Image processing method and device, equipment, medium and product

Abstract

The invention discloses an image processing method and device, equipment, a medium and a product, and relates to the technical field of computers, and the method comprises the steps: obtaining a to-be-processed feature map; processing the to-be-processed feature map based on a preset super-resolution model to obtain a target feature map; wherein the super-resolution model comprises at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, and the at least one edge enhancement super-resolution operator module is used for carrying out feature extraction and enhancement on the to-be-processed feature map; the at least one global edge feature perception attention module is used for performing multi-layer feature extraction and fusion on the to-be-processed feature map, and processing the obtained to-be-processed feature map through a super-resolution model to obtain a target feature map; through at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module in the super-resolution model, the calculation cost in the image processing process can be reduced, and the image processing efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to an image processing method, apparatus, device, medium and product. Background Technology

[0002] One-dimensional barcodes, as a method of information encoding, play a crucial role in many industries. They enable the efficient and low-cost encoding and storage of data such as product information, identification, and asset records.

[0003] Existing one-dimensional barcode image recognition technologies typically require super-resolution algorithms to process low-resolution, blurry images. However, traditional super-resolution algorithms are often too complex and require a lot of computing resources and time, which limits their application on resource-constrained edge devices. The high computational cost and low efficiency of these algorithms result in high costs in real-time processing scenarios.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide an image processing method, apparatus, device, medium, and product, which aims to reduce the computational cost in the image processing process and improve the image processing efficiency.

[0006] To achieve the above objectives, this application provides an image processing method, the method comprising:

[0007] Obtain the feature map to be processed;

[0008] The feature map to be processed is processed based on a preset super-resolution model to obtain a target feature map; wherein, the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module is used to extract and enhance features of the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features of the feature map to be processed.

[0009] In one embodiment, the global edge feature awareness attention module includes a batch normalization layer, a first convolutional layer, a parameterized corrected linear unit, a second convolutional layer, and an adaptive weighting module.

[0010] In one embodiment, the step of processing the feature map to be processed based on a preset super-resolution model to obtain the target feature map includes:

[0011] The feature map to be processed is input into the batch normalization layer to obtain a normalized feature map;

[0012] The first convolutional layer is used to perform a first convolution operation on the normalized feature map to obtain a first-dimensional feature map.

[0013] The first dimension feature map is extracted by the at least one edge enhancement super-resolution operator module to obtain an edge enhancement feature map.

[0014] The edge enhancement feature map is nonlinearly transformed by the parameterized modified linear unit to obtain a nonlinear feature map;

[0015] The second convolutional layer performs a second convolution operation on the nonlinear feature map to obtain a second-dimensional feature map;

[0016] The second-dimensional feature map is adaptively adjusted by the adaptive weighting module to obtain the target feature map.

[0017] In one embodiment, the step of adaptively adjusting the second-dimensional feature map using the adaptive weighting module to obtain the target feature map includes:

[0018] In the adaptive weighting module, the second-dimensional feature map is normalized using an activation function to generate attention weights;

[0019] The attention weights are multiplied element-wise with the second dimension feature map to obtain the target feature map.

[0020] In one embodiment, the step of extracting features from the first-dimensional feature map using the at least one edge-enhancing super-resolution operator module to obtain the edge-enhanced feature map further includes:

[0021] Obtain several convolutional branches, wherein the several convolutional branches include ordinary convolution, Sobel convolution and Laplacian convolution;

[0022] The structure of the several convolutional branches is reparameterized to obtain the at least one edge-enhancing super-resolution operator module.

[0023] In one embodiment, the at least one edge enhancement super-resolution operator module includes a Sobel convolution kernel and a Laplacian convolution kernel. The step of extracting features from the first dimension feature map using the at least one edge enhancement super-resolution operator module to obtain an edge enhancement feature map includes:

[0024] The first-dimensional feature map is subjected to horizontal and vertical edge extraction using the Sobel convolution kernel to obtain first-order edge information.

