CN115909582A - Entrance guard equipment and system for face recognition of wearing mask - Google Patents
Entrance guard equipment and system for face recognition of wearing mask Download PDFInfo
- Publication number
- CN115909582A CN115909582A CN202211409397.0A CN202211409397A CN115909582A CN 115909582 A CN115909582 A CN 115909582A CN 202211409397 A CN202211409397 A CN 202211409397A CN 115909582 A CN115909582 A CN 115909582A
- Authority
- CN
- China
- Prior art keywords
- face
- detection image
- feature
- classification
- face detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The utility model relates to an intelligent recognition field, it specifically discloses a wear gauze mask face identification's entrance guard's equipment and system, it through with the face detection image through the gauze mask remover based on antagonism generation network in order to obtain behind the generation face detection image with the face detection image carries out polymerization in order to obtain multichannel face detection image, adopts the convolution neural network model based on feature extractor to extract image local region high dimension in the multichannel face detection image implies the characteristic and classifies, improves classification accuracy, and then improves the precision of wearing gauze mask face identification.
Description
Technical Field
The application relates to the field of intelligent recognition, and more particularly relates to an entrance guard equipment and system for face recognition of a wearing mask.
Background
The face recognition technology is applied to a plurality of fields such as public security, intelligent security, mobile phone privacy protection and the like at present. The entrance guard equipment with the face recognition function in daily life almost becomes the standard configuration of each business access place. However, due to some requirements, people need to wear masks when going out, the accuracy of the face recognition system is greatly reduced when people wear masks, and exposure risks exist if the masks are removed.
Therefore, an optimized entrance guard device and system for face recognition of a wearing mask are desired.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a wear gauze mask face identification's entrance guard's equipment, it through with the face detection image through the gauze mask remover based on the antagonism generates the network in order to obtain generate the face detection image after with the face detection image carries out polymerization in order to obtain multichannel face detection image, adopts the convolutional neural network model based on the feature extractor to extract image local region high-dimensional implicit feature in the multichannel face detection image is classified, improves classification accuracy, and then improves the precision of wearing gauze mask face identification.
According to an aspect of the present application, there is provided an entrance guard apparatus for face recognition of a wearer's mask, comprising:
an apparatus main body;
a camera disposed in the apparatus body; and
a processor disposed in the device body and communicably connected to the camera;
wherein the processor comprises:
the image receiving module is used for receiving a face detection image of the mask wearing object collected by the camera;
the mask removal generation module is used for enabling the face detection image to pass through a mask remover based on a confrontation generation network so as to obtain a generated face detection image;
the multi-channel image aggregation module is used for aggregating the face detection image and the generated face detection image along the channel dimension to obtain a multi-channel face detection image;
the feature extraction module is used for enabling the multi-channel face detection image to pass through a convolutional neural network model using a space attention mechanism to obtain a classification feature map;
the characteristic optimization module is used for carrying out feature clustering-based de-focusing fuzzy optimization on the classification characteristic graph to obtain an optimized classification characteristic graph; and
the identification module is used for enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of the mask wearing object; and
and the entrance guard control module is used for determining whether to generate an entrance guard opening instruction or not based on the classification result.
In the above entrance guard device with mask face recognition, the confrontation generation network includes an identifier and a generator, and the mask removal generation module is further configured to input the face detection image into the generator of the mask remover to obtain the generated face detection image.
In above-mentioned entrance guard's equipment of wearing gauze mask face identification, the feature extraction module includes: the depth convolution coding unit is used for inputting the multi-channel face detection image into the convolution neural network model to obtain a face feature map; the spatial attention unit is used for enabling the face feature map to pass through a spatial attention module to obtain a spatial attention map; and the attention applying unit is used for multiplying the spatial attention diagram and the face feature diagram by position points to obtain the classification feature diagram.
In the above entrance guard's equipment of wearing gauze mask face identification, the degree of depth convolution coding unit is further used for: using each layer of the convolutional neural network model to respectively perform the following steps on input data in forward transmission of the layer: carrying out convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the convolutional neural network is the face feature map, and the input of the first layer of the convolutional neural network is the multi-channel face detection image.
In the above entrance guard's equipment of wearing gauze mask face identification, the space attention unit is further used for: performing global mean pooling and global maximum pooling along a channel dimension on the face feature map to obtain a first face global feature matrix and a second face global feature matrix; cascading the first face global feature matrix and the second face global feature matrix and then passing through a convolutional layer to obtain an attention score map; and non-linearly activating the attention score map using a Softmax activation function to obtain the spatial attention map.
In the above-mentioned entrance guard's equipment of wearing gauze mask face identification, the feature optimization module is further used for: performing feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map;
wherein the formula is:
where μ and δ are the mean and standard deviation, respectively, of the feature set, where f i,j,k Is the feature value of the (i, j, k) th position of the classification feature map, f i,j,k E F is the feature set.
