Human body image age estimation method, system, medium and device

Abstract

The application provides a human body image age estimation method, system, medium and device, comprising: acquiring a human body image and a corresponding age label; training a human body image age estimation model based on the human body image and the age label, so as to estimate the age of the human body image based on the trained human body image age estimation model. The application uses a convolutional neural network as a backbone model to extract human body image age-related features, and adds a design scheme of a plurality of age stage classifier network structures that overlap with each other at the end of the backbone model; the learning of the model on different age difference characteristics is strengthened, the correlation between different ages is related through the mutual overlap between the age stages of each classifier, and the accuracy of the human body image age estimation model is effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology in artificial intelligence applications, and relates to an estimation method, particularly a method, system, medium and device for estimating the age of human images. Background Technology

[0002] Estimating age based on human image information has broad application prospects in fields such as smart commerce, video surveillance, recommendation systems, and demographics. In the past, facial image information was often used for highly accurate age recognition, and the technology was mature. However, given the strict legal protections for facial information privacy under current regulations, current age recognition systems primarily rely on human images. For example, in a common smart commerce system like a shopping mall customer flow analysis system, passive camera images of customers are used to analyze and identify their ages, compiling age information for the customer group. This, combined with customer flow statistics, yields age distribution data for customers visiting each store. Based on this data, it's possible to predict which products users are truly interested in and in need, assisting in the planning and development of corresponding marketing strategies, providing crucial data support for mall operational decisions, and ultimately leading to profitability.

[0003] In the task of estimating the age of a human image, predicting the age of a human image, from the perspective of human subjective consciousness, often involves a comprehensive assessment of various aspects such as a person's clothing, hairstyle, hair color, hair volume, posture, height, and body shape.

[0004] Currently, the mainstream methods for estimating human age mainly use deep learning convolutional neural networks, which can be divided into two categories:

[0005] (1) Treat human age estimation as an ordered regression task, design a convolutional neural network model, regress a continuous value to represent the predicted age of human images, establish a human image age dataset, train and learn, and use the optimal regression model obtained from the training to identify the age value of human images.

[0006] (2) Treat human age estimation as a classification task, divide human age into a fixed number of categories (e.g., 1-100 years old), design a convolutional neural network model, establish a human image age dataset, train and learn, and use the best classification model obtained from the training to identify which age category the human image belongs to.

[0007] From the above two types of situations, we can conclude that:

[0008] The first method regresses a continuous value, which aligns with the continuous nature of human age. However, this method treats the aging process as a linear one, which is inconsistent with reality. Obviously, the changes in appearance caused by aging are faster in children than in young adults. Furthermore, humans are more likely to distinguish the age difference between two people than their absolute age. Convolutional neural networks are derived from the neuronal structure of the human brain, making this direct regression model more difficult to train and resulting in relatively low accuracy.

[0009] The second method discretizes continuous ages into equally spaced, non-overlapping intervals from 1 to 100 years old. Compared to the first method, this highlights age differences, which better aligns with human subjective perception and improves accuracy. However, this discretization method, covering an equal width across the target output range, ignores the relationships between different ages, resulting in limitations on model accuracy. Furthermore, practical applications present several labeling challenges, such as the difficulty in labeling age data and the unavoidable errors. In particular, different images of the same person may be labeled with different ages, and different annotators may make different subjective judgments, all of which affect the improvement of model accuracy.

[0010] Therefore, existing human age estimation techniques suffer from problems such as low estimation accuracy and large computational load due to the inability to take into account the interrelationships between different age categories and the high requirements for the accuracy of dataset labels. Summary of the Invention

[0011] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a method, system, medium and device for estimating the age of human images, in order to solve the problems of low estimation accuracy and large computational load in the process of implementing human age estimation technology, due to the inability to take into account the relationship between different age categories and the high requirements for the accuracy of dataset labels.

[0012] To achieve the above and other related objectives, in a first aspect, this application provides a method for estimating the age of a human image, comprising the following steps: acquiring a human image and a corresponding age label; training a human image age estimation model based on the human image and the age label, and estimating the age of the human image based on the trained human image age estimation model.

