Face recognition method and system based on big data, terminal device and storage medium
By employing multi-region preprocessing and an original neural network model, the problem of low accuracy in face recognition under conditions of sample imbalance and significant scene differences has been solved. This has enabled fast and accurate face recognition while reducing the number of model parameters, making it suitable for terminal devices.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHENGDU MENGMA XINGYUAN TECH CO LTD
- Filing Date
- 2023-05-08
- Publication Date
- 2026-06-02
AI Technical Summary
In applications such as identity verification, existing face recognition methods perform poorly when there is an imbalance in the number of sample images and significant differences in the scene. Furthermore, deep models increase memory requirements, which affects accuracy.
By expanding face image samples through multi-region preprocessing, adopting an original neural network model, reducing model parameters, and combining multi-scale feature stitching and multi-class convolutional layer feature extraction, the model's generalization performance and accuracy are improved.
It achieves fast and accurate face recognition in terminal devices, reduces the number of model parameters and improves recognition accuracy, and is suitable for deployment on terminal devices.
Smart Images
Figure CN116469149B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a face recognition method, system, terminal device, and storage medium based on big data, belonging to the field of face recognition technology. Background Technology
[0002] With the development of technologies such as big data and deep learning, the effectiveness of facial recognition has been significantly improved, and it has been widely used in fields such as identity authentication and video surveillance. However, in certain specific facial recognition scenarios, such as identity verification, when the number of sample images is severely imbalanced or the scene differences are significant, the facial recognition performance and accuracy of existing technologies can be affected. Furthermore, to improve the accuracy of facial recognition models, deeper and more complex models are often required, which also increases the model's memory requirements. Summary of the Invention
[0003] In view of this, the present invention provides a face recognition method, system, terminal device, and storage medium based on big data. It addresses the characteristics of large-scale face data by performing multi-region preprocessing to expand face image samples and improve the generalization performance of the neural network model. Furthermore, by using an original model, the number of model parameters is reduced, and the model's accuracy is improved, thereby achieving rapid face image recognition and facilitating deployment in terminal devices.
[0004] The first objective of this invention is to provide a face recognition method based on big data.
[0005] The second objective of this invention is to provide a face recognition system based on big data.
[0006] A third objective of this invention is to provide a terminal device.
[0007] A fourth objective of this invention is to provide a storage medium.
[0008] The first objective of this invention can be achieved by adopting the following technical solution:
[0009] A face recognition method based on big data, the method comprising:
[0010] Obtain a first dataset, which includes multiple different face images;
[0011] For each face image, the image is divided according to preset rules and a face key point recognition model to obtain a first region image, a second region image and a third region image. The first region image, the second region image and the third region image are processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0012] The first and second datasets are input into the neural network model for training to obtain the face recognition model;
[0013] Acquire the image of the face to be identified;
[0014] The face image to be identified is input into the face recognition model to obtain the recognition result.
[0015] In one possible embodiment, the neural network model includes an input layer, a first convolutional module, a second convolutional module, a third convolutional module, a first adaptive average pooling layer, and a first... flatten Layer, First dropout Layer, first fully connected module, second adaptive average pooling layer, second flatten Layer, Second dropout The system consists of a first layer, a second fully connected module, a feature fusion layer, and an output layer.
[0016] The input layer is connected to the first convolutional module and the second convolutional module, respectively.
[0017] The first convolutional module, the first adaptive average pooling layer, and the first flatten Layer, First dropout The first layer and the first fully connected module are connected in sequence;
[0018] The second convolutional module, the third convolutional module, the second adaptive average pooling layer, and the second... flatten Layer, Second dropout The first layer and the second fully connected module are connected in sequence;
[0019] The first fully connected module and the second fully connected module are respectively connected to the feature fusion layer;
[0020] The feature fusion layer is connected to the output layer.
[0021] In one possible embodiment, the first convolutional module includes a first input sub-layer, a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, a fourth 1×1 convolutional layer, a first max pooling layer, and a first output sub-layer.