In the entrance guard equipment for face recognition of the mask, the expansion unit is used for expanding the optimized classification feature map into classification feature vectors based on row vectors or column vectors; a full-concatenation encoding unit, configured to perform full-concatenation encoding on the classification feature vector using a plurality of full-concatenation layers of the classifier to obtain an encoded classification feature vector; and the classification result generating unit is used for enabling the coded classification feature vector to pass through a Softmax classification function of the classifier to obtain the classification result.
In above-mentioned entrance guard's equipment of wearing gauze mask face identification, include: the image receiving module is used for receiving a face detection image of the mask wearing object collected by the camera; the mask removal generation module is used for enabling the face detection image to pass through a mask remover based on a countermeasure generation network to obtain a generated face detection image; the multi-channel image aggregation module is used for aggregating the face detection image and the generated face detection image along the channel dimension to obtain a multi-channel face detection image; the feature extraction module is used for enabling the multi-channel face detection image to pass through a convolutional neural network model using a space attention mechanism to obtain a classification feature map; the characteristic optimization module is used for carrying out feature clustering-based de-focusing fuzzy optimization on the classification characteristic graph to obtain an optimized classification characteristic graph; the identification module is used for enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of the mask wearing object; and the entrance guard control module is used for determining whether an entrance guard opening instruction is generated or not based on the classification result.
According to another aspect of the present application, there is provided a face recognition method for a respirator, including:
receiving a face detection image of a mask wearing object acquired by the camera;
passing the face detection image through a mask remover based on a countermeasure generation network to obtain a generated face detection image;
aggregating the face detection image and the generated face detection image along a channel dimension to obtain a multi-channel face detection image;
obtaining a classification feature map by passing the multi-channel face detection image through a convolutional neural network model using a spatial attention mechanism;
performing feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map; and
enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, wherein the classification result is used for representing an identity label of a mask wearing object; and
and determining whether to generate an entrance guard opening instruction or not based on the classification result.
Compared with the prior art, the entrance guard's equipment and system of wearing gauze mask face identification that this application provided, it through with face detection image through the gauze mask remover based on the antagonism network that generates with behind the face detection image that obtains generating, face detection image with face detection image gathers in order to obtain multichannel face detection image, adopts the convolution neural network model based on the feature extractor to extract image local region high-dimensional implicit feature in the multichannel face detection image is categorised, improves categorised precision, and then improves the precision of wearing gauze mask face identification.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is an application scene diagram of an access control device for face recognition of a mask according to an embodiment of the present application;
fig. 2A is a block diagram of an access control device for face recognition on a mask according to an embodiment of the present application;
fig. 2B is a block diagram of an access control device for face recognition of a mask according to an embodiment of the present application;
fig. 3 is a system architecture diagram of an access control device for face recognition of a mask according to an embodiment of the present application;
fig. 4 is a block diagram of a feature extraction module in an access control device for face recognition of a mask worn according to an embodiment of the present application;
fig. 5 is a flowchart of convolutional neural network coding in an access control device for face recognition of a mask worn according to an embodiment of the present application;
fig. 6 is a flowchart of space attention unit coding in an access control device for face recognition with a mask according to an embodiment of the present application;
fig. 7 is a block diagram of an identification module in an access control device for face recognition of a wearer mask according to an embodiment of the present application;
fig. 8 is a flowchart of a face recognition method for a wearer mask according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Overview of scenes
As described above, in daily life, an entrance guard apparatus having a face recognition function is a standard in almost every business place. However, due to some requirements, people need to wear masks when going out, the accuracy of the face recognition system is greatly reduced when people wear masks, and exposure risks exist if the masks are removed. Therefore, an optimized entrance guard device and system for face recognition of a wearer's mouth is desired.
Specifically, when the mask blocks most of the face, the face recognition system cannot extract complete facial features of the face, and can only capture partial information of the face, and if the face image stored in the system is a complete face without wearing the mask, the difference between the two images is large, and at this time, the face recognition fails.
In view of the above technical problems, in the technical solution of the present application, after a face detection image of a mask wearing object is collected by a camera, the face detection image is generated by a mask remover that generates a network based on confrontation. Here, the countermeasure generation network includes a discriminator and a generator, wherein the generator is configured to generate a mask removed face image, the discriminator is configured to measure a difference between the mask removed face image and a real face image, and the generator is subjected to a back propagation training based on gradient descent with a discriminator loss function value so that the mask removed face image generated by the generator is brought closer to the real face image. That is, in the technical solution of the present application, a complete face image is restored based on the countermeasure generation idea.
However, considering that the generated face detection image is not a real face image, the face detection image and the generated face detection image are further aggregated along the channel dimension to obtain a multi-channel face detection image. Namely, the source domain image and the generated image are aggregated along the channel dimension, so that the source domain information and the generated information are aggregated to expand the input of the network end so as to improve the accuracy of face recognition.
