CN106326857A - Gender identification method and gender identification device based on face image - Google Patents

Abstract

The invention discloses a gender recognition method and device based on a human face image. The method includes: performing gender classification and discrimination on a target human face image through a preset prior model in a deep convolutional neural network to obtain the target human face image For the corresponding gender discrimination result, by comparing the cosine similarity between the features of the target face image and the features of the image sample, the cosine comparison result corresponding to the target face image is obtained, and the gender discrimination result is compared with the cosine Perform Bayesian classifier operations on the results to obtain the gender of the object in the face image. In this way, in complex imaging situations, the key parts of the face image are re-spliced, and the deep convolutional neural network is used to distinguish and The combination of image feature comparison can use the features of key parts of the face, depth texture, edge and color features for auxiliary recognition, increasing the accuracy of gender recognition.

Description

Gender recognition method and device based on face image

technical field

The invention belongs to the technical field of image recognition, and in particular relates to a gender recognition method and device based on a human face image.

Background technique

Face recognition is a biometric technology for identification based on human facial feature information. Use a video camera or camera to collect images or video streams containing human faces, and automatically detect and track human faces in the images, usually also called portrait recognition and facial recognition. Face recognition products have been widely used in finance, justice, military, public security, border inspection, government, aerospace, electric power, factories, education, medical care and many enterprises and institutions. For example, face recognition access control time attendance system and face recognition anti-theft door, computer login, e-government and e-commerce regarding information security. With the further maturity of technology and the improvement of social recognition, face recognition technology will be applied in more fields. Among them, image-based gender recognition can effectively assist face recognition, and the accuracy of gender recognition can directly affect the accuracy of final face recognition.

In the prior art, the process of identifying the gender of a face image is as follows: first face detection, then feature extraction of the face image, and finally identify the gender of the face image through a classifier according to the extracted features. Due to the relationship between the imaging equipment, the image may not clearly reflect the complete face of the person. Most of the images will appear blurry, strong light, dark, etc. Face posture, under such complex imaging conditions, existing technologies cannot accurately determine the gender of a person by simply extracting the features of the face image, which in turn will increase the error rate of the recognition result.

Contents of the invention

The embodiment of the present invention provides a gender recognition method and device based on a face image, aiming to solve the problem that the features of the face image cannot be accurately extracted due to changes in external imaging factors, which will increase the error rate of recognition.

A gender recognition method based on a face image provided by an embodiment of the present invention includes: performing gender classification and discrimination on a target face image through a preset prior model in a deep convolutional neural network to obtain the target face image Corresponding gender discrimination results, wherein the target face image is an image composed of key parts in the face image; by comparing the cosine similarity between the features of the target face image and the features of the image samples, Obtaining a cosine comparison result corresponding to the target face image; performing a Bayesian classifier operation on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image.

A gender recognition device based on a human face image provided by an embodiment of the present invention includes: a gender discrimination module for performing gender classification and discrimination on a target human face image through a preset prior model in a deep convolutional neural network to obtain the The gender discrimination result corresponding to the target human face image, wherein the target human face image is an image composed of each key part area in the human face image; the comparison processing module is used to compare the features of the target human face image with the The cosine similarity between the features of the image samples is used to obtain the corresponding cosine comparison result of the target face image; the identification module is used to perform Bayesian classifier operations on the gender discrimination result and the cosine comparison result, Obtain the gender of the object in the face image.

The gender recognition method and device based on human face images provided by the embodiments of the present invention use the preset deep convolutional neural network prior model to classify and judge the gender of the target human face image, and obtain the corresponding gender of the target human face image. Gender discrimination results, wherein the target face image is an image composed of key parts in the face image, by comparing the cosine similarity between the features of the target face image and the features of the image sample, the obtained The cosine comparison result corresponding to the target face image, the Bayesian classifier operation is performed on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image, so in complex imaging situations Next, by re-splicing each key part of the face image, and combining the deep convolutional neural network discrimination and image feature comparison, the features, depth texture, edge and color features of each key part of the face can be used Assisted identification increases the accuracy of gender identification.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only are some embodiments of the present invention.

Fig. 1 is the realization flow diagram of the gender recognition method based on face image that the first embodiment of the present invention provides;

Fig. 2 is a schematic diagram of the implementation flow of the gender recognition method based on the face image provided by the second embodiment of the present invention;

Fig. 3 is the schematic diagram of face detection in the embodiment of the present invention;

Fig. 4 is the schematic diagram of face image in the embodiment of the present invention;

FIG. 5 is a schematic diagram of a target face image obtained by reconstruction of each key part area in an embodiment of the present invention;

6 is a schematic structural diagram of a gender recognition device based on a face image provided by a third embodiment of the present invention;

Fig. 7 is a schematic structural diagram of a gender recognition device based on a face image according to a fourth embodiment of the present invention.

FIG. 8 is a schematic diagram of a hardware structure of an electronic device for a gender recognition method based on a face image according to a fifth embodiment of the present invention.

detailed description

In order to make the purpose, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described The embodiments are only some of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts belong to the protection scope of the present invention.

