus according to claim 12, comprising:   The video feature extraction unit may determine whether the collected audio is an adult or a child, and then extract one or more feature values for each group of men, women, and children. Item 17. The gender-age identification device according to Item 16.   The gender-age identification apparatus according to claim 17, wherein the video calculation unit determines representative characteristic values for each of the male, female, and child groups, and identifies gender and age using the representative characteristic values.   The final identification unit calculates each mutual similarity for at least one age and sex identified by the audio processing unit or the video processing unit, and finally determines the gender and age having the highest mutual similarity. The gender-age identification apparatus according to claim 12.

Images