[0025] The first-dimensional feature map is subjected to edge detection and feature enhancement using the Laplacian convolution kernel to obtain second-order edge information.

[0026] The edge enhancement feature map is obtained based on the first-order edge information and the second-order edge information.

[0027] Furthermore, to achieve the above objectives, this application also proposes an image processing apparatus, the image processing apparatus comprising:

[0028] The acquisition module is used to acquire the feature map to be processed;

[0029] The processing module is used to process the feature map to be processed based on a preset super-resolution model to obtain a target feature map; wherein, the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module is used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features from the feature map to be processed.

[0030] In addition, to achieve the above objectives, this application also proposes an image processing apparatus, the apparatus comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the image processing method as described above.

[0031] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and which, when executed by a processor, implements the steps of the image processing method described above.

[0032] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the image processing method described above.

[0033] One or more technical solutions proposed in this application have at least the following technical effects:

[0034] The process involves acquiring a feature map to be processed; processing the feature map based on a preset super-resolution model to obtain a target feature map; wherein the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module. The at least one edge enhancement super-resolution operator module is used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features from the feature map to be processed. By processing the acquired feature map to be processed through the super-resolution model to obtain the target feature map, and by using at least one edge enhancement super-resolution operator module in the super-resolution model to extract and enhance features from the feature map to be processed, it is ensured that important features can still be accurately recovered and extracted under low-contrast and complex background image conditions. By using at least one global edge feature perception attention module to extract and fuse multi-layer features from the feature map to be processed, the model can adaptively perceive global features in the feature map to be processed, thereby reducing the computational cost in the image processing process and improving image processing efficiency and processing effect. Attached Figure Description

[0035] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0036] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a flowchart illustrating an embodiment of the image processing method of this application.

[0038] Figure 2 This is a schematic diagram of the overall network structure in the embodiments of this application;

[0039] Figure 3 Examples of embodiments in this application Figure 1 A flowchart illustrating step S20;

[0040] Figure 4 This is a schematic diagram of the model structure in the embodiments of this application;

[0041] Figure 5 This is a flowchart illustrating the image processing method of Embodiment 3 of this application;

[0042] Figure 6 This is a schematic diagram of the edge enhancement super-resolution operator structure in an embodiment of this application;

[0043] Figure 7 This is a schematic diagram illustrating the principle of structural reparameterization in the embodiments of this application;

[0044] Figure 8 This is a schematic diagram of the network structure in an embodiment of this application;

[0045] Figure 9 This is a schematic diagram of the module structure of the image processing device according to an embodiment of this application;

[0046] Figure 10 This is a schematic diagram of the device structure of the hardware operating environment involved in the image processing method in the embodiments of this application.

[0047] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0048] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0049] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0050] The main solution of this application embodiment is as follows: A feature map to be processed is acquired; the feature map to be processed is processed based on a preset super-resolution model to obtain a target feature map; wherein, the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module. The at least one edge enhancement super-resolution operator module is used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features from the feature map to be processed. By processing the acquired feature map to be processed through the super-resolution model to obtain the target feature map, and by extracting and enhancing features from the feature map to be processed through at least one edge enhancement super-resolution operator module in the super-resolution model, it is ensured that important features can still be accurately recovered and extracted under low-contrast and complex background image conditions; by extracting and fusing multi-layer features from the feature map to be processed through at least one global edge feature perception attention module, the model can adaptively perceive global features in the feature map to be processed, thereby reducing the computational cost in the image processing process and improving image processing efficiency and processing effect.

[0051] In this embodiment, for ease of description, the following description uses an image processing device as the execution subject.

[0052] The image processing method in this application is applicable to the recognition and processing of various types of images. This application uses a barcode image as an example for illustration.