[0013] In this application, human images within a target area are first acquired, and the human image data are manually labeled. Training and testing datasets are then established. A convolutional neural network model is used to train the human images and age labels on the samples, thereby obtaining a human image age estimation model. Finally, the trained human image age estimation model is used to estimate the age of the human images. This application enhances the model's learning of different age-related characteristics, and the overlapping relationships between age groups in each classifier effectively improve the accuracy of the human image age estimation task by linking the correlations between different ages.

[0014] In one implementation of the first aspect, obtaining a human body image and its corresponding age tag includes the following steps: obtaining a target image containing a human body; extracting a human body image based on the target image; obtaining multiple age tags for the human body image; and constructing the age tag based on the average of the multiple age tags.

[0015] In this implementation, a human target image is acquired, and a human image is extracted from the human target image; then, age labels are manually labeled to obtain several age labels; finally, the age label corresponding to the human image is obtained by averaging the several age labels.

[0016] In one implementation of the first aspect, training a human image age estimation model based on the human image and the age label includes the following steps: preprocessing the human image; constructing a human image age estimation model, the human image age estimation model including a backbone network, fully connected layers, multiple fully connected layer groups, and a computation module; the backbone network is used to obtain age features of the human image; the fully connected layers are used to provide the age features to each fully connected layer group; the multiple fully connected layer groups are used to predict age based on the age features in multiple randomly overlapping age ranges; the computation module is used to obtain the estimated age of the human image based on the predicted age output by each fully connected layer group; and inputting the human image and the age label into the human image age estimation model to train the human image age estimation model.

[0017] In this implementation, human images are scaled to a uniform size, and a human image age estimation training model is built based on human images and age labels in the training dataset. Then, human images and age labels in the test dataset are input into the human image age estimation training model for testing, and finally, a human image age estimation model is obtained.

[0018] In one implementation of the first aspect, preprocessing the human body image includes the following steps: scaling the human body image to a preset size; and normalizing the pixel values ​​of the scaled human body image to the range of [-1, 1].

[0019] In one implementation of the first aspect, the backbone network is a ResNet18 network, which outputs an age feature of length 512.

[0020] In one implementation of the first aspect, the loss function of the human image age estimation model is:

[0021]

[0022] Where M represents the network being divided into M age groups for prediction; N represents the age being randomly divided into N age categories; The predicted age for the nth age segment of the mth group; is the age label for the nth segment of the mth group; the softmax function is used to transform the N-dimensional array of network prediction output into an array with a sum of 1, where each value represents the probability score of the corresponding predicted age segment.

[0023] In one implementation of the first aspect, the estimated age obtained by the calculation module is:

[0024]

[0025] Among them, age m M represents the median age predicted for each age group in the fully connected layer grouping in the model; M is the number of fully connected layer groups.

[0026] Secondly, this application provides a human image age estimation system, comprising: an acquisition module for acquiring a human image and a corresponding age label; a training module for training a human image age estimation model based on the human image and the age label; and an age estimation module for estimating the age of the human image based on the trained human image age estimation model.

[0027] In this application, human images within a target area are acquired through a data acquisition module, and these images are manually labeled. A convolutional neural network model is then used to train the human images and age labels on the samples, thereby obtaining a human image age estimation model. Finally, the trained human image age estimation model is used to estimate the age of the human images. This application enhances the model's ability to learn the characteristics of different age differences, and the overlapping relationships between age groups in each classifier effectively improve the accuracy of the human image age estimation task by linking the correlations between different ages.

[0028] Thirdly, this application provides a human image age estimation device, comprising: a processor and a memory. The memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the human image age estimation device performs the human image age estimation method.

[0029] Finally, this application provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the human image age estimation method.