Then, the multi-channel face detection image is processed through a convolution neural network model using a spatial attention mechanism to obtain a classification feature map. That is, a convolutional neural network model is used as an image feature extractor to capture high-dimensional implicit features of local areas of images in the multi-channel face detection image. Considering that the influence weights of pixel points at various positions in the multi-channel face detection image on final face recognition judgment are different, in order to enable the characteristics of different pixel points in the multi-channel face detection image to have stronger identifiability, a spatial attention mechanism is integrated in the convolutional neural network model.
And after the classification feature map is obtained, the classification feature map is passed through a classifier to obtain a classification result of the identity label representing the mask wearing object. And then, determining whether to generate an entrance guard opening instruction or not based on the classification result. That is, if the identification tag of the mask wearing object belongs to the entered identification tag, an entrance guard opening instruction is generated, and if the identification tag of the mask wearing object does not belong to the entered identification tag, an intrusion object prompt is generated.
Particularly, in the technical scheme of the application, the original image of the wearing mask and the generated image of the mask removing mask are aggregated at a source domain end, and feature extraction and encoding are performed by using a convolutional neural network model of a spatial attention mechanism, so that the image semantic information of the original image of the wearing mask and the generated image of the mask removing mask can be fully utilized. However, at the same time, the feature information is enriched, and the clustering effect of the feature distribution of the classification feature map is also deteriorated, thereby affecting the classification effect of the classification feature map.
Here, the applicant of the present application considers that, in the classification feature map, the image semantic relation distribution extracted by the convolutional neural network model using the spatial attention mechanism exhibits a gaussian distribution in a natural state, that is, the average degree of correlation between the image semantic information of the original image with the mask worn and the generated image with the mask removed has the highest probability density, and the higher degree and the lower degree of correlation have lower probability densities. Therefore, based on the high-frequency distribution characteristics which follow the Gaussian point distribution, the classification characteristic diagram can be subjected to the focus-removing fuzzy optimization of characteristic clustering, which is represented as:
mu and delta are feature sets f, respectively i,j,k E mean and standard deviation of F, and F i,j,k Is the feature value of the (i, j, k) th position of the classification feature map F.
The focusing fuzzy optimization of the feature clustering compensates the dependency similarity of the high-frequency distribution features which follow the Gaussian point distribution relative to the uniform representation of the overall feature distribution by performing the feature clustering index based on the statistical information on the focusing stack representation for estimating the clustering metric value, thereby avoiding the focusing fuzzy of the overall feature distribution caused by the low dependency similarity, and thus, improving the classification effect of the classification feature map. That is, the accuracy of face recognition of the gauze mask is improved.
Based on this, this application has provided a wear gauze mask face identification's entrance guard's equipment, it includes: an apparatus main body; a camera disposed in the apparatus main body; and a processor disposed in the device body and communicably connected to the camera; wherein the processor comprises: the image receiving module is used for receiving a face detection image of the mask wearing object collected by the camera; the mask removal generation module is used for enabling the face detection image to pass through a mask remover based on a countermeasure generation network to obtain a generated face detection image; the multi-channel image aggregation module is used for aggregating the face detection image and the generated face detection image along the channel dimension to obtain a multi-channel face detection image; the feature extraction module is used for enabling the multi-channel face detection image to pass through a convolutional neural network model using a space attention mechanism to obtain a classification feature map; the characteristic optimization module is used for carrying out feature clustering-based de-focusing fuzzy optimization on the classification characteristic graph to obtain an optimized classification characteristic graph; the identification module is used for enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of the mask wearing object; and the entrance guard control module is used for determining whether to generate an entrance guard opening instruction or not based on the classification result.
Fig. 1 is an application scene diagram of an access control device for face recognition of a mask according to an embodiment of the present application. As shown in fig. 1, in the application scenario, the access control device (e.g., 10 as illustrated in fig. 1) includes a device body (e.g., 11 as illustrated in fig. 1), and a camera (e.g., 12 as illustrated in fig. 1) disposed in the device body; and a processor (e.g., 13 as illustrated in fig. 1) disposed at the apparatus main body and communicably connected with the camera. A face detection image (e.g., F as illustrated in fig. 1) of the mask wearing subject is acquired by a camera disposed in the apparatus main body. Then, the image is input into the processor (for example, 13 in fig. 1) deployed with a mask wearing face recognition algorithm, wherein the server can process the input image with the mask wearing face recognition algorithm to generate a classification result representing an identity tag of a mask wearing subject, and determine whether to generate an entrance guard opening instruction based on the classification result.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary System
Fig. 2A is a block diagram of an access control device for face recognition of a mask wearing the mask according to an embodiment of the present application. As shown in fig. 2A, the entrance guard device 300 for face recognition of a mask according to an embodiment of the present application includes: a device main body 310; a camera 320 disposed in the apparatus main body; and a processor 330 disposed in the apparatus main body and communicably connected to the camera.