[0022] The first input sub-layer, the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are connected in sequence;
[0023] The first input sub-layer, the fourth 1×1 convolutional layer, and the first max pooling layer are connected;
[0024] The first output sub-layer is used to output the feature matrix output by the concatenated third 1×1 convolutional layer and the feature matrix output by the first max pooling layer;
[0025] The first 1×1 convolutional layer and the second 1×1 convolutional layer include a 2x upsampling operation;
[0026] The second 1×1 convolutional layer and the third 1×1 convolutional layer include a 2x upsampling operation.
[0027] In one possible embodiment, the second convolutional module includes a second input sub-layer, a fifth 1×1 convolutional layer, a first 2×2 convolutional layer, a second 2×2 convolutional layer, a third 2×2 convolutional layer, a first 3×3 convolutional layer, and a second output sub-layer.
[0028] The second input sub-layer, the fifth 1×1 convolutional layer, and the first 2×2 convolutional layer are connected in sequence;
[0029] The second input sub-layer, the second 2×2 convolutional layer, and the third 2×2 convolutional layer are connected in sequence;
[0030] The second input sub-layer and the first 3×3 convolutional layer are connected;
[0031] The second output sub-layer is used to output the feature matrix output by the first 2×2 convolutional layer, the feature matrix output by the third 2×2 convolutional layer, the feature matrix output by the first 3×3 convolutional layer, and the feature matrix output by the fifth 1×1 convolutional layer after being concatenated.
[0032] In one possible embodiment, the third convolutional module includes a third input sub-layer, a second 3×3 convolutional layer, a third 3×3 convolutional layer, a fourth 3×3 convolutional layer, a fifth 3×3 convolutional layer, a sixth 1×1 convolutional layer, a seventh 1×1 convolutional layer, a second max pooling layer, and a third output sub-layer.
[0033] The third input sub-layer, the second 3×3 convolutional layer, the third 3×3 convolutional layer, the fourth 3×3 convolutional layer, and the fifth 3×3 convolutional layer are connected in sequence;
[0034] The third input sub-layer, the sixth 1×1 convolutional layer, and the second max pooling layer are connected in sequence.
[0035] The third input sub-layer and the seventh 1×1 convolutional layer are connected;
[0036] The third output sub-layer is used to output the feature matrix output by the fifth 3×3 convolutional layer, the feature matrix output by the second max pooling layer, the feature matrix output by the seventh 1×1 convolutional layer, the feature matrix output by the second 3×3 convolutional layer, the feature matrix output by the third 3×3 convolutional layer, and the feature matrix output by the fourth 3×3 convolutional layer after being spliced.
[0037] In one possible embodiment, the step of dividing the image according to preset rules and a facial landmark recognition model to obtain a first region image, a second region image, and a third region image includes:
[0038] Based on the facial landmark recognition model, the jaw point, the two temple points, and multiple auricle points of the two ears in the image are identified.
[0039] The triangular region formed by the mandibular point and the two temple points is taken as the first region image;
[0040] Based on both ears, the ear region enclosed by multiple auricular points is respectively used as the second region image and the third region image.
[0041] In one possible embodiment, processing the first region image, the second region image, and the third region image according to multiple image interference methods includes:
[0042] For the first region of the first region image, noise processing and / or compression processing are performed;
[0043] For the second region of the first region image, image obfuscation processing and / or random cropping processing are performed;
[0044] For the second and third region images, image obfuscation and / or random cropping are performed.
[0045] The second objective of this invention can be achieved by adopting the following technical solution:
[0046] A face recognition system based on big data, the system comprising:
[0047] The first acquisition unit is used to acquire a first dataset, which includes multiple different face images;
[0048] The second acquisition unit is used to divide each face image into a first region image, a second region image, and a third region image according to preset rules and a face key point recognition model. The first region image, the second region image, and the third region image are then processed according to various image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0049] The training unit is used to input the first and second datasets into the neural network model for training, thereby obtaining the face recognition model.
[0050] The third acquisition unit is used to acquire the face image to be identified;
[0051] The recognition unit is used to input the face image to be recognized into the face recognition model and obtain the recognition result.
[0052] The third objective of this invention can be achieved by adopting the following technical solution:
[0053] A terminal device includes a processor and a memory for storing processor-executable programs. When the processor executes the program stored in the memory, it implements the above-described face recognition method based on big data.