[0030] As described above, the human image age estimation method, system, medium, and apparatus of the present invention have the following beneficial effects:

[0031] (1) The design scheme of this application adds multiple overlapping grouped classifiers at the end of the convolutional neural network. This overlapping design for different age groups takes into account the relationship between different age categories, fully represents the age in a diverse segment, and provides structural support for improving accuracy. Moreover, the model has a higher learning capacity and is more likely to learn the differences in human age, thereby improving the accuracy of the human image age estimation model.

[0032] (2) The human image age estimation method of this application can improve the accuracy of model age estimation while ensuring model performance, and the amount of computation is small.

[0033] (3) The age group design scheme of the human image age estimation model of this application with overlapping grouping at the end can reduce the requirements for the accuracy of the dataset label and alleviate the decrease in model accuracy caused by unavoidable errors in the process of manually labeling the dataset. Attached Figure Description

[0034] Figure 1 The diagram shows an implementation of the human image age estimation method of the present invention in an application scenario.

[0035] Figure 2 The diagram shown is a flowchart of an embodiment of the human image age estimation method of the present invention.

[0036] Figure 3 The diagram shown is a flowchart of step S11 in the human image age estimation method of the present invention.

[0037] Figure 4A The diagram shown is a flowchart of step S12 in the human image age estimation method of the present invention.

[0038] Figure 4B The diagram shown is a flowchart of step S121 in the human image age estimation method of the present invention.

[0039] Figure 5 The diagram shows a backbone model in one embodiment of the human image age estimation method of the present invention.

[0040] Figure 6 The diagram shows different discretization methods for the age regression problem in one embodiment of the human image age estimation method of the present invention.

[0041] Figure 7 The diagram shows a network structure of an overlapping age group classifier in one embodiment of the human image age estimation method of the present invention.

[0042] Figure 8 The diagram shown is a schematic representation of the principle structure of the human image age estimation system of the present invention in one embodiment.

[0043] Figure 9 The diagram shown is a schematic representation of the principle structure of the human image age estimation device of the present invention in one embodiment.

[0044] Component designation explanation

[0045] 81 Acquisition Module

[0046] 82 Training Module

[0047] 83 Age Estimation Module

[0048] 91 processor

[0049] 92 Memory

[0050] Steps S11 to S12 Detailed Implementation

[0051] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0052] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0053] The method for estimating the age of human images provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0054] Please see Figure 1 and Figure 2 The figures are respectively shown as schematic diagrams illustrating the application scenario of the human image age estimation method of the present invention and flowcharts illustrating an embodiment of the human image age estimation method of the present invention. Figure 1 and Figure 2 As shown, this embodiment provides a method for estimating the age of a human body image.

[0055] The method for estimating the age of a human image specifically includes the following steps:

[0056] S11, obtain the human image and its corresponding age label. Please refer to [link / reference]. Figure 3 The diagram shows a flowchart of step S11 in the human image age estimation method of the present invention. Figure 3 As shown, S11 includes the following steps:

[0057] S111, Obtain the target image containing the human body.

[0058] The probe acquires target images containing human figures. The probe is an image acquisition device capable of capturing images of the target area.

[0059] S112, Extract human body image based on target image.

[0060] Based on the target image of the human body acquired by the probe, the human body image is extracted from the target image.

[0061] In this embodiment, a human detection tool is used to extract the human body region image from the target image of the human body, and removes images of poor quality such as blurry, overly dark, overly bright, and overexposed images, retaining clean, clear, and identifiable human body region images for age labeling in the next step. The human detection tool can employ a matting method to extract the human body region image.

[0062] S113, Obtain multiple age tags for the human body image.

[0063] Age labels are manually assigned to human body images. In this embodiment, the age label for each human body image is manually assigned. The judgment criteria for manually assigning age labels are mainly based on visual judgment, such as factors like clothing, hairstyle, hair color, and body shape, to complete the age determination and labeling of the human body image, thereby obtaining several age labels.

[0064] S114, construct the age label based on the average of the multiple age labels.

[0065] Based on the several age labels obtained in step S113, a new age label is obtained by calculating their average value.