Please refer to Fig. 1, Fig. 1 is a schematic diagram of the implementation flow of the gender recognition method based on face images provided by the first embodiment of the present invention, which can be applied to face recognition equipment, face anti-theft systems, computers and other terminals for recognizing face images middle. The gender recognition method based on face images shown in Figure 1 mainly includes the following steps:

S101. Perform gender classification and discrimination on the target face image through the preset prior model in the deep convolutional neural network, and obtain a gender discrimination result corresponding to the target face image.

The target face image is an image composed of key parts in the face image. The key part areas may be the areas of various organ parts of the human face and the facial contour area, and the key part areas in the target human face image are not limited here. It should be noted that the relationship between the face image and the target person image is: the target face image is an image composed of the key parts in the face image. The face image is used to represent an image of the whole face.

The deep convolutional neural network is a convolutional neural network (CNN, Convolutional Neural Networks,), which is a deep neural network with a convolutional structure, including at least two nonlinear trainable convolutional layers, two nonlinear Fixed convolutional layer and fully connected layer, a total of at least 5 hidden layers, mainly used in speech analysis and image recognition.

A priori model can be obtained by training image samples through a deep convolutional neural network. The process of training image samples to obtain a priori model is the process of classifying image samples according to preset categories, and the prior model can be used to classify images. Classification models.

S102. Obtain a cosine comparison result corresponding to the target face image by comparing the cosine similarity between the features of the target face image and the features of the image sample.

In the field of image recognition, the feature of an image is a proper term in the field of image recognition, and the extraction of image features is a concept in computer vision and image processing.

The features of the image samples can be stored in the built-in storage module of the terminal, or can be stored on the cloud server. Cosine similarity uses the cosine value of the angle between two vectors in a vector space as a measure of the difference between two individuals.

S103. Perform a Bayesian classifier operation on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image.

It should be noted that the object in the face image is the person depicted in the face image. The Bayesian classifier is an algorithm that calculates the probability of which category an object belongs to through the Bayesian formula. The principle of Bayesian classifier operation is to use the Bayesian formula to calculate the posterior probability of the object through the prior probability of an object. Probability, that is, the probability that the object belongs to a certain class, and the class with the largest posterior probability is selected as the class to which the object belongs.

In the embodiment of the present invention, the gender discrimination result and the cosine comparison result are used as the prior probability of the object in the face image, and the posterior probability of the male category and the female category to which the object belongs is calculated by Bayesian formula, And select the gender category with the largest posterior probability as the gender category of the object.

In the embodiment of the present invention, the gender classification and discrimination are performed on the target face image through the preset deep convolutional neural network prior model, and the gender discrimination result corresponding to the target face image is obtained, wherein the target face image is composed of The image composed of each key part area in the face image, and then by comparing the cosine similarity between the features of the target face image and the features of the image sample, the cosine comparison result corresponding to the target face image is obtained, and finally The gender discrimination result and the cosine comparison result are subjected to a Bayesian classifier operation to obtain the gender of the object in the face image. In this way, in complex imaging situations, the key parts in the face image are re-spliced, and Combining deep convolutional neural network discrimination and image feature comparison, it can use the features of key parts of the face, depth texture, edge and color features for auxiliary recognition, increasing the accuracy of gender recognition.

Please refer to Fig. 2, Fig. 2 is a schematic diagram of the implementation flow of the gender recognition method based on face images provided by the second embodiment of the present invention, which can be applied to face recognition equipment, face anti-theft systems, computers and other terminals for recognizing face images , mainly includes the following steps:

S201. According to the position of each key point of the face image in the face image, extract the key part area in the face image.

S202. Reconstruct the face image according to the key parts in the face image to obtain the target face image.

The target face image is an image composed of key parts in the face image.

Optionally, before step S201, the method further includes: determining a face image in the image to be recognized through face detection and face key point location; setting the face image in the image to be recognized as a detection area.

Use the wavelet (HAAR) classifier or DLIB (C++ library) algorithm to detect the face of the input image, and then use the supervised descent algorithm (SDM, Supervised Descent Method) to locate the key points of the face on the detected image. Among them, the key points of the face located by the SDM algorithm include: eyebrows, eyes, nose, mouth and facial contours. Of course, face detection and face key point positioning can also be realized by other algorithms.

The HAAR classifier includes an adaptive enhancement (Adaboost) algorithm. In the field of image recognition, a classifier refers to an algorithm for classifying human faces and non-human faces. DLIB is a C++ algorithm library that can be applied to face detection and face key point location.

Figure 3 is a schematic diagram of face detection. As shown in Figure 3, the dark black square frame is the face detection frame, the circle represents the face, the triangle represents the animal, and the polygon represents the tree. After the face detection, the face image can be extracted from the image. And set the face image in the face detection frame.