[0053] One-dimensional barcodes, as a method of information encoding, play a crucial role in numerous industries. They enable the efficient and low-cost encoding and storage of data such as product information, identification, and asset records. In the retail industry, one-dimensional barcodes facilitate rapid checkout and inventory management, significantly improving transaction efficiency and accuracy. In logistics and transportation, they aid in real-time package tracking and supply chain management, ensuring timely and safe delivery of goods. Furthermore, the application of one-dimensional barcodes in healthcare, food safety, ticketing systems, and many other fields has greatly improved the efficiency and security of information management.

[0054] While barcodes play a significant role in information management, in practical applications, limitations of scanning equipment, environmental factors, or wear and tear on the barcode itself can cause images to become blurry and unreadable, making them unrecognizable. Super-resolution algorithms improve the recognizability of barcodes by increasing image resolution and enhancing image details. This is crucial for improving overall system performance and user experience. Super-resolution algorithms can make blurry barcode images clear, ensuring accurate information reading, and this technology is particularly important in automated, unattended scenarios.

[0055] Edge devices, such as smartphones and handheld scanners, are playing an increasingly important role in 1D barcode image recognition due to their portability and flexibility. These devices typically have limited computing power and storage space; therefore, traditional super-resolution algorithms, due to their high computational complexity and resource consumption, are often difficult to run effectively on these devices. To address this issue, developing lightweight super-resolution algorithms is crucial. By designing a super-resolution network specifically for 1D barcodes and reparameterizing its network structure, the computational requirements of the algorithm can be significantly reduced. Furthermore, for 1D barcode super-resolution tasks, a global edge feature-aware attention mechanism is designed to maintain or improve image enhancement effects, enabling the effective application of super-resolution technology on edge devices and achieving real-time, efficient 1D barcode image processing.

[0056] In conclusion, the widespread application of 1D barcode images across multiple fields highlights their importance, and super-resolution algorithms play a crucial role in improving the accuracy of 1D barcode recognition. With the rise of edge computing, lightweight super-resolution algorithms are of great significance for achieving efficient adaptation to edge devices. This not only improves user experience but also promotes the application of related technologies in a wider range of scenarios.

[0057] Existing one-dimensional barcode image recognition technologies often suffer from insufficient accuracy when processing low-resolution, blurry images. Furthermore, traditional super-resolution algorithms are often overly complex, requiring significant computational resources and time, which limits their application on resource-constrained edge devices. The high computational cost and low efficiency of these algorithms result in high costs for real-time processing scenarios.

[0058] This application provides a solution to address the following technical problem:

[0059] 1. Improve the accuracy of barcode image recognition: Super-resolution computing can significantly improve the recognition accuracy of low-resolution or blurry barcode images, and maintain high accuracy even under adverse scanning or printing conditions.

[0060] 2. Optimize computational efficiency: To address the resource constraints of edge devices, a lightweight super-resolution model is developed using techniques such as structured reconstruction and global channel pruning. This reduces the complexity and number of parameters of the model, thereby reducing computation time and resource consumption, and ensuring that the algorithm can run quickly on edge devices.

[0061] 3. Meet real-time processing requirements: Design a super-resolution algorithm that enables fast image processing to meet the needs of real-time barcode image recognition in retail, logistics and other fields.

[0062] 4. Enhance the model's versatility and adaptability: Develop a super-resolution model that can adapt to one-dimensional barcode images with different characteristics, ensuring that the algorithm remains efficient and accurate in various application scenarios.

[0063] By addressing the aforementioned technical problems, this application aims to provide a lightweight, efficient, and accurate method for super-resolution enhancement of one-dimensional barcode images. This method can not only significantly improve the recognition accuracy of one-dimensional barcode images, but also meet the real-time processing requirements of edge devices, and has broad versatility and adaptability.

[0064] In this embodiment, structural reparameterization technology is used to transform the network structure between the training and inference stages, thereby improving its inference efficiency. Furthermore, in super-resolution (SR) tasks, pixel-level optimization of images is required, necessitating the use of specific operators. This patent designs an Edge-Guided Operator (EGO) module and a Global Edge Feature Sensing Attention (GEFSA) structure specifically for super-resolution tasks to enhance their inference performance.