[0066] In this embodiment, the average value of several manually labeled age tags is taken to obtain an average age tag. This age tag is the average age judged by humans, thereby reducing the error caused by human subjectivity.

[0067] For example: A probe captures an image of a target area, obtaining a human target image. This image includes a pedestrian and surrounding scenery. Then, a human detection tool extracts the human body image from this target image; this human body image only shows the portion of the human body. Based on factors such as clothing, hairstyle, and posture, the extracted human body image is manually labeled with age tags, resulting in multiple age labels. For instance, based on a pedestrian's upright posture, dark hair, slender build, and tight summer clothing, the manually labeled ages are: 20, 18, 19, 18, 25, 18…21, 22, 24, 18, etc. Finally, the average age of these manually labeled ages is calculated, resulting in an average age of 22 years.

[0068] Therefore, training and testing datasets can be established based on the above. These datasets consist of human images and age labels.

[0069] For example, 10,000 human images and age labels were collected; of these, 8,000 human images and age labels were used to construct a training dataset for model training; and 2,000 human images and age labels were used to construct a test dataset for evaluation and selection of the optimal model.

[0070] S12, a human image age estimation model is trained based on the human image and the age label, so as to estimate the age of the human image based on the trained human image age estimation model.

[0071] Please see Figure 4A The diagram shows a flowchart of step S12 in the human image age estimation method of the present invention. Figure 4A As shown, step S12 includes the following steps:

[0072] S121, Preprocess the human body image.

[0073] Please see Figure 4B The diagram shows a flowchart of step S121 in the human image age estimation method of the present invention. Figure 4B As shown, step S121 includes the following steps:

[0074] S121A, the human body image is scaled to a preset size.

[0075] The acquired human images are scaled to obtain human images of a uniform size. In this embodiment, the acquired human images are RGB images; all human images are scaled to a uniform fixed size of 256x128.

[0076] S121B normalizes the pixel values ​​of the scaled human body image to the range [-1, 1].

[0077] In this embodiment, the pixel values ​​of the human body image are converted into unitless standard data and then processed in machine learning. Ultimately, these data are mapped to the interval [-1, 1]. Therefore, the value range of these data is within the same interval, which facilitates data processing.

[0078] S122, Construct a human image age estimation model by inputting the human image and the age label into the human image age estimation model to train the human image age estimation model.

[0079] A human image age estimation model is constructed, comprising a backbone network, fully connected layers, multiple fully connected layer groups, and a computation module. The backbone network is used to acquire age features of the human image. The fully connected layers are used to provide the age features to each fully connected layer group. The multiple fully connected layer groups are used to predict age based on the age features in multiple randomly overlapping age ranges. The computation module is used to obtain the estimated age of the human image based on the predicted age output by each fully connected layer group.

[0080] In this embodiment, to improve the performance and efficiency of the age estimation model, the backbone network in the model is used to extract globally shared features, and the CNN network employs a lightweight network. A CNN network is a neural network specifically designed to process data with a grid-like structure. A convolutional network is a neural network that uses convolution operations to replace general matrix multiplication operations in at least one layer. The structure of a convolutional neural network includes: an input layer, convolutional layers, pooling layers, activation function layers, and fully connected layers.

[0081] In this embodiment, the backbone network in the CNN network model is taken as an example for illustration.

[0082] ResNet18 is a deep learning neural network architecture. Its network depth is 18 layers, meaning the network depth refers to the weight layers. This includes pooling layers, convolutional layers, activation layers, and so on.

[0083] Please see Figure 5 The image shown is a schematic diagram of the backbone model in one embodiment of the human image age estimation method of the present invention. Figure 5 As shown, a convolutional neural network model for estimating the age of human images is used.

[0084] Specifically, initialize the network parameters. The initial weight parameters of the network are initialized using a normal distribution with a mean of 0 and a variance of 0.01, and the bias is initialized to 0.