Optionally, the detection area is enlarged according to a preset multiple, so that the face image includes an image of a hair area. Expanding the detection area according to the preset multiple is to expand the hair area into the face image. The way to expand the detection area can be to expand the entire detection area, or to expand the upper part, lower part, and left and right parts of the detection area. It is also possible to expand only the upper and lower parts of the detection area by setting the multiple. The preset multiple is related to the length of the hair and the height of the hair on the top of the head. In this implementation, the preferred value of the preset multiple is 0.15, that is, the upper part and the lower part of the detection area are each enlarged by 0.15 times. Fig. 4 is a schematic diagram of a human face image, wherein the detection area is within the rectangular frame in Fig. 4 (a) and Fig. 4 (b), and the detection area in Fig. 4 (a) includes the face image before expanding the detection area, Fig. 4 The detection area in (b) includes the face image after expanding the detection area.

Since the color, length, and hairstyle of hair are very important basis for identifying gender, including the hair region in the face image can identify gender more accurately.

Optionally, the eye area in the detection area is corrected so that the eyes in the eye area in the face image are on the same horizontal line. The way to correct the eyes is not limited, the eyes can be corrected by geometric transformation, and the angle between the two eyes can also be changed to correct the eyes. The ultimate goal is to put the eyes in the eye area on the same horizontal line.

The key points of the face include: eyes, nose, mouth, eyebrows, hair and facial contour, and the key parts of the face image are extracted through the position of each key point in the face image. Wherein, the key part areas in the face image include: a hair area, an eyebrow area, an eye area, a nose area, and a mouth area. The target face image obtained by reconstructing each key part area can more accurately describe the local area of the face such as hairstyle, skin, and facial contour of faces of different genders.

As shown in FIG. 5 , FIG. 5 is a schematic diagram of a target face image reconstructed from each key part area. For the convenience of explanation and display, the key parts in the schematic diagram shown in Figure 5 only include the eye area, eyebrow area, nose area and mouth area, and other key parts are not shown in Figure 5, and the schematic diagram shown in Figure 5 is only An example does not constitute a limitation to key parts and regions in the present invention.

S203. Perform gender classification and discrimination on the target face image through the preset deep convolutional neural network prior model, and obtain a gender discrimination result corresponding to the target face image.

A deep convolutional neural network is a deep neural network with a convolutional structure, including at least two non-linear trainable convolutional layers, two non-linear fixed convolutional layers and a fully connected layer, with a total of at least 5 hidden It is mainly used in the fields of speech analysis and image recognition. For the explanation of the convolutional neural network and the prior model here, please refer to the relevant description in the first embodiment of the present invention, and details are not repeated here.

Optionally, through the prior model in the preset deep convolutional neural network, gender classification and discrimination are performed on the target face image, and the gender discrimination result corresponding to the target face image is obtained, specifically:

Classifying and discriminating the target face image through the prior model in the deep convolutional neural network, obtaining a maximum probability value corresponding to the target face image and a first target gender category corresponding to the maximum probability value;

The maximum probability value and the first target gender category are used as the gender discrimination result.

The gender category includes a female category and a male category. Multiple image samples of key parts are trained in a deep convolutional neural network, and a prior model is obtained after training. A priori model can be obtained by training image samples through a deep convolutional neural network. The process of training image samples to obtain a priori model is the process of classifying image samples according to preset gender categories. The prior model can be used to classify images A model for classifying discriminants. The first target gender category is a male category or a female category.

S204. Obtain a cosine comparison result corresponding to the target face image by comparing the cosine similarity between the features of the target face image and the features of the image sample.

The image sample is an image sample composed of various key parts in the overall face image sample, wherein the key parts in the overall face image sample include: hair area, eyebrow area, eye area, nose area, and mouth area. The key part area in the overall human face image sample corresponds to the area included in the key part area in the human face image, that is, if the overall human face image sample includes the eyebrow area, then the human face image includes the eyebrow area , if the overall face image sample includes the glasses area, then the face image includes the glasses area. The composition of the image samples obtained after reconstruction is similar to the image shown in Figure 5. The method of extracting the key part region from the whole face image sample is the same as the method of extracting the key part region from the face image, and will not be repeated here. Similarly, the reconstruction method of the reconstructed image sample is the same as the reconstruction method of the reconstructed target face image, which will not be repeated here.

The cosine similarity uses the cosine value of the angle between two vectors in the vector space as a measure to measure the size of the difference between two individuals.

Optionally, by comparing the cosine similarity between the features of the target face image and the features of the image sample, the cosine comparison result corresponding to the target face image is obtained, specifically:

By performing a cosine similarity comparison between the feature of the target face image and the feature of the image sample, a plurality of cosine similarity values corresponding to the target face image are obtained;

According to the order of the cosine similarity value from high to low, select a preset number of target face image samples;

Perform quantitative statistics on the gender categories of the target face image sample, and use the gender category with the largest number as the second target gender category;

Select a similar face image sample whose gender category is the second target gender category in the target face image sample, and extract the cosine similarity value between the similar face image sample and the target face image as the target cosine similarity degree value;

Calculate the average of the target cosine similarity values;

The second target gender category and the average value are used as the cosine comparison result.