Fig. 2B is a block diagram of an access control device for face recognition of a mask wearing the mask according to an embodiment of the present application. As shown in fig. 2B, according to the entrance guard device 300 for face recognition of a mask wearing an embodiment of the present application, the processor 330 includes: an image receiving module 331; a mask removal generation module 332; a multi-channel image aggregation module 333; a feature extraction module 334; a feature optimization module 335; and, an identification module 336; and an access control module 337.
The image receiving module 331 is configured to receive a face detection image of a mask wearing object collected by a camera; the mask removal generation module 332 is configured to pass the face detection image through a mask remover based on an adversarial generation network to obtain a generated face detection image; the multi-channel image aggregation module 333 is configured to aggregate the face detection image and the generated face detection image along a channel dimension to obtain a multi-channel face detection image; the feature extraction module 334 is configured to pass the multi-channel face detection image through a convolutional neural network model using a spatial attention mechanism to obtain a classification feature map; the feature optimization module 335 is configured to perform feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map; the identification module 336 is used for enabling the optimized classified feature map to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of a mask wearing object; and the access control module 337 is configured to determine whether to generate an access opening instruction based on the classification result.
Fig. 3 is a system architecture diagram of an access control device for face recognition of a wearing mask according to an embodiment of the present application. As shown in fig. 3, in the system architecture of the entrance guard device 300 for face recognition of a mask wearing person, firstly, a face detection image of a mask wearing person collected by the camera is received by the image receiving module 331; the mask removal generation module 332 passes the face detection image received by the image receiving module 331 through a mask remover based on a challenge generation network to obtain a generated face detection image; the multi-channel image aggregation module 333 aggregates the face detection image received by the image receiving module 331 and the generated face detection image obtained by the mask removing and generating module 332 along a channel dimension to obtain a multi-channel face detection image; then, the feature extraction module 334 obtains a classification feature map by passing the multi-channel face detection image obtained by the multi-channel image aggregation module 333 through a convolutional neural network model using a spatial attention mechanism; the feature optimization module 335 performs feature-clustering-based de-focusing fuzzy optimization on the classification feature map obtained by the feature extraction module 334 to obtain an optimized classification feature map; then, the recognition module 336 passes the optimized classification feature map obtained by the feature optimization module 335 through a classifier to obtain a classification result, where the classification result is used to represent an identity tag of the mask wearing subject; further, the access control module 337 determines whether to generate an access opening instruction based on the classification result.
Specifically, in the operation process of the entrance guard device 300 with mask face recognition, the image receiving module 331 and the mask removing generation module 332 are configured to receive the face detection image of the mask object collected by the camera, and generate the face detection image through a mask remover based on an confrontation generation network. When a mask shields most of the face, the face recognition system cannot extract complete facial features of the face and can only capture partial information of the face, and if the face picture stored in the system is a complete face without wearing the mask, the difference between the two pictures is large, and at this time, the face recognition fails. In order to solve the technical problem, according to the technical scheme, after a face detection image of a mask wearing object is collected by a camera, the face detection image is subjected to face detection image generation through a mask remover based on an confrontation generation network. Here, the countermeasure generation network includes a discriminator and a generator, wherein the generator is configured to generate a mask removed face image, the discriminator is configured to measure a difference between the mask removed face image and a real face image, and the generator is subjected to a back propagation training based on gradient descent with a discriminator loss function value so that the mask removed face image generated by the generator is brought closer to the real face image. That is, in the technical solution of the present application, a complete face image is restored based on the countermeasure generation idea.
Specifically, in the operation process of the entrance guard device 300 with mask face recognition, the multi-channel image aggregation module 333 is configured to aggregate the face detection image and the generated face detection image along a channel dimension to obtain a multi-channel face detection image. However, considering that the generated face detection image is not a real face image, the face detection image and the generated face detection image are further aggregated along the channel dimension to obtain a multi-channel face detection image. Namely, the source domain image and the generated image are aggregated along the channel dimension, so that the source domain information and the generated information are aggregated to expand the input of a network end to improve the accuracy of face recognition.