[0054] The fourth objective of this invention can be achieved by adopting the following technical solution:
[0055] A storage medium storing a program that, when executed by a processor, implements the aforementioned face recognition method based on big data.
[0056] The present invention has the following advantages over the prior art:
[0057] 1. This invention, tailored to the characteristics of large-scale facial data, performs multi-region preprocessing to expand facial image samples and improve the generalization performance of the neural network model. Furthermore, by using an original model, the number of model parameters is reduced, and the model's accuracy is improved, thereby achieving rapid facial image recognition and facilitating deployment on terminal devices.
[0058] 2. The face recognition model provided in this embodiment of the invention adopts a method of multi-scale feature splicing and multi-class convolutional layer joint feature extraction, which works together on the image to extract features at different levels, thereby increasing the expressive power of the model and improving the accuracy of the model. Attached Figure Description
[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.
[0060] Figure 1 This is a flowchart of the face recognition method based on big data according to Embodiment 1 of the present invention.
[0061] Figure 2 This is a division diagram of the first region image in Embodiment 1 of the present invention.
[0062] Figure 3 This is a structural diagram of the neural network model of Embodiment 1 of the present invention.
[0063] Figure 4 This is a structural diagram of the first convolution module in Embodiment 1 of the present invention.
[0064] Figure 5 This is a structural diagram of the second convolution module in Embodiment 1 of the present invention.
[0065] Figure 6This is a structural diagram of the third convolution module in Embodiment 1 of the present invention.
[0066] Figure 7 This is a structural block diagram of the face recognition system based on big data according to Embodiment 2 of the present invention.
[0067] Figure 8 This is a structural block diagram of the terminal device according to Embodiment 3 of the present invention. Detailed Implementation
[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0069] In the specification and claims of this application, the terms "first," "second," etc., are used to distinguish similar objects and not to describe a specified order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, not limited in number; for example, a first object can be one or more. Furthermore, in the specification and claims, "and" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.
[0070] Example 1:
[0071] like Figure 1 As shown in the figure, this embodiment provides a face recognition method based on big data, which includes the following steps:
[0072] S100. Obtain the first dataset, which includes multiple different face images.
[0073] Prior to S100, the following were included:
[0074] 1. Obtain facial images from big data on identity verification. However, the categories of these facial images are relatively simple. There is usually only one facial image for each person. Even if there are multiple facial images, the differences between the images are not significant.
[0075] 2. Set the face image size to 112x112 pixels.
[0076] 3. Use the pre-set multiple different face images as the first dataset.
[0077] S200. For each face image, the image is divided according to preset rules and a face key point recognition model to obtain a first region image, a second region image, and a third region image. The first region image, the second region image, and the third region image are processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0078] In one possible embodiment, the step of dividing the image according to preset rules and a facial landmark recognition model to obtain a first region image, a second region image, and a third region image includes:
[0079] S211. Based on the facial key point recognition model, identify the chin point, two temple points, and multiple auricle points of the two ears in the image.
[0080] In S211, the facial landmark recognition model is based on a convolutional neural network (CNN).
[0081] S212. The triangular region formed by the mandibular point and the two temple points is taken as the first region image.
[0082] S213. Based on the two ears, the ear region enclosed by the multiple auricle points is respectively used as the second region image and the third region image.
[0083] S213 specifically involves extracting pixels from the ear region based on multiple auricle points to form a binarized ear mask. For each ear, the mask is applied to the original image to obtain an image of that ear region. The image of the left ear region is used as the second region image, and the image of the right ear region is used as the third region image.
[0084] In one possible embodiment, processing the first region image, the second region image, and the third region image according to multiple image interference methods includes:
[0085] S221. For the first region of the first region image, perform noise processing and / or compression processing.
[0086] In S221, noise processing includes at least one of the following: Gaussian noise processing, salt-and-pepper noise processing, and speckle noise processing. In practical applications, generally only one type of noise processing is performed, and then it is merged with the second region described below. At this time, the first region can undergo different noise processing, thus obtaining more image sample augmentation data. The division of the first region 10 and the second region 20 is referenced. Figure 2 .
[0087] S222. For the second region of the first region image, perform image obfuscation processing and / or random cropping processing.
[0088] S223. Perform image obfuscation processing and / or random cropping processing on the second region image and the third region image.