The larger the value of the cosine similarity value, the higher the similarity. First, according to the calculated cosine similarity value, the image samples are arranged in order of similarity from high to low, and then a preset number of target face image samples are selected starting from the high similarity. The preset number can be selected arbitrarily. Of course, if the number selected is large, the accuracy of final discrimination will be improved. The preferred number of presets in this embodiment is 20.

The gender category of this target human face image sample is carried out quantitative statistics, and the gender category with the largest number of samples is used as the target gender category, if the number of samples of the male category in the target human face image sample is large, then the target gender category is male; If the number of samples of the female category in the target face image sample is large, the target gender category is female. If the target gender category is male, then select the image samples of men in the target face image samples as similar face image samples; if the target gender category is female, then select the image samples of women in the target face image samples as similar face images sample.

The following is a specific example to illustrate how to identify the gender of the object according to the cosine similarity comparison between the features of the target face image and the features of the image sample. The specific instructions are as follows:

For the convenience of explanation, let the characteristics of five image samples be: sample 1, sample 2, sample 3, sample 4, and sample 5, the target face image is image 1, and the preset number is 3; suppose, the gender of sample 1 The category is male, the gender category of sample 2 is female, the gender category of sample 3 is male, the gender category of sample 4 is female, and the gender category of sample 5 is male.

Compare the cosine similarity between the features of image 1 and the features of samples 1-5;

Calculate the cosine similarity value between image 1 and sample 1 as 1, calculate the cosine similarity value between image 1 and sample 2 as 2, and calculate the cosine similarity value between image 1 and sample 3 as 3 , Calculate the cosine similarity value between image 1 and sample 4 as a value 4, calculate the cosine similarity value between image 1 and sample 5 as a value 5, where value 5>value 3>value 4>value 1>value 2 ;

According to the order of the cosine similarity value calculated above from high to low, select a preset number of target face image samples as sample 5, sample 3 and sample 4;

The gender category of this target face image sample is carried out quantity statistics, the sample number of male category is 2, the sample number of female category is 1, then male category is the second target gender category;

Determine the target face image samples of the male category as samples 5 and 3, and calculate the average value 1 of the values 5 and 3;

The mean value 1 and the second target gender category are used as the cosine comparison result.

S205 , performing a Bayesian classifier operation on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image.

The Bayesian classifier is an algorithm that calculates the probability of which category an object belongs to through the Bayesian formula. The principle of Bayesian classifier operation is to use the Bayesian formula to calculate the posterior probability of the object through the prior probability of an object. Probability, that is, the probability that the object belongs to a certain class, and the class with the largest posterior probability is selected as the class to which the object belongs.

In the embodiment of the present invention, the gender discrimination result and the cosine comparison result are used as the prior probability of the object in the face image, and the posterior probability of the object belonging to the male category and the female category is calculated by the Bayesian formula, and The gender category with the largest posterior probability is selected as the gender category of the object.

In the embodiment of the present invention, according to the position of each key point of the face image in the face image, the key part area in the face image is extracted, and the image of the face is reconstructed according to the key part area in the face image to obtain The target face image, through the preset deep convolutional neural network prior model, performs gender classification and discrimination on the target face image, and obtains the gender discrimination result corresponding to the target face image. By comparing the target face image The cosine similarity between the features of the target face image and the features of the image sample is obtained, and the cosine comparison result corresponding to the target face image is obtained. Finally, the Bayesian classifier operation is performed on the gender discrimination result and the cosine comparison result to obtain the person The gender of the object in the face image, so that in complex imaging situations, by re-stitching the key parts of the face image, and combining the deep convolutional neural network discrimination with the feature comparison of the image, it is possible to use the face The features, depth texture, edge and color features of the key parts are assisted in the recognition, which increases the accuracy of gender recognition.

Please refer to FIG. 6 . FIG. 6 is a schematic structural diagram of a gender recognition device based on a face image according to a third embodiment of the present invention. For convenience of description, only parts related to the embodiment of the present invention are shown. The face image-based gender recognition device illustrated in FIG. 6 may be the subject of execution of the face image-based gender recognition method provided in the embodiment shown in FIG. 1 , and may be a terminal or a control module in the terminal. The gender recognition device based on the face image shown in FIG. 6 mainly includes: a gender discrimination module 601 , a comparison processing module 602 and a recognition module 603 . The above functional modules are described in detail as follows:

The gender discrimination module 601 is used to perform gender classification and discrimination on the target face image through the prior model in the preset deep convolutional neural network, and obtain the gender discrimination result corresponding to the target human face image;

Comparison processing module 602, for obtaining the cosine comparison result corresponding to the target human face image by comparing the cosine similarity between the features of the target human face image and the features of the image sample;

The identification module 603 is configured to perform a Bayesian classifier operation on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image.

The target face image is an image composed of key parts in the face image. The key part areas may be the areas of various organ parts of the human face and the facial contour area, and the key part areas in the target human face image are not limited here. It should be noted that the relationship between the face image and the target person image is: the target face image is an image composed of the key parts in the face image. The face image is used to represent an image of the whole face.

A deep convolutional neural network is a deep neural network with a convolutional structure, including at least two non-linear trainable convolutional layers, two non-linear fixed convolutional layers and a fully connected layer, with a total of at least 5 hidden It is mainly used in the fields of speech analysis and image recognition.