[0085] Then, human images with a pixel size of 256x128 from the training dataset are input into the ResNet18 network. Figure 5 In this context, `conv` represents a convolutional layer, `maxpool` represents a max-pooling layer, and `avgpool` represents an average-pooling layer. Specifically, a 3x256x128 human image is input into the ResNet18 network. This data represents three channels, each with a size of 256x128. The input data first passes through a 7x7 convolutional layer with a stride of 2, padding of 3, and 64 output channels, resulting in an output data size of 64x128x64. Next, the 64x128x64 human image is passed through a pooling layer with a 3x3 kernel, a stride of 2, and padding of 1. Finally, the output data size is 64x64x32. This pooling process does not change the number of channels; it only halves the data size. After passing through convolutional layers, the network will finally output feature data of length 512, that is, a set of 512-dimensional age-related feature vectors are obtained by progressive dimensionality reduction.

[0086] The ResNet18 network used in this embodiment can also be replaced by other lightweight networks, such as MobelNet, EfficientNet, ShuffleNet, and combinations of one or more methods, including but not limited to the above networks.

[0087] Since this application employs a CNN network, the obtained 512-dimensional age-related feature vector is input into a fully connected layer. See [link to relevant documentation]. Figure 6 and Figure 7 The figures show schematic diagrams of different discretization methods for age regression problems in one embodiment of the human image age estimation method of the present invention, and a schematic diagram of the network structure of an overlapping age group classifier in one embodiment of the human image age estimation method of the present invention.

[0088] In a CNN, the fully connected layer (FC layer) acts as a "classifier" within the entire convolutional neural network. It maps the learned distributed feature representations to the sample space.

[0089] like Figure 6 As shown, in this embodiment, since standard age classifiers and equally spaced overlapping discrete age segment classifiers are insufficient to reflect the complexity of human image age estimation structures and the diversity of age label representations, we adopt a structure of randomly overlapping age segment classifiers to achieve this process. Compared with standard age classifiers and direct age regression, this model, which discretely represents age segments, has lower requirements for the accuracy of dataset label age annotations, mitigating some of the problems caused by manual labeling errors leading to decreased model accuracy. Each classifier's randomly discrete age segment is an encoding of the output age range. This encoding diversity and overlapping structure make the model easier to learn for age estimation tasks. This discrete method is used in conjunction with convolutional neural networks. At the end of the model, a fully connected layer connects M fully connected layers of size N to form a parallel structure of multiple classifiers. Each classifier has its own softmax activation, and the model is trained to have age estimation capabilities through a cross-entropy loss function between each classifier and its corresponding age segment label.

[0090] Please continue reading. Figure 7 The fully connected layer of this application employs a structure of several classifiers.

[0091] In this embodiment, the human age classification task is divided into M groups, each group being responsible for classifying and predicting an age range that is randomly divided into N segments. M and N are two hyperparameters. The parameters M and N are obtained through comparison of multiple sets of experiments. The number of classifiers M and the number of output categories N will affect the computational performance of the model, but compared with the computational cost of the convolutional neural network backbone model, the increased computational cost of a few more classifiers is still a very small part and has almost no impact on the overall computational performance of the model.

[0092] The larger the hyperparameters M and N are set, the more information the network can express, the more complex the calculation becomes, and the poorer the convergence will be. Conversely, setting them smaller will lead to large errors due to excessive discretization.

[0093] Specifically, based on the fully connected layers of the trained model, the network is divided into M age group prediction groups, and each prediction group is randomly divided into N age categories. A loss function is used to optimize the relationship between the predicted age value and the age label of the human image, thereby improving the accuracy of the discretized age group classification for each classifier.

[0094] The loss function of the human image age estimation model is:

[0095]

[0096] Where M represents the network being divided into M age groups for prediction; N represents the age being randomly divided into N age categories; The predicted age for the nth age segment of the mth group; Let n be the age label of the m-th group; the softmax function is expressed as: This is used to transform the N-dimensional array of network prediction output into an array that sums to 1, where each value represents the probability score for the corresponding predicted age group.