Specifically, in the operation process of the entrance guard equipment 300 with mask face recognition, the feature extraction module 334 is configured to obtain a classification feature map from the multi-channel face detection image through a convolutional neural network model using a spatial attention mechanism. In the technical scheme of the application, the multi-channel face detection image is subjected to a convolutional neural network model using a spatial attention mechanism to obtain a classification feature map. That is, a convolutional neural network model is used as an image feature extractor to capture high-dimensional implicit features of local areas of images in the multi-channel face detection image. Considering that the influence weights of pixel points at various positions in the multi-channel face detection image on the final face recognition judgment are different, in order to enable the characteristics of different pixel points in the multi-channel face detection image to have stronger identifiability, spatial attention is integrated in the convolutional neural network model. More specifically, the feature extraction module includes: the depth convolution coding unit is used for inputting the multi-channel face detection image into the convolution neural network model to obtain a face feature map; the spatial attention unit is used for enabling the face feature map to pass through a spatial attention module to obtain a spatial attention map; and the attention applying unit is used for multiplying the spatial attention diagram and the human face feature diagram by position points to obtain the classification feature diagram. Wherein the depth convolutional coding unit is further configured to: performing, using the layers of the convolutional neural network model, in a layer forward pass, input data separately: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the convolutional neural network is the face feature map, and the input of the first layer of the convolutional neural network is the multi-channel face detection image. The spatial attention unit is further configured to: performing global mean pooling and global maximum pooling along a channel dimension on the face feature map to obtain a first face global feature matrix and a second face global feature matrix; cascading the first face global feature matrix and the second face global feature matrix, and then obtaining an attention score map through a convolution layer; and non-linearly activating the attention score map using a Softmax activation function to obtain the spatial attention map.
Fig. 4 is a block diagram of a feature extraction module in an access control device for face recognition of a wearing mask according to an embodiment of the present application. As shown in fig. 4, the feature extraction module 334 includes: a depth convolution coding unit 3341, configured to input the multi-channel face detection image into the convolution neural network model to obtain a face feature map; a spatial attention unit 3342, configured to pass the facial feature map through a spatial attention module to obtain a spatial attention map; and an attention applying unit 3343, configured to multiply the spatial attention map and the face feature map by location point to obtain the classification feature map.
Fig. 5 is a flowchart of convolutional neural network coding in an access control device for face recognition of a wearing mask according to an embodiment of the present application. As shown in fig. 5, in the convolutional neural network coding process, the convolutional neural network coding method includes: s210, carrying out convolution processing on input data to obtain a convolution characteristic diagram; s220, performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and S230, carrying out nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the convolutional neural network is the face feature map, and the input of the first layer of the convolutional neural network is the multi-channel face detection image.
Fig. 6 is a flowchart of space attention unit coding in an access control device for face recognition of a wearer mask according to an embodiment of the present application. As shown in fig. 6, in the spatial attention unit coding process, the method includes: s310, performing global mean pooling and global maximum pooling along channel dimensions on the face feature map S320 to obtain a first face global feature matrix and a second face global feature matrix; s330, cascading the first human face global feature matrix and the second human face global feature matrix, and then obtaining an attention score map through a convolutional layer; and performing nonlinear activation on the attention score map by using a Softmax activation function to obtain the spatial attention map.
Specifically, in the operation process of the entrance guard equipment 300 with mask face recognition, the feature optimization module 335 is configured to perform feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map. Particularly, in the technical scheme of the application, the original image of the wearing mask and the generated image of the mask removing mask are aggregated at a source domain end, and feature extraction and encoding are performed by using a convolutional neural network model of a spatial attention mechanism, so that the image semantic information of the original image of the wearing mask and the generated image of the mask removing mask can be fully utilized. However, at the same time, the feature information is enriched, and the clustering effect of the feature distribution of the classification feature map is also deteriorated, thereby affecting the classification effect of the classification feature map.
Here, the applicant of the present application considers that, in the classification feature map, the image semantic relation distribution extracted by the convolutional neural network model using the spatial attention mechanism exhibits a gaussian distribution in a natural state, that is, the average degree of correlation between the image semantic information of the original image with the mask worn and the generated image with the mask removed has the highest probability density, and the higher degree and the lower degree of correlation have lower probability densities. Therefore, based on the high-frequency distribution characteristics which follow the Gaussian point distribution, the classification characteristic diagram can be subjected to the focus-removing fuzzy optimization of characteristic clustering, which is represented as:
mu and delta are feature sets f, respectively i,j,k E mean and standard deviation of F, and F i,j,k Is the feature value of the (i, j, k) th position of the classification feature map F.
The focusing fuzzy optimization of the feature clustering compensates the dependency similarity of the high-frequency distribution features following the Gaussian point distribution relative to the uniform representation of the overall feature distribution by performing feature clustering index based on statistical information on the focusing stack representation used for estimating the clustering metric value, thereby avoiding the focusing fuzzy of the overall feature distribution caused by low dependency similarity, and thus, improving the classification effect of the classification feature map. That is, the accuracy of face recognition of the gauze mask is improved.
Specifically, in the operation process of the entrance guard device 300 for face recognition of the mask wearing type, the recognition module 336 is configured to pass the optimized classification feature map through a classifier to obtain a classification result, where the classification result is used to represent an identity tag of a mask wearing type object. In a specific example of the present application, the identification module includes: the expansion unit is used for expanding the optimized classification feature map into classification feature vectors on the basis of row vectors or column vectors; a full-concatenation encoding unit, configured to perform full-concatenation encoding on the classification feature vector using a plurality of full-concatenation layers of the classifier to obtain an encoded classification feature vector; and the classification result generating unit is used for enabling the coded classification feature vector to pass through a Softmax classification function of the classifier to obtain the classification result. More specifically, the passing the optimized classification feature map through a classifier to obtain a classification result includes: processing the optimized classification feature map by using the classifier according to the following formula to obtain a classification result, wherein the formula is as follows:
O＝softmax{(W n ,B n ):…:(W 1 ,B 1 ) L Project (F) }, where Project (F) denotes projection of the optimized classification feature map as a vector, W 1 To W n As a weight matrix for each fully connected layer, B 1 To B n Representing the bias vectors of the fully connected layers of each layer.