A priori model can be obtained by training image samples through a deep convolutional neural network. The process of training image samples to obtain a priori model is the process of classifying image samples according to preset categories, and the prior model can be used to classify images. Classification models.

In the field of image recognition, the feature of an image is a proper term in the field of image recognition, and the extraction of image features is a concept in computer vision and image processing.

The features of the image samples can be stored in the built-in storage module of the terminal, or can be stored on the cloud server. Cosine similarity uses the cosine value of the angle between two vectors in a vector space as a measure of the difference between two individuals. It should be noted that the object in the face image is the person depicted in the face image.

The Bayesian classifier is an algorithm that calculates the probability of which category an object belongs to through the Bayesian formula. The principle of Bayesian classifier operation is to use the Bayesian formula to calculate the posterior probability of the object through the prior probability of an object. Probability, that is, the probability that the object belongs to a certain class, and the class with the largest posterior probability is selected as the class to which the object belongs. In the embodiment of the present invention, the gender discrimination result and the cosine comparison result are used as the prior probability of the object in the face image, and the posterior probability of the male category and the female category to which the object belongs is calculated by Bayesian formula, And select the gender category with the largest posterior probability as the gender category of the object.

It should be noted that, in the implementation of the gender recognition device based on the face image shown in FIG. 6 above, the division of each functional module is only an example. In actual applications, it can be based on the needs, such as the configuration requirements of the corresponding hardware or the realization of the software. Considering the convenience, the above-mentioned functions are allocated by different functional modules. Moreover, in practical applications, the corresponding functional modules in this embodiment may be implemented by corresponding hardware, or may be completed by corresponding hardware executing corresponding software. Each of the embodiments provided in this specification can apply the above-mentioned description principle, which will not be described in detail below.

In the embodiment of the present invention, the gender discrimination module 602 performs gender classification and discrimination on the target face image through the preset deep convolutional neural network prior model, and obtains the gender discrimination result corresponding to the target face image, wherein the target The face image is an image composed of key parts in the face image, and the comparison processing module 603 obtains the target face image by comparing the cosine similarity between the features of the target face image and the features of the image sample For the corresponding cosine comparison result, the recognition module 604 performs Bayesian classifier operation on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image. The key parts of the face image are re-stitched, and the deep convolutional neural network discrimination is combined with the feature comparison of the image, which can use the features, depth texture, edge and color features of each key part of the face for auxiliary recognition, increasing the accuracy of gender recognition.

Please refer to FIG. 7 , which is a schematic structural diagram of a gender recognition device based on a face image provided by a fourth embodiment of the present invention. For convenience of description, only the parts related to the embodiment of the present invention are shown. The face image-based gender recognition device illustrated in FIG. 7 may be the subject of execution of the face image-based gender recognition method provided in the embodiment shown in FIG. 2 , and may be a terminal or a control module in the terminal. The gender recognition device based on the face image shown in Figure 7 mainly includes: an extraction module 701, a reconstruction module 702, a gender discrimination module 703, a comparison processing module 704, and a recognition module 705, wherein the gender discrimination module 703 includes: a discrimination module 7031 and setting module 7032; comparison processing module 704 includes: comparison module 7041, selection module 7042, statistics module 7043 and setting module 7044. The above functional modules are described in detail as follows:

The extraction module 701 is configured to extract the key part area in the face image according to the position of each key point of the face image in the face image.

The reconstruction module 702 is configured to reconstruct the image of the face according to the key parts in the face image to obtain the target face image.

The target face image is an image composed of key parts in the face image.

Optionally, the device further includes: a determination module for determining the face image in the image to be recognized through face detection and face key point positioning; the determination module is also used for setting the face image in the image to be recognized as the detection area.

Use the HAAR classifier or DLIB algorithm to detect the face of the input image, and then use the SDM algorithm to locate the key points of the face on the detected image. The key points of the face located by the SDM algorithm include: eyebrows, eyes, nose, Mouth and facial contours. Of course, face detection and face key point positioning can also be realized by other algorithms.

The HAAR classifier includes the Adaboost algorithm. In the field of image recognition, a classifier refers to an algorithm for classifying human faces and non-human faces. DLIB is a C++ algorithm library that can be applied to face detection and face key point location.

Figure 3 is a schematic diagram of face detection. As shown in Figure 3, the dark black square frame is the face detection frame, the circle represents the face, the triangle represents the animal, and the polygon represents the tree. After the face detection, the face image can be extracted from the image. And set the face image in the face detection frame.

Optionally, the device further includes an expansion module for expanding the detection area according to a preset multiple, so that the face image includes an image of a hair area. The expansion module expands the detection area according to the preset multiple to expand the hair area into the face image. The way to expand the detection area can be to expand the entire detection area, or to expand the upper, lower, and left and right parts of the detection area. Enlarging the preset multiple can also expand only the upper and lower parts of the detection area. The preset multiple is related to the length of the hair and the height of the hair on the top of the head. In this implementation, the preferred value of the preset multiple is 0.15, that is, the upper part and the lower part of the detection area are each enlarged by 0.15 times. Fig. 4 is a schematic diagram of a human face image, wherein the detection area is within the rectangular frame in Fig. 4 (a) and Fig. 4 (b), and the detection area in Fig. 4 (a) includes the face image before expanding the detection area, Fig. 4 The detection area in (b) includes the face image after expanding the detection area.