[0089] The reason for performing S222 and S223 is that these areas are easily occluded and missing, so that more image samples can be obtained to expand the data.
[0090] S300. Input the first and second datasets into the neural network model for training to obtain the face recognition model.
[0091] In one possible embodiment, such as Figure 3 As shown, the neural network model includes an input layer, a first convolutional module, a second convolutional module, a third convolutional module, a first adaptive average pooling layer, and a first... flatten Layer, First dropout Layer, first fully connected module, second adaptive average pooling layer, second flatten Layer, Second dropout The system consists of a first layer, a second fully connected module, a feature fusion layer, and an output layer.
[0092] The input layer is connected to the first convolutional module and the second convolutional module respectively; the first convolutional module, the first adaptive average pooling layer, and the first... flatten Layer, First dropout The first fully connected module and the second convolutional module are connected in sequence; the second convolutional module, the third convolutional module, the second adaptive average pooling layer, and the second... flatten Layer, Second dropout The first fully connected module and the second fully connected module are connected in sequence; the first fully connected module and the second fully connected module are respectively connected to the feature fusion layer; the feature fusion layer is connected to the output layer.
[0093] In the model, adaptive average pooling layer and dropout Layers can reduce the number of parameters and computational load to some extent, thereby improving the model's generalization ability and efficiency.
[0094] In one possible embodiment, such as Figure 4 As shown, the first convolutional module includes a first input sub-layer, a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, a fourth 1×1 convolutional layer, a first max pooling layer, and a first output sub-layer.
[0095] The first input sub-layer, the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are connected in sequence; the first input sub-layer, the fourth 1×1 convolutional layer, and the first max pooling layer are connected; the first output sub-layer is used to output the feature matrix output by the concatenated third 1×1 convolutional layer and the feature matrix output by the first max pooling layer; the first 1×1 convolutional layer and the second 1×1 convolutional layer include a 2x upsampling operation; the second 1×1 convolutional layer and the third 1×1 convolutional layer include a 2x upsampling operation.
[0096] In one possible embodiment, such as Figure 5 As shown, the second convolutional module includes a second input sub-layer, a fifth 1×1 convolutional layer, a first 2×2 convolutional layer, a second 2×2 convolutional layer, a third 2×2 convolutional layer, a first 3×3 convolutional layer, and a second output sub-layer.
[0097] The second input sub-layer, the fifth 1×1 convolutional layer, and the first 2×2 convolutional layer are connected in sequence; the second input sub-layer, the second 2×2 convolutional layer, and the third 2×2 convolutional layer are connected in sequence; the second input sub-layer and the first 3×3 convolutional layer are connected; the second output sub-layer is used to output the concatenated feature matrices output by the first 2×2 convolutional layer, the third 2×2 convolutional layer, the first 3×3 convolutional layer, and the fifth 1×1 convolutional layer.
[0098] In one possible embodiment, such as Figure 6 As shown, the third convolutional module includes a third input sub-layer, a second 3×3 convolutional layer, a third 3×3 convolutional layer, a fourth 3×3 convolutional layer, a fifth 3×3 convolutional layer, a sixth 1×1 convolutional layer, a seventh 1×1 convolutional layer, a second max pooling layer, and a third output sub-layer.
[0099] The third input sub-layer, the second 3×3 convolutional layer, the third 3×3 convolutional layer, the fourth 3×3 convolutional layer, and the fifth 3×3 convolutional layer are connected in sequence; the third input sub-layer, the sixth 1×1 convolutional layer, and the second max pooling layer are connected in sequence; the third input sub-layer and the seventh 1×1 convolutional layer are connected; the third output sub-layer is used to output the concatenated feature matrices output by the fifth 3×3 convolutional layer, the second max pooling layer, the seventh 1×1 convolutional layer, the second 3×3 convolutional layer, the third 3×3 convolutional layer, and the fourth 3×3 convolutional layer.
[0100] For example, the input layer has an input size of 112x112x3, where 3 represents the three RGB channels.