Fig. 7 is a block diagram of an identification module in an entrance guard device for face recognition of a wearing mask according to an embodiment of the present application. As shown in fig. 7, the identification module 336 includes: an expansion unit 3361, configured to expand the optimized classification feature map into classification feature vectors based on row vectors or column vectors; a full-concatenation coding unit 3362, configured to perform full-concatenation coding on the classification feature vector using a plurality of full-concatenation layers of the classifier to obtain a coded classification feature vector; and a classification result generating unit 3363, configured to pass the encoded classification feature vector through a Softmax classification function of the classifier to obtain the classification result.
Specifically, in the operation process of the entrance guard device 300 with the face recognition function, the entrance guard control module 337 is configured to determine whether to generate an entrance guard opening instruction based on the classification result. In a specific example of the present application, it is further determined whether to generate an entrance guard opening instruction based on the classification result. That is, if the identification tag of the mask wearing object belongs to the entered identification tag, an entrance guard opening instruction is generated, and if the identification tag of the mask wearing object does not belong to the entered identification tag, an intrusion object prompt is generated.
To sum up, wear gauze mask face identification's entrance guard's equipment 300 according to this application embodiment is elucidated, its through with face detection image through the gauze mask remover based on the antagonism network that generates with behind the face detection image of obtaining generation with face detection image carries out polymerization in order to obtain multichannel face detection image, adopts the convolutional neural network model based on the feature extractor to extract image local region high dimension in the multichannel face detection image implies the characteristic and classifies, improves classification accuracy, and then improves the precision of wearing gauze mask face identification.
As described above, the entrance guard device for face recognition of a mask wearing part according to the embodiment of the present application can be implemented in various terminal devices. In one example, the entrance guard device 300 for face recognition of a wearing mask according to the embodiment of the present application may be integrated into a terminal device as a software module and/or a hardware module. For example, the door access device 300 for face recognition of the mask may be a software module in an operating system of the terminal device, or may be an application developed for the terminal device; of course, the degassed hydrogen conductivity measurement device 300 may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the gauze mask face recognition entrance guard device 300 and the terminal device may be separate devices, and the gauze mask face recognition entrance guard device 300 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information according to an agreed data format.
Exemplary method
Fig. 8 is a flowchart of a face recognition method for a mask according to an embodiment of the present application. As shown in fig. 8, the method for recognizing a face with a mask according to the embodiment of the present application includes the steps of: s110, receiving a face detection image of the mask wearing object collected by the camera; s120, enabling the face detection image to pass through a mask remover based on an antagonistic generation network to obtain a generated face detection image; s130, the face detection image and the generated face detection image are aggregated along the channel dimension to obtain a multi-channel face detection image; s140, obtaining a classification characteristic image of the multi-channel face detection image through a convolutional neural network model using a space attention mechanism; s150, performing feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map; s160, the optimized classification feature map is processed by a classifier to obtain a classification result, and the classification result is used for representing the identity label of the mask wearing object; and S170, determining whether to generate an entrance guard opening instruction or not based on the classification result.
In one example, in the above method for recognizing a wearer' S face, the step S140 includes: inputting the multi-channel face detection image into the convolutional neural network model to obtain a face feature map; passing the face feature map through a spatial attention module to obtain a spatial attention map; and multiplying the spatial attention diagram and the face feature diagram according to position points to obtain the classification feature diagram. Wherein, the inputting the multi-channel face detection image into the convolution neural network model to obtain the face feature map comprises: performing, using the layers of the convolutional neural network model, in a layer forward pass, input data separately: performing convolution processing on input data to obtain a convolution characteristic diagram; performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the convolutional neural network is the face feature map, and the input of the first layer of the convolutional neural network is the multi-channel face detection image. The human face feature map is processed by a spatial attention module to obtain a spatial attention map, which comprises the following steps: performing global mean pooling and global maximum pooling along a channel dimension on the face feature map to obtain a first face global feature matrix and a second face global feature matrix; cascading the first face global feature matrix and the second face global feature matrix, and then obtaining an attention score map through a convolution layer; and performing nonlinear activation on the attention score map by using a Softmax activation function to obtain the spatial attention map.
In one example, in the above method for recognizing a wearer' S face, the step S150 includes: performing feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map;
wherein the formula is:
where μ and δ are the mean and standard deviation, respectively, of the feature set, where f i,j,k Is the feature value of the (i, j, k) th position of the classification feature map, f ,j,k E F is the feature set.