[0097] The network loss is calculated using the loss function. Each gradient backpropagation causes the loss to decrease until the model accuracy reaches its maximum on the test dataset, at which point training stops, thus obtaining a high-precision human image age estimation model.

[0098] Finally, based on the age estimation formula, the predicted ages output by the human image age estimation model are calculated to obtain the specific age of the human body.

[0099] Specifically, the average value of the human image age estimates obtained after processing by each classifier is calculated, and finally a high-precision human image age estimate can be obtained.

[0100] The estimated age obtained by the calculation module is:

[0101]

[0102] Among them, age m M represents the median age predicted for each age group in the fully connected layer grouping in the model; M is the number of fully connected layer groups.

[0103] A human image age estimation system was built based on a human image age estimation model, and it was used as an age recognition module for intelligent customer analysis systems such as shopping malls to meet different user needs.

[0104] The human image age estimation model of this application can be used as a module for the intelligent analysis system of shopping mall customer groups. Input a human image of a customer and return the corresponding age attribute information of the customer. Then, the age attribute information of the customer group visiting the store within a certain time period can be statistically analyzed. In conjunction with the customer flow statistics system, a group profile of the customers visiting the store can be drawn through this big data of age distribution. This provides data support for important measures such as the performance operation of the shopping mall, product deployment, and store planning, thereby bringing profitability.

[0105] The scope of protection for the human image age estimation method described in this application is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principles of this application is included within the scope of protection of this application.

[0106] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following... Figure 1 The method for estimating the age of human images.

[0107] At any possible level of technical detail, this application can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this application.

[0108] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, (but not limited to) electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0109] The computer-readable program described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards these instructions to a computer-readable storage medium in the respective computing / processing device. The computer program instructions used to perform the operations of this application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and procedural programming languages ​​such as "C" or similar programming languages. Computer-readable program instructions may execute entirely on a user's computer, partially on a user's computer, as a standalone software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may 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 may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing state information from the computer-readable program instructions. This electronic circuitry can execute the computer-readable program instructions to implement various aspects of this application.

[0110] This application also provides a human image age estimation system, which can implement the human image age estimation method described in this application. However, the implementation device of the human image age estimation method described in this application includes, but is not limited to, the structure of the human image age estimation system listed in this embodiment. All structural modifications and substitutions of the prior art made based on the principles of this application are included within the protection scope of this application.

[0111] The human image age estimation system provided in this embodiment will be described in detail below with reference to the illustrations.

[0112] This embodiment provides a human image age estimation system, including:

[0113] The acquisition module is used to acquire human images and corresponding age tags;

[0114] The training module is used to train a human image age estimation model based on the human image and the age label;

[0115] The age estimation module is used to estimate the age of human images based on a trained human image age estimation model.

[0116] Please see Figure 8 The diagram shows a schematic representation of the human image age estimation system of the present invention in one embodiment. Figure 8 As shown, the human image age estimation system includes: an acquisition module 81, a training module 82, and an age estimation module 83.

[0117] The acquisition module 81 is used to acquire human images and corresponding age tags.

[0118] Specifically, a target image containing a human body is acquired using a probe. The probe is an image acquisition device capable of capturing images of the target area. A human detection tool is used to extract the human body region from the target image, removing images of poor quality such as blurry, overly dark, overly bright, or overexposed areas, retaining clean, clear, and identifiable human body region images for age labeling in the next step.

[0119] Age labels are manually assigned to human body images. In this embodiment, the age label for each human body image is manually assigned. The judgment criteria for manually assigning age labels are mainly based on visual judgment, such as factors like clothing, hairstyle, hair color, and body shape, to complete the age determination and labeling of the human body image, thereby obtaining several age labels.

[0120] By averaging several manually labeled age tags, an average age tag can be obtained. This age tag represents the average age judged by humans, thereby reducing errors caused by human subjectivity.

[0121] Therefore, training and testing datasets can be established based on the above. These datasets consist of human images and age labels.

[0122] The training module 82 is connected to the acquisition module 81 and is used to train a human image age estimation model based on the human image and the age label.