[0065] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or image processing device capable of performing the above functions. The following description uses an image processing device as an example to illustrate this embodiment and the subsequent embodiments.

[0066] Based on this, embodiments of this application provide an image processing method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the image processing method of this application.

[0067] In this embodiment, the image processing method includes steps S10 to S20:

[0068] Step S10: Obtain the feature map to be processed;

[0069] Specifically, the image processing method in this application embodiment is applicable to the recognition and processing of various types of images. This application embodiment uses a one-dimensional barcode image as an example for illustration.

[0070] Optionally, one-dimensional barcode images typically have the following characteristics:

[0071] Lines and edges: One-dimensional barcodes consist of parallel lines (bars) and spaces (spaces) that form specific patterns to encode information. The edges of the lines and spaces are the most prominent features of one-dimensional barcodes.

[0072] High contrast: To ensure readability, barcodes typically have high contrast between the lines and spaces.

[0073] Regularity: The spacing and width of the lines in a one-dimensional barcode follow specific encoding rules, exhibiting regularity;

[0074] Directionality: The lines of a one-dimensional code are horizontal, and the entire code is read in a horizontal direction.

[0075] Optionally, the feature map to be processed obtained in the embodiments of this application, such as the one-dimensional barcode feature map to be processed, can be an image with relatively blurry lines and edges. By processing it through a preset super-resolution model, an image with clear lines and edges can be obtained.

[0076] Step S20: Process the feature map to be processed based on the preset super-resolution model to obtain the target feature map;

[0077] The super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module.

[0078] Furthermore, after obtaining the feature map to be processed, it can be processed based on a preset super-resolution model to obtain the target feature map.

[0079] Optionally, the global edge feature awareness attention module includes a batch normalization layer, a first convolutional layer, a parameterized corrected linear unit, a second convolutional layer, and an adaptive weighting module.

[0080] Reference Figure 2 , Figure 2 This is a schematic diagram of the overall network structure in the embodiments of this application, such as... Figure 2 As shown, by stacking multiple Edge-Guided Operator (EGO) modules with Global Edge Feature Sensing Attention (GEFSA) modules, a super-resolution algorithm for one-dimensional codes is formed, which can effectively improve the detection performance of one-dimensional codes.

[0081] In this embodiment of the application, a dedicated super-resolution network is designed for one-dimensional barcode features to address the resource limitations of edge devices. Techniques such as EGO and GEFSA are employed to reduce the complexity and number of parameters of the model, thereby reducing computation time and resource consumption and ensuring that the algorithm can run quickly on edge devices.

[0082] This implementation, through the above scheme, specifically involves acquiring a feature map to be processed; processing the feature map based on a preset super-resolution model to obtain a target feature map; wherein, the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module. The at least one edge enhancement super-resolution operator module is used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features from the feature map to be processed. By processing the acquired feature map to be processed through the super-resolution model to obtain the target feature map, and by using at least one edge enhancement super-resolution operator module in the super-resolution model to extract and enhance features from the feature map to be processed, it is ensured that important features can still be accurately recovered and extracted under low-contrast and complex background image conditions; by using at least one global edge feature perception attention module to extract and fuse multi-layer features from the feature map to be processed, the model can adaptively perceive global features in the feature map to be processed, thereby reducing the computational cost in the image processing process and improving image processing efficiency and processing effect.