Since the color, length, and hairstyle of hair are very important basis for identifying gender, including the hair region in the face image can identify gender more accurately.

Optionally, the device further includes a correction module for correcting the eye area in the detection area, so that the eyes in the eye area in the face image are on the same horizontal line. The way to correct the eyes is not limited, the eyes can be corrected by geometric transformation, and the angle between the two eyes can also be changed to correct the eyes. The ultimate goal is to put the eyes in the eye area on the same horizontal line.

The key points of the human face include: eyes, nose, mouth, eyebrows, hair and facial contour. The extraction module 701 extracts the key parts of the human face image based on the positions of each key point in the human face image. Wherein, the key part areas in the face image include: a hair area, an eyebrow area, an eye area, a nose area, and a mouth area. The target face image obtained by reconstructing each key part area through the reconstruction module 702 can more accurately describe the local area of the face such as hairstyle, skin, and facial contour of faces of different genders.

As shown in FIG. 5 , FIG. 5 is a schematic diagram of a target face image reconstructed from each key part area. For the convenience of explanation and display, the key parts in the schematic diagram shown in Figure 5 only include the eye area, eyebrow area, nose area and mouth area, and other key parts are not shown in Figure 5, and the schematic diagram shown in Figure 5 is only An example does not constitute a limitation to key parts and regions in the present invention.

The gender discrimination module 703 is configured to perform gender classification and discrimination on the target face image through the preset prior model in the deep convolutional neural network, and obtain a gender discrimination result corresponding to the target face image.

Optionally, the gender discrimination module 703 includes: a discrimination module 7031 and a setting module 7032;

The discrimination module 7031 is used to classify and discriminate the target face image through the prior model in the deep convolutional neural network, and obtain the maximum probability value corresponding to the target human face image and the first target gender corresponding to the maximum probability value category;

A setting module 7032, configured to use the maximum probability value and the first target gender category as the gender discrimination result.

The first target gender category is a male category or a female category. A deep convolutional neural network is a deep neural network with a convolutional structure, including at least two non-linear trainable convolutional layers, two non-linear fixed convolutional layers and a fully connected layer, with a total of at least 5 hidden It is mainly used in the fields of speech analysis and image recognition. For the explanation of the convolutional neural network and the prior model here, please refer to the relevant description in the first embodiment of the present invention, and details are not repeated here.

The comparison processing module 704 is configured to obtain a cosine comparison result corresponding to the target human face image by comparing the cosine similarity between the features of the target human face image and the features of the image sample.

The image sample is an image sample composed of various key parts in the overall face image sample, wherein the key parts in the overall face image sample include: hair area, eyebrow area, eye area, nose area, and mouth area. The key part area in the overall human face image sample corresponds to the area included in the key part area in the human face image, that is, if the overall human face image sample includes the eyebrow area, then the human face image includes the eyebrow area , if the overall face image sample includes the glasses area, then the face image includes the glasses area. The composition of the image samples obtained after reconstruction is similar to the image shown in Figure 5. The method of extracting the key part region from the whole face image sample is the same as the method of extracting the key part region from the face image, and will not be repeated here. Similarly, the reconstruction method of the reconstructed image sample is the same as the reconstruction method of the reconstructed target face image, which will not be repeated here.

The cosine similarity uses the cosine value of the angle between two vectors in the vector space as a measure to measure the size of the difference between two individuals.

Optionally, the comparison processing module 704 includes: a comparison module 7041 , a selection module 7042 , a statistics module 7043 , a calculation module 7044 and a setting module 7045 .

The comparison module 7041 is used to obtain a plurality of cosine similarity values corresponding to the target face image by performing a cosine similarity comparison between the features of the target face image and the features of the image sample;

The selection module 7042 is used to select a preset number of target face image samples according to the order of the cosine similarity value from high to low;

Statistical module 7043, used for counting the gender category of the target face image sample, and using the gender category with the largest number as the second target gender category;

The selecting module 7042 is also used to select a similar human face image sample whose gender category is the second target gender category in the target human face image sample, and extract the cosine between the similar human face image sample and the target human face image The similarity value is used as the target cosine similarity value;

Calculation module 7044, for calculating the average value of the target cosine similarity value;

A setting module 7045, configured to use the second target gender category and the average value as the cosine comparison result.

The larger the value of the cosine similarity value, the higher the similarity. The preset number can be selected arbitrarily. Of course, if the number selected is large, the accuracy of final discrimination will be improved. The preferred number of presets in this embodiment is 20.