[0101] The first convolutional module has four output channels: a first 1x1 convolutional layer with 64 output channels, a second 1x1 convolutional layer with 64 output channels, a third 1x1 convolutional layer with 128 output channels, and a fourth 1x1 convolutional layer with 128 output channels. The first max-pooling layer has a pooling size of 2x2, a stride of 2, and uses valid padding. In this first convolutional module, no stride is set; instead, the size of the output data is controlled by the first max-pooling layer. This approach reduces the number of model parameters while preserving important data features, thereby improving model performance.
[0102] The second convolutional module has 128 output channels for the fifth 1x1 convolutional layer, 256 output channels for the first 2x2 convolutional layer, 256 output channels for the second 2x2 convolutional layer, 256 output channels for the third 2x2 convolutional layer, and 512 output channels for the first 3x3 convolutional layer.
[0103] The third convolutional module has 512 output channels for the second, third, fourth, and fifth 3x3 convolutional layers, 512 output channels for the sixth, and 1024 output channels for the seventh 1x1 convolutional layer. The pooling size of the second max pooling layer is 2x2.
[0104] In the second and third convolutional modules, the stride of the convolutional layers is 1 or 2, and the padding method is the same.
[0105] In this process, an activation function and a batch normalization layer need to be added after each convolutional layer to improve the performance and stability of the model. The activation function is the ReLU function.
[0106] The first fully connected module contains a fully connected layer with 1024 output channels. After the fully connected layer, an activation function and a batch normalization layer need to be added to improve the performance and stability of the model. The activation function is the ReLU function.
[0107] The second fully connected module contains a fully connected layer with 1024 output channels. After the fully connected layer, an activation function and a batch normalization layer need to be added to improve the performance and stability of the model. The activation function is the ReLU function.
[0108] Feature fusion layer: fuses the features of the first fully connected module and the second fully connected module, with an output channel count of 2048.
[0109] Output layer: The number of output channels is the number of face identifications, which can be set according to specific circumstances.
[0110] In one possible embodiment, the cross-entropy loss function and the Adam optimizer can be selected when training the model.
[0111] S400: Obtain the image of the face to be identified.
[0112] S500: Input the face image to be recognized into the face recognition model to obtain the recognition result.
[0113] Under the same dataset and training method, the number of parameters of the face recognition model in this application is 15.52% of the number of parameters of the DeepFace neural network model and 17.30% of the number of parameters of the VGGFace neural network model.
[0114] Under the same dataset and training method, the face recognition model of this application has a maximum face recognition accuracy of 99.32%, which is 2.91% higher than the neural network model DeepFace and 0.5% higher than the neural network model VGGFace.
[0115] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.
[0116] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0117] Example 2:
[0118] like Figure 7 As shown, this embodiment provides a face recognition system based on big data. The system includes a first acquisition unit 701, a second acquisition unit 702, a training unit 703, a third acquisition unit 704, and a recognition unit 705. The specific functions of each unit are as follows:
[0119] The first acquisition unit 701 is used to acquire a first dataset, which includes multiple different face images;
[0120] The second acquisition unit 702 is used to divide each face image into a first region image, a second region image, and a third region image according to preset rules and a face key point recognition model. The first region image, the second region image, and the third region image are then processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0121] Training unit 703 is used to input the first dataset and the second dataset into the neural network model for training to obtain a face recognition model;
[0122] The third acquisition unit 704 is used to acquire the face image to be identified;
[0123] The recognition unit 705 is used to input the face image to be recognized into the face recognition model to obtain the recognition result.
[0124] Example 3:
[0125] like Figure 8 As shown, this embodiment provides a terminal device, which includes a processor 802, a memory, an input device 803, a display device 804, and a network interface 805 connected via a system bus 701. The processor 802 provides computing and control capabilities. The memory includes a non-volatile storage medium 806 and internal memory 807. The non-volatile storage medium 806 stores an operating system, computer programs, and a database. The internal memory 807 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium 806. When the computer program is executed by the processor 802, it implements the big data-based face recognition method of Embodiment 1 described above, as follows:
[0126] Obtain a first dataset, which includes multiple different face images;
[0127] For each face image, the image is divided according to preset rules and a face key point recognition model to obtain a first region image, a second region image and a third region image. The first region image, the second region image and the third region image are processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0128] The first and second datasets are input into the neural network model for training to obtain the face recognition model;
[0129] Acquire the image of the face to be identified;
[0130] The face image to be identified is input into the face recognition model to obtain the recognition result.