[0123] In this embodiment, the human body image is first preprocessed.

[0124] Specifically, the acquired human body images are scaled to obtain human body images of a uniform size. In this embodiment, the acquired human body images are RGB images; all human body images are scaled to a uniform fixed size of 256x128. Then, the pixel values ​​of the scaled human body images are normalized to the range [-1, 1].

[0125] Then, a human image age estimation model is constructed. The human image and the age label are input into the human image age estimation model to train the human image age estimation model.

[0126] In this embodiment, a human image age estimation model is constructed, which includes: a backbone network, fully connected layers, multiple fully connected layer groups, and a calculation module; the backbone network is used to obtain the age features of the human image; the fully connected layers are used to provide the age features to each fully connected layer group; the multiple fully connected layer groups are used to predict the age based on the age features in multiple randomly overlapping age ranges; and the calculation module is used to obtain the estimated age of the human image based on the predicted age output by each fully connected layer group.

[0127] Specifically, to improve the performance and efficiency of the age estimation model, the backbone network in the model uses a lightweight CNN network for extracting globally shared features. A CNN is a neural network specifically designed to process data with a grid-like structure. A convolutional network is a neural network that uses convolution operations to replace general matrix multiplication operations in at least one layer. The structure of a convolutional neural network includes: an input layer, convolutional layers, pooling layers, activation function layers, and fully connected layers. The backbone network used is the ResNet18 network.

[0128] Specifically, initialize the network parameters. The initial weight parameters of the network are initialized using a normal distribution with a mean of 0 and a variance of 0.01, and the bias is initialized to 0.

[0129] Then, the human images with a pixel size of 256x128 in the training dataset are input into the ResNet18 network, and the final output is feature data of length 512. That is, a set of 512-dimensional age-related feature vectors are obtained by stepwise dimensionality reduction.

[0130] The networks used in this embodiment include algorithms such as MobelNet, EfficientNet, and ShuffleNet, and these algorithms can run on processing servers or terminals such as CPUs, NPUs, or GPUs.

[0131] The model will be trained to have the ability to estimate age by connecting M fully connected layers of size N to form a parallel structure of multiple classifiers. Each classifier has its own softmax activation. The model will be trained to have the ability to estimate age by using the cross-entropy loss function between each classifier and the corresponding age group label.

[0132] Specifically, the human age classification task is divided into M groups, each group responsible for classifying and predicting an age range that is randomly divided into N segments. M and N are two hyperparameters. The parameters M and N are determined by setting up multiple sets of experiments for comparison.

[0133] For example, based on the fully connected layers of the trained model, the network is divided into M age prediction groups, and each prediction group is randomly divided into N age categories. A loss function is used to optimize the relationship between the predicted age value and the age label of the human image, thereby improving the accuracy of the discretized age category classification for each classifier.

[0134] The network loss is calculated using the loss function. Each gradient backpropagation causes the loss to decrease until the model accuracy reaches its maximum on the test dataset, at which point training stops, thus obtaining a high-precision human image age estimation model.

[0135] The age estimation module 83 is used to estimate the age of human images based on a trained human image age estimation model.

[0136] The predicted ages output by the human image age estimation model are calculated based on the age estimation formula to obtain the specific age of the human body.

[0137] Specifically, the average value of the human image age estimates obtained after processing by each classifier is calculated, and finally a high-precision human image age estimate can be obtained.

[0138] A human image age estimation system was built based on a human image age estimation model, and it was used as an age recognition module for intelligent customer analysis systems such as shopping malls to meet different user needs.

[0139] It should be noted that the division of the various modules in the above system is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software through processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, module x can be a separate processing element, or it can be integrated into a chip within the system. Alternatively, it can be stored as program code in the system's memory, and its function can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element mentioned here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0140] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SoC).