In one example, in the above wearing mask face recognition method, the step S160 includes: expanding the optimized classification feature map into classification feature vectors based on row vectors or column vectors; performing full-join encoding on the classification feature vectors using a plurality of full-join layers of the classifier to obtain encoded classification feature vectors; and passing the encoding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.
In summary, the mask wearing face recognition method according to the embodiment of the application is clarified, the generated face detection image is obtained through the mask remover based on the countermeasure generation network, then the generated face detection image and the face detection image are aggregated to obtain a multi-channel face detection image, a convolutional neural network model based on a feature extractor is adopted to extract high-dimensional implicit features of image local areas in the multi-channel face detection image for classification, the classification precision is improved, and the mask wearing face recognition precision is further improved.
Claims (8)
1. The utility model provides a wear gauze mask face identification's entrance guard's equipment which characterized in that includes:
an apparatus main body;
a camera disposed in the apparatus body; and
a processor disposed in the device body and communicably connected to the camera;
wherein the processor comprises:
the image receiving module is used for receiving a face detection image of the mask wearing object collected by the camera;
the mask removal generation module is used for enabling the face detection image to pass through a mask remover based on a confrontation generation network so as to obtain a generated face detection image;
the multi-channel image aggregation module is used for aggregating the face detection image and the generated face detection image along the channel dimension to obtain a multi-channel face detection image;
the feature extraction module is used for enabling the multi-channel face detection image to pass through a convolutional neural network model using a space attention mechanism to obtain a classification feature map;
the characteristic optimization module is used for carrying out focusing-removing fuzzy optimization based on characteristic clustering on the classification characteristic diagram to obtain an optimized classification characteristic diagram; and
the identification module is used for enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of the mask wearing object; and
and the entrance guard control module is used for determining whether to generate an entrance guard opening instruction or not based on the classification result.
2. The door access device for face recognition on a mask according to claim 1, wherein the countermeasure generation network comprises a discriminator and a generator, and the mask removal generation module is further configured to input the face detection image into the generator of the mask remover to obtain the generated face detection image.
3. The entrance guard's equipment of wear gauze mask face identification of claim 2, characterized in that, the feature extraction module includes:
the depth convolution coding unit is used for inputting the multi-channel face detection image into the convolution neural network model to obtain a face feature map;
the spatial attention unit is used for enabling the face feature map to pass through a spatial attention module to obtain a spatial attention map; and
and the attention applying unit is used for multiplying the spatial attention diagram and the human face feature diagram by position points to obtain the classification feature diagram.
4. The entrance guard's equipment of wear gauze mask face identification of claim 3, characterized in that, the depth convolution coding unit, further is used for: using each layer of the convolutional neural network model to respectively perform the following steps on input data in forward transmission of the layer:
carrying out convolution processing on input data to obtain a convolution characteristic diagram;
performing pooling processing based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and
performing nonlinear activation on the pooled feature map to obtain an activated feature map;
and the output of the last layer of the convolutional neural network is the face feature map, and the input of the first layer of the convolutional neural network is the multi-channel face detection image.
5. The entrance guard's equipment of wear gauze mask face identification of claim 4, characterized in that, the space attention unit is further used for:
performing global mean pooling and global maximum pooling along a channel dimension on the face feature map to obtain a first face global feature matrix and a second face global feature matrix;
cascading the first face global feature matrix and the second face global feature matrix, and then obtaining an attention score map through a convolution layer; and
nonlinearly activating the attention score map using a Softmax activation function to obtain the spatial attention map.
6. The entrance guard's equipment of wear gauze mask face identification of claim 5, characterized in that, the characteristic optimization module is further used for: performing feature clustering-based de-focusing fuzzy optimization on the classification feature map to obtain an optimized classification feature map;
wherein the formula is:
where μ and δ are the mean and standard deviation, respectively, of the feature set, where f i,j,k Is the feature value of the (i, j, k) th position of the classification feature map, f i,j,k E F is the feature set.
7. The entrance guard's equipment of wear gauze mask face identification of claim 6, characterized in that, the identification module includes:
the expansion unit is used for expanding the optimized classification feature map into classification feature vectors on the basis of row vectors or column vectors;
a full-concatenation encoding unit, configured to perform full-concatenation encoding on the classification feature vector using a plurality of full-concatenation layers of the classifier to obtain an encoded classification feature vector; and
a classification result generating unit, configured to pass the encoded classification feature vector through a Softmax classification function of the classifier to obtain the classification result.