Statistical module 7043 performs quantitative statistics on the gender category of the target face image sample, and uses the gender category with the largest number of samples as the target gender category. If the number of samples of the male category in the target human face image sample is large, the target gender category is a male; if the number of samples of the female category in the target face image sample is large, the target gender category is female. If the target gender category is male, then the selection module 7042 selects the image samples of men in the target face image samples as similar face image samples, and if the target gender category is female, then the selection module 7042 selects the images of women in the target face image samples samples as similar face image samples.

The recognition module 705 is configured to perform Bayesian classifier operations on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image.

The Bayesian classifier is an algorithm that calculates the probability of which category an object belongs to through the Bayesian formula. The principle of Bayesian classifier operation is to use the Bayesian formula to calculate the posterior probability of the object through the prior probability of an object. Probability, that is, the probability that the object belongs to a certain class, and the class with the largest posterior probability is selected as the class to which the object belongs.

In the embodiment of the present invention, the gender discrimination result and the cosine comparison result are used as the prior probability of the object in the face image, and the identification module 705 calculates the posterior probability of the object belonging to the male category and the female category through the Bayesian formula probability, and select the gender category with the largest posterior probability as the gender category of the object.

In the embodiment of the present invention, the extraction module 701 extracts the key part area in the face image according to the position of each key point of the face image in the face image, wherein the key part area in the face image Including: hair area, eyebrow area, eye area, nose area, mouth area, the reconstruction module 702 reconstructs the image of the human face according to the key part area in the human face image, and obtains the target human face image, and the gender discrimination module 703 Perform gender classification and discrimination on the target face image through the preset prior model in the deep convolutional neural network, and obtain the gender discrimination result corresponding to the target face image, wherein the target face image is composed of the face image The image composed of each key part area in the image, and then the comparison processing module 704 obtains the cosine comparison corresponding to the target face image by comparing the cosine similarity between the features of the target face image and the features of the image sample As a result, the final recognition module 705 performs Bayesian classifier operations on the gender discrimination result and the cosine comparison result to obtain the gender of the object in the face image. The key parts are re-stitched, and the deep convolutional neural network discrimination is combined with the feature comparison of the image, which can use the features, depth texture, edge and color features of each key part of the face for auxiliary recognition, increasing the gender recognition. accuracy.

Fig. 8 is a schematic diagram of the hardware structure of the electronic device of the gender recognition method based on the face image provided by the fifth embodiment of the present invention. As shown in Fig. 8, the electronic device includes:

One or more processors 810 and memory 820, one processor 810 is taken as an example in FIG. 8 .

The electronic device may further include: an input device 830 and an output device 840 .

The processor 810, the memory 820, the input device 830, and the output device 840 may be connected through a bus or in other ways. In FIG. 8, connection through a bus 850 is taken as an example.

The memory 820, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs and modules, such as program instructions corresponding to the data storage method in the embodiment of the present application /modules (for example, gender discrimination module 601, comparison processing module 602 and recognition module 603 shown in FIG. 6). The processor 810 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 820, that is, realizes the face image-based gender recognition method of the above-mentioned method embodiment.

The memory 820 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required by a function; data etc. In addition, the memory 820 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the storage 820 may optionally include storages that are remotely located relative to the processor 810, and these remote storages may be connected to the gender recognition device based on facial images through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The input device 830 can receive input numbers or character information, and generate key signal input related to user settings and function control of the gender recognition device based on facial images. The output device 840 may include a display device such as a display screen.

The one or more modules are stored in the memory 820, and when executed by the one or more processors 810, perform the gender recognition method based on a face image in any method embodiment above.

The above-mentioned products can execute the method provided by the embodiment of the present application, and have corresponding functional modules and beneficial effects for executing the method. For technical details not described in detail in this embodiment, refer to the method provided in the embodiment of this application.

The electronic equipment of the embodiment of the present invention exists in various forms, including but not limited to:

(1) Mobile communication equipment: This type of equipment is characterized by mobile communication functions, and its main goal is to provide voice and data communication. Such terminals include: smart phones, multimedia phones, feature phones, and low-end phones.

(2) Ultra-mobile personal computer equipment: This type of equipment belongs to the category of personal computers, has computing and processing functions, and generally has the characteristics of mobile Internet access. Such terminals include: PDA Personal Digital Assistant, Mobile Internet Device (MID, Mobile Internet Device) and Ultra-mobile Personal Computer (UMPC, Ultra-mobile Personal Computer) equipment, etc.

(3) Portable entertainment equipment: This type of equipment can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, as well as smart toys and portable car navigation devices.

(4) Server: A device that provides computing services. The composition of a server includes a processor, hard disk, memory, system bus, etc. The server is similar to a general-purpose computer architecture, but due to the need to provide high-reliability services, it is important in terms of processing power and stability. , Reliability, security, scalability, manageability and other aspects have high requirements.

(5) Other electronic devices with data interaction function.

In the multiple embodiments provided in this application, it should be understood that the disclosed system, terminal and method may be implemented in other ways. For example, the terminal embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods. For example, multiple modules or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication link shown or discussed may be through some interfaces, and the indirect coupling or communication link between modules may be in electrical, mechanical or other forms.