[0083] Based on the first embodiment of this application, a second embodiment of this application is proposed. In this second embodiment, content that is the same as or similar to that in the first embodiment described above can be referred to the above description and will not be repeated hereafter. Based on this, please refer to... Figure 3 , Figure 3 Examples of embodiments in this application Figure 1 The flowchart of step S20 is shown below. In one feasible implementation, step S20 may include steps S201 to S206:

[0084] S201: Input the feature map to be processed into the batch normalization layer to obtain a normalized feature map;

[0085] S202: Perform a first convolution operation on the normalized feature map through the first convolutional layer to obtain a first-dimensional feature map;

[0086] S203: Extract features from the first dimension feature map using the at least one edge enhancement super-resolution operator module to obtain an edge enhancement feature map;

[0087] S204: The edge enhancement feature map is subjected to a nonlinear transformation by the parameterized modified linear unit to obtain a nonlinear feature map;

[0088] S205: The second convolutional layer performs a second convolution operation on the nonlinear feature map to obtain a second-dimensional feature map;

[0089] S206: The second dimension feature map is adaptively adjusted by the adaptive weighting module to obtain the target feature map.

[0090] Optionally, the step of adaptively adjusting the second-dimensional feature map using the adaptive weighting module to obtain the target feature map includes:

[0091] In the adaptive weighting module, the second-dimensional feature map is normalized using an activation function to generate attention weights;

[0092] The attention weights are multiplied element-wise with the second dimension feature map to obtain the target feature map.

[0093] Optionally, this application proposes an innovative global edge feature-aware attention mechanism. This mechanism aims to enhance the network's ability to perceive key information in barcode images. By adaptively adjusting the degree of attention to different regions, it ensures that even with poor image quality, high-frequency detail features such as line edges and textures can be accurately extracted and enhanced.

[0094] Reference Figure 4 , Figure 4 This is a schematic diagram of the model structure in the embodiments of this application, such as... Figure 4As shown, the global edge feature awareness attention mechanism includes a batch normalization (BN) module, which captures one-dimensional code features from local to global; a 3×3 convolution module for reducing dimensionality and a 3×3 convolution module for increasing dimensionality; an EGO edge information extraction module; a PruLU module for increasing the network's nonlinearity; a final adaptive weighting module, Sigmoid, for dynamically adjusting the importance of features; and a fusion module that reintegrates the enhanced features into the network to guide subsequent super-resolution processing.

[0095] Optionally, the input feature maps are passed through a Batch Normalization (BN) layer. This layer normalizes each mini-batch of data by subtracting the mean and dividing by the standard deviation, then multiplying by a trainable scaling factor and adding an offset. The normalization process ensures that feature maps of different batches and scales have a consistent distribution, reducing gradient oscillations during training and accelerating convergence. As the initial module for global feature awareness, the BN layer's output provides stable and standardized feature inputs for subsequent convolutional operations, helping to maintain feature consistency and ensuring the effectiveness of subsequent feature extraction.

[0096] Optionally, in this embodiment, the first convolutional layer is a first 3x3 convolutional layer (dimensionality reduction convolution), used to perform a 3×3 convolution operation on the feature map output by the BN layer, reducing spatial dimensions while increasing feature depth. The convolutional kernel extracts low-level and mid-level features by weighted summation of local regions, achieving spatial compression and information condensation. This convolutional layer connects the BN layer output to the subsequent attention weight learning module, reducing the resolution of the feature map and the computational load, thus creating conditions for the execution of the attention mechanism.

[0097] Optionally, the Edge Enhancement Super-Operator (EGO) module extracts edge information from the reduced-dimensional feature map, primarily through convolution operations combined with a special filter kernel design to extract high-frequency details such as edges and textures in the image. The edge information is further enhanced by fusing high-frequency features to recover and amplify key lines and textures in the image. The EGO module directly processes the reduced-dimensional feature map and enhances edge and detail information during the feature extraction stage, providing crucial high-frequency information for subsequent feature enhancement and weighting. EGO provides more detailed feature maps for attention weight calculation, enabling the attention mechanism to better highlight important image regions.