8. A face recognition system for a respirator, comprising:
the image receiving module is used for receiving a face detection image of the mask wearing object collected by the camera;
the mask removal generation module is used for enabling the face detection image to pass through a mask remover based on a countermeasure generation network to obtain a generated face detection image;
the multi-channel image aggregation module is used for aggregating the face detection image and the generated face detection image along the channel dimension to obtain a multi-channel face detection image;
the feature extraction module is used for enabling the multi-channel face detection image to pass through a convolutional neural network model using a space attention mechanism to obtain a classification feature map;
the characteristic optimization module is used for carrying out feature clustering-based de-focusing fuzzy optimization on the classification characteristic graph to obtain an optimized classification characteristic graph; and
the identification module is used for enabling the optimized classification characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for representing an identity label of the mask wearing object; and
and the entrance guard control module is used for determining whether to generate an entrance guard opening instruction or not based on the classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211409397.0A CN115909582A (en) | 2022-11-11 | 2022-11-11 | Entrance guard equipment and system for face recognition of wearing mask |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211409397.0A CN115909582A (en) | 2022-11-11 | 2022-11-11 | Entrance guard equipment and system for face recognition of wearing mask |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115909582A true CN115909582A (en) | 2023-04-04 |
Family
ID=86494929
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211409397.0A Withdrawn CN115909582A (en) | 2022-11-11 | 2022-11-11 | Entrance guard equipment and system for face recognition of wearing mask |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115909582A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116363738A (en) * | 2023-06-01 | 2023-06-30 | 成都睿瞳科技有限责任公司 | Face recognition method, system and storage medium based on multiple moving targets |
| CN116563795A (en) * | 2023-05-30 | 2023-08-08 | 北京天翊文化传媒有限公司 | Doll production management method and doll production management system |
| CN116580444A (en) * | 2023-07-14 | 2023-08-11 | 广州思林杰科技股份有限公司 | Method and equipment for testing long-distance running timing based on multi-antenna radio frequency identification technology |
| CN116664961A (en) * | 2023-07-31 | 2023-08-29 | 东莞市将为防伪科技有限公司 | Intelligent identification method and system for anti-counterfeit label based on signal code |
-
2022
- 2022-11-11 CN CN202211409397.0A patent/CN115909582A/en not_active Withdrawn
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116563795A (en) * | 2023-05-30 | 2023-08-08 | 北京天翊文化传媒有限公司 | Doll production management method and doll production management system |
| CN116363738A (en) * | 2023-06-01 | 2023-06-30 | 成都睿瞳科技有限责任公司 | Face recognition method, system and storage medium based on multiple moving targets |
| CN116580444A (en) * | 2023-07-14 | 2023-08-11 | 广州思林杰科技股份有限公司 | Method and equipment for testing long-distance running timing based on multi-antenna radio frequency identification technology |
| CN116664961A (en) * | 2023-07-31 | 2023-08-29 | 东莞市将为防伪科技有限公司 | Intelligent identification method and system for anti-counterfeit label based on signal code |
| CN116664961B (en) * | 2023-07-31 | 2023-12-05 | 东莞市将为防伪科技有限公司 | Intelligent identification method and system for anti-counterfeit label based on signal code |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115909582A (en) | Entrance guard equipment and system for face recognition of wearing mask | |
| Liu et al. | Recent advances of image steganography with generative adversarial networks | |
| CN110569731B (en) | Face recognition method and device and electronic equipment | |
| CN108537743A (en) | A kind of face-image Enhancement Method based on generation confrontation network | |
| CN110188829B (en) | Neural network training method, target recognition method and related products | |
| CN111523668A (en) | Training method and device of data generation system based on differential privacy | |
| Schmid et al. | Performance analysis of iris-based identification system at the matching score level | |
| CN112801054B (en) | Face recognition model processing method, face recognition method and device | |
| Inzunza-González et al. | Double hyperchaotic encryption for security in biometric systems | |
| CN113705290A (en) | Image processing method, image processing device, computer equipment and storage medium | |
| CN108564061B (en) | Image identification method and system based on two-dimensional pivot analysis | |
| Huang et al. | Exploring feature coupling and model coupling for image source identification | |
| CN115439809A (en) | Subway people stream density real-time monitoring system and method based on digital twins | |
| CN114036553A (en) | K-anonymity-combined pedestrian identity privacy protection method | |
| Mohamed et al. | Automated face recogntion system: Multi-input databases | |
| CN112528077B (en) | Video face retrieval method and system based on video embedding | |
| CN116543419B (en) | Hotel health personnel wearing detection method and system based on embedded platform | |
| CN118053232A (en) | Enterprise safety intelligent management system and method thereof | |
| KR100876786B1 (en) | System and method for verifying user's face using light masks | |
| CN111737688A (en) | Attack defense system based on user portrait | |
| Kimura et al. | Quality-dependent score-level fusion of face, gait, and the height biometrics | |
| CN115775404A (en) | Training method, device, equipment and storage medium of face recognition model | |
| CN111506691B (en) | Track matching method and system based on depth matching model | |
| CN114882582A (en) | Gait recognition model training method and system based on federal learning mode | |
| CN110956098B (en) | Image processing method and related equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20230404 |