[0131] Example 4:
[0132] This embodiment provides a storage medium, which is a computer-readable storage medium, storing a computer program. When the computer program is executed by a processor, it implements the big data-based face recognition method of Embodiment 1 above, as follows:
[0133] Obtain a first dataset, which includes multiple different face images;
[0134] For each face image, the image is divided according to preset rules and a face key point recognition model to obtain a first region image, a second region image and a third region image. The first region image, the second region image and the third region image are processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset.
[0135] The first and second datasets are input into the neural network model for training to obtain the face recognition model;
[0136] Acquire the image of the face to be identified;
[0137] The face image to be identified is input into the face recognition model to obtain the recognition result.
[0138] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.
[0139] In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this embodiment, the computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.
[0140] The computer-readable storage medium described above can be used to write computer programs for executing this embodiment in one or more programming languages or combinations thereof. These programming languages include object-oriented programming languages—such as Java, Python, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0141] In summary, this invention, tailored to the characteristics of large-scale facial data, performs multi-region preprocessing to expand facial image samples and improve the generalization performance of the neural network model. Furthermore, by using an original model, the number of model parameters is reduced, and the model's accuracy is improved, thereby achieving rapid facial image recognition and facilitating deployment on terminal devices.
[0142] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.
Claims
1. A face recognition method based on big data, characterized in that, The method includes: Obtain a first dataset, which includes multiple different facial images, the facial images being facial images from a large dataset of identity verification data; For each face image, the image is divided according to preset rules and a face key point recognition model to obtain a first region image, a second region image and a third region image. The first region image, the second region image and the third region image are processed according to multiple image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset. The first and second datasets are input into the neural network model for training to obtain the face recognition model; Acquire the image of the face to be identified; Input the image of the face to be identified into the face recognition model to obtain the recognition result; The neural network model includes an input layer, a first convolutional module, a second convolutional module, a third convolutional module, a first adaptive average pooling layer, a first flattening layer, a first dropout layer, a first fully connected module, a second adaptive average pooling layer, a second flattening layer, a second dropout layer, a second fully connected module, a feature fusion layer, and an output layer. The input layer is connected to the first convolutional module and the second convolutional module. The first convolutional module, the first adaptive average pooling layer, the first flattening layer, the first dropout layer, and the first fully connected module are connected sequentially. The second convolutional module, the third convolutional module, the second adaptive average pooling layer, the second flattening layer, the second dropout layer, and the second fully connected module are connected sequentially. The first fully connected module and the second fully connected module are connected to the feature fusion layer. The feature fusion layer is connected to the output layer. The step of dividing the image according to preset rules and a facial key point recognition model to obtain a first region image, a second region image, and a third region image includes: identifying the chin point, two temple points, and multiple auricle points of the two ears in the image according to the facial key point recognition model; using the triangular region enclosed by the chin point and the two temple points as the first region image; and using the two ears as the ear regions enclosed by the multiple auricle points as the second region image and the third region image, respectively. The processing of the first region image, the second region image, and the third region image according to multiple image interference methods includes: performing noise processing and / or compression processing on the first region image; performing image obfuscation processing and / or random cropping processing on the second region image; and performing image obfuscation processing and / or random cropping processing on the second region image and the third region image.
2. The method according to claim 1, characterized in that, The first convolutional module includes a first input sub-layer, a first 1×1 convolutional layer, a second 1×1 convolutional layer, a third 1×1 convolutional layer, a fourth 1×1 convolutional layer, a first max pooling layer, and a first output sub-layer; The first input sub-layer, the first 1×1 convolutional layer, the second 1×1 convolutional layer, and the third 1×1 convolutional layer are connected in sequence; The first input sub-layer, the fourth 1×1 convolutional layer, and the first max pooling layer are connected; The first output sub-layer is used to output the feature matrix output by the concatenated third 1×1 convolutional layer and the feature matrix output by the first max pooling layer; The first 1×1 convolutional layer and the second 1×1 convolutional layer include a 2x upsampling operation; The second 1×1 convolutional layer and the third 1×1 convolutional layer include a 2x upsampling operation.