[0098] Optionally, the Parametric Rectified Linear Unit (PReLU) activation function is applied after the EGO module output. By introducing learnable parameters to the activation function for each channel, the nonlinear transformation is adjusted. This adaptive adjustment of the activation function allows the network to adaptively recover and highlight detailed information in different scenarios, especially enhancing weak edges. PReLU operates on the edge-enhanced feature map, further improving the nonlinear expressive power and laying the foundation for subsequent convolutional operations and feature fusion. It ensures that the network can flexibly handle the diversity and complexity of features during processing.

[0099] Optionally, in this embodiment, the second convolutional layer is a second 3x3 convolutional layer (dimensionality-restored convolution). This module performs 3×3 convolutions, gradually restoring the spatial dimensions of the feature map to their original state while reducing the number of channels, achieving a compact and efficient feature representation. The convolutional kernel reconstructs features through the combination of multi-level features, creating conditions for subsequent attention weighting. The dimensionality-restored convolution is not only closely connected to the features processed by PReLU, but also provides a complete feature representation for the final attention weighting. The output of this module is a concise and information-rich feature map, which can better match the attention weights.

[0100] Optionally, the output of the second convolutional layer is normalized using the Sigmoid activation function, mapping the feature map values ​​to the range (0,1) to generate attention weights. These weights are used to adjust the response of the feature map, multiplying it with the feature map point-by-point to highlight features in key regions. The weighted multiplication of the Sigmoid-generated weights with the original features is the core step of the entire attention mechanism. This weighting process ensures that the network prioritizes high-confidence regions and weakens noisy or irrelevant features during super-resolution.

[0101] Optionally, through the compression operation of the sigmoid function described above, the generated weight map assigns different weight values ​​(usually in the channel dimension or spatial dimension) to different regions of the feature map. These weight values ​​reflect the network's attention to a certain region or specific feature. Specifically, regions with high weights indicate that the region or channel contains more important feature information and should be strengthened in subsequent processing. Regions with low weights indicate that the feature information in the region or channel is less important, and may be background or noise, and should be suppressed.

[0102] Optionally, after generating attention weights, the network applies these weights to the feature map through element-wise multiplication. This process is adaptive; that is, the network dynamically adjusts the weights based on the features of different input images, highlighting key regions and suppressing unimportant regions. This dynamic adjustment capability enables the network to handle barcode images of varying quality and in different scenarios, automatically focusing on the most informative parts of the image. For example, when processing barcodes with complex backgrounds or low contrast, the network will highlight edge information using sigmoid weights, while in relatively simple images, the network may focus more on the sharp parts of the barcode and suppress noise.

[0103] Based on any of the above embodiments of this application, a third embodiment of this application is proposed. In this third embodiment, content that is the same as or similar to the above embodiments can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 Before step S203, the image processing method further includes steps S01 to S02:

[0104] S01: Obtain several convolutional branches, wherein the several convolutional branches include ordinary convolution, Sobel convolution and Laplacian convolution;

[0105] S02: Perform structural reparameterization on the plurality of convolutional branches to obtain the at least one edge enhancement super-resolution operator module.

[0106] Optionally, the at least one edge-enhancing super-resolution operator module includes a Sobel convolution kernel and a Laplace convolution kernel.

[0107] Optionally, the step of extracting features from the first-dimensional feature map using the at least one edge-enhancing super-resolution operator module to obtain an edge-enhanced feature map includes:

[0108] The first-dimensional feature map is subjected to horizontal and vertical edge extraction using the Sobel convolution kernel to obtain first-order edge information.

[0109] The first-dimensional feature map is subjected to edge detection and feature enhancement using the Laplacian convolution kernel to obtain second-order edge information.

[0110] The edge enhancement feature map is obtained based on the first-order edge information and the second-order edge information.

[0111] Reference Figure 6 , Figure 6 This is a schematic diagram of the edge enhancement super-resolution operator structure in an embodiment of this application, as shown below. Figure 6 As shown, in this embodiment, first-order edge information extraction (extracting horizontal and vertical edges through Sobel convolution kernels) and second-order edge feature extraction (using Laplacian convolution kernels) are used for edge detection and feature enhancement.

[0