The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or may be distributed to multiple network modules. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, each module may exist separately physically, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules.

If the integrated modules are realized in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk, and other media that can store program codes.

It should be noted that, for the sake of simplicity of description, all the aforementioned method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence. Because of the present invention, certain steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

In the foregoing embodiments, the descriptions of each embodiment have their own emphases, and for parts not described in detail in a certain embodiment, reference may be made to relevant descriptions of other embodiments.

The above is the description of the face image-based gender recognition method and device provided by the present invention. For those skilled in the art, according to the idea of the embodiment of the present invention, there will be changes in the specific implementation and application scope. In summary, the content of this specification should not be construed as limiting the present invention.

Claims (10)

1. a gender identification method based on facial image, it is characterised in that including:

By prior model in preset degree of depth convolutional neural networks, target facial image is carried out sex discriminant classification, obtains The Sex Discrimination result that described target facial image is corresponding, wherein said target facial image is by key portion each in facial image The image of region, position composition；

By the cosine similarity between feature and the feature of image pattern of target facial image described in comparison, obtain described mesh The cosine comparison result that mark facial image is corresponding；

Described Sex Discrimination result and described cosine comparison result are carried out Bayes classifier computing, obtains described facial image The sex of middle object.

Method the most according to claim 1, it is characterised in that in described facial image, each key position region includes: head Send out region, brow region, eye areas, nasal area, face region.

Method the most according to claim 2, it is characterised in that described by priori in preset degree of depth convolutional neural networks Model, carries out sex discriminant classification to target facial image, obtains the Sex Discrimination result that described target facial image is corresponding, tool Body includes:

By prior model in described degree of depth convolutional neural networks, described target facial image is carried out discriminant classification, obtains institute State most probable value corresponding to target facial image and first object sex classification corresponding to described most probable value；

Using described most probable value and described first object sex classification as described Sex Discrimination result.

Method the most according to claim 2, it is characterised in that described image pattern is by each in overall face image pattern The image pattern of key position region composition, in wherein said overall face image pattern, key position region includes: hair district Territory, brow region, eye areas, nasal area, face region.

Method the most according to claim 4, it is characterised in that described by the feature of target facial image described in comparison with Cosine similarity between the feature of image pattern, obtains the cosine comparison result that described target facial image is corresponding, specifically wraps Include:

By by carrying out cosine similarity comparison between feature and the feature of image pattern of described target facial image, obtaining many Individual cosine similarity value；

According to described cosine similarity value order from high to low, choose the target face image pattern of preset number；

The sex classification of described target face image pattern is carried out quantity statistics, and using sex classifications most for quantity as Two target gender classifications；

The similar face image that sex classification is described second target gender classification is chosen in described target face image pattern Sample, and extract cosine similarity value between described similar face image sample and described target facial image as more than target String Similarity value；

Calculate the meansigma methods of described target cosine similarity value；

Using described second target gender classification and described meansigma methods as described cosine comparison result.

6. a sex identification device based on facial image, it is characterised in that including:

Sex Discrimination module, for by prior model in preset degree of depth convolutional neural networks, is carried out target facial image Gender Classification differentiate, obtain the Sex Discrimination result that described target facial image is corresponding, wherein said target facial image be by The image of each key position region composition in facial image；

Comparison processing module, for by the cosine between feature and the feature of image pattern of target facial image described in comparison Similarity, obtains the cosine comparison result that described target facial image is corresponding；

Identification module, for described Sex Discrimination result and described cosine comparison result are carried out Bayes classifier computing, The sex of object in described facial image.

Device the most according to claim 6, it is characterised in that in described facial image, each key position region includes: head Send out region, brow region, eye areas, nasal area, face region.

Device the most according to claim 7, it is characterised in that described Sex Discrimination module includes:

Discrimination module, for by prior model in described degree of depth convolutional neural networks, is carried out described target facial image point Class differentiates, obtains most probable value corresponding to described target facial image and first object corresponding to described most probable value Other classification；

Setting module, for using described most probable value and described first object sex classification as described Sex Discrimination result.

Device the most according to claim 7, it is characterised in that described image pattern is by each in overall face image pattern The image pattern of key position region composition, in wherein said overall face image pattern, key position region includes: hair district Territory, brow region, eye areas, nasal area, face region.

Device the most according to claim 8, it is characterised in that described comparison processing module includes:

Comparing module, for by similar by carrying out cosine between feature and the feature of image pattern of described target facial image Degree comparison, obtains multiple cosine similarity value；

Choose module, for according to described cosine similarity value order from high to low, choosing the target face figure of preset number Decent；

Statistical module is for carrying out quantity statistics to the sex classification of described target face image pattern and quantity is most Sex classification is as the second target gender classification；

Described choosing module, being additionally operable to choose sex classification in described target face image pattern is described second target gender The similar face image sample of classification, and extract the cosine between described similar face image sample and described target facial image Similarity value is as target cosine similarity value；

Computing module, for calculating the meansigma methods of described target cosine similarity value；

Module is set, for using described second target gender classification and described meansigma methods as described cosine comparison result.