3. The method according to claim 1, characterized in that, The second convolutional module includes a second input sub-layer, a fifth 1×1 convolutional layer, a first 2×2 convolutional layer, a second 2×2 convolutional layer, a third 2×2 convolutional layer, a first 3×3 convolutional layer, and a second output sub-layer; The second input sub-layer, the fifth 1×1 convolutional layer, and the first 2×2 convolutional layer are connected in sequence; The second input sub-layer, the second 2×2 convolutional layer, and the third 2×2 convolutional layer are connected in sequence; The second input sub-layer and the first 3×3 convolutional layer are connected; The second output sub-layer is used to output the feature matrix output by the first 2×2 convolutional layer, the feature matrix output by the third 2×2 convolutional layer, the feature matrix output by the first 3×3 convolutional layer, and the feature matrix output by the fifth 1×1 convolutional layer after being concatenated.
4. The method according to claim 1, characterized in that, The third convolutional module includes a third input sub-layer, a second 3×3 convolutional layer, a third 3×3 convolutional layer, a fourth 3×3 convolutional layer, a fifth 3×3 convolutional layer, a sixth 1×1 convolutional layer, a seventh 1×1 convolutional layer, a second max pooling layer, and a third output sub-layer. The third input sub-layer, the second 3×3 convolutional layer, the third 3×3 convolutional layer, the fourth 3×3 convolutional layer, and the fifth 3×3 convolutional layer are connected in sequence; The third input sub-layer, the sixth 1×1 convolutional layer, and the second max pooling layer are connected in sequence. The third input sub-layer and the seventh 1×1 convolutional layer are connected; The third output sub-layer is used to output the feature matrix output by the fifth 3×3 convolutional layer, the feature matrix output by the second max pooling layer, the feature matrix output by the seventh 1×1 convolutional layer, the feature matrix output by the second 3×3 convolutional layer, the feature matrix output by the third 3×3 convolutional layer, and the feature matrix output by the fourth 3×3 convolutional layer after being spliced.
5. A face recognition system based on big data, characterized in that, The system includes: The first acquisition unit is used to acquire a first dataset, which includes multiple different face images, and the face images are face images from big data on identity verification. The second acquisition unit is used to divide each face image into a first region image, a second region image, and a third region image according to preset rules and a face key point recognition model. The first region image, the second region image, and the third region image are then processed according to various image interference methods and randomly combined into multiple multi-interference face images to obtain a second dataset. The training unit is used to input the first and second datasets into the neural network model for training, thereby obtaining the face recognition model. The third acquisition unit is used to acquire the face image to be identified; The recognition unit is used to input the face image to be recognized into the face recognition model and obtain the recognition result; The neural network model includes an input layer, a first convolutional module, a second convolutional module, a third convolutional module, a first adaptive average pooling layer, a first flattening layer, a first dropout layer, a first fully connected module, a second adaptive average pooling layer, a second flattening layer, a second dropout layer, a second fully connected module, a feature fusion layer, and an output layer. The input layer is connected to the first convolutional module and the second convolutional module. The first convolutional module, the first adaptive average pooling layer, the first flattening layer, the first dropout layer, and the first fully connected module are connected sequentially. The second convolutional module, the third convolutional module, the second adaptive average pooling layer, the second flattening layer, the second dropout layer, and the second fully connected module are connected sequentially. The first fully connected module and the second fully connected module are connected to the feature fusion layer. The feature fusion layer is connected to the output layer. The step of dividing the image according to preset rules and a facial key point recognition model to obtain a first region image, a second region image, and a third region image includes: identifying the chin point, two temple points, and multiple auricle points of the two ears in the image according to the facial key point recognition model; using the triangular region enclosed by the chin point and the two temple points as the first region image; and using the two ears as the ear regions enclosed by the multiple auricle points as the second region image and the third region image, respectively. The processing of the first region image, the second region image, and the third region image according to multiple image interference methods includes: performing noise processing and / or compression processing on the first region image; performing image obfuscation processing and / or random cropping processing on the second region image; and performing image obfuscation processing and / or random cropping processing on the second region image and the third region image.
6. A terminal device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the method described in any one of claims 1-4.
7. A storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1-4.