[0141] Please see Figure 9 The diagram shows a schematic representation of the human image age estimation device of the present invention in one embodiment. Figure 9 As shown, this embodiment provides a human image age estimation device, which includes: a processor 91 and a memory 92; the memory 92 is used to store a computer program; the processor 91 is connected to the memory 92 and is used to execute the computer program stored in the memory 92, so that the human image age estimation device performs each step of the human image age estimation method as described above.

[0142] Preferably, the memory may include random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0143] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0144] In summary, the human image age estimation method, system, medium, and apparatus provided in this application have the following beneficial effects:

[0145] The design scheme in this application adds multiple overlapping grouped classifiers to the end of the convolutional neural network. This overlapping design for different age groups considers the relationships between different age categories, fully representing age in diverse segments, and providing structural support for improving accuracy. Furthermore, this model has a higher learning capacity and is more likely to learn the differences in human age, thereby improving the accuracy of the human image age estimation model. While mitigating the decrease in model accuracy caused by unavoidable errors in manually labeled datasets, it also improves the model's age estimation accuracy while maintaining model performance, and with relatively low computational cost.

[0146] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for estimating the age of a human body image, characterized in that, Includes the following steps: Obtain human body images and corresponding age labels; A human image age estimation model is trained based on the human image and the age label, and the human image age is estimated based on the trained human image age estimation model. The method includes: preprocessing the human image; constructing a human image age estimation model, the human image age estimation model including a backbone network, fully connected layers, multiple fully connected layer groups, and a computation module; the backbone network is used to obtain the age features of the human image; the fully connected layers are used to provide the age features to each fully connected layer group; the multiple fully connected layer groups are used to predict the age based on the age features in multiple randomly overlapping age ranges; the computation module is used to obtain the estimated age of the human image based on the predicted age output by each fully connected layer group; and inputting the human image and the age label into the human image age estimation model to train the human image age estimation model.

2. The method for estimating the age of a human image according to claim 1, characterized in that, Obtaining a human image and its corresponding age tag includes the following steps: Acquire a target image containing a human body; Extract human images based on the target image; Obtain multiple age tags for the human body image; The age label is constructed based on the average of the multiple age labels.

3. The method for estimating the age of a human image according to claim 1, characterized in that, Preprocessing the human body image includes the following steps: Scale the human body image to a preset size; The pixel values ​​of the scaled human body image are normalized to the range [-1, 1].

4. The method for estimating the age of a human image according to claim 1, characterized in that, The backbone network uses the ResNet18 network, which outputs an age feature of length 512.

5. The method for estimating the age of a human image according to claim 1, characterized in that, The loss function of the human image age estimation model is: Where M represents the network being divided into M age groups for prediction; N represents the age being randomly divided into N age categories; The predicted age for the nth age segment of the mth group; is the age label for the nth segment of the mth group; the softmax function is used to transform the N-dimensional array of network prediction output into an array with a sum of 1, where each value represents the probability score of the corresponding predicted age segment.

6. The method for estimating the age of a human image according to claim 1, characterized in that, The estimated age obtained by the calculation module is: Among them, age m M represents the median age predicted for each age group in the fully connected layer grouping in the model; M is the number of fully connected layer groups.

7. A human image age estimation system, characterized in that, include: The acquisition module is used to acquire human images and corresponding age tags; The training module is used to train a human image age estimation model based on the human image and the age label; The process includes: preprocessing the human image; constructing a human image age estimation model, which includes a backbone network, fully connected layers, multiple fully connected layer groups, and a computation module; the backbone network is used to obtain age features of the human image; the fully connected layers are used to provide the age features to each fully connected layer group; the multiple fully connected layer groups are used to predict age based on the age features in multiple randomly overlapping age ranges; the computation module is used to obtain the estimated age of the human image based on the predicted age output by each fully connected layer group; and inputting the human image and the age label into the human image age estimation model for training. The age estimation module is used to estimate the age of human images based on a trained human image age estimation model.

8. A human image age estimation device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is connected to the memory and is used to execute the computer program stored in the memory so that the human image age estimation device performs the human image age estimation method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the human image age estimation method according to any one of claims 1 to 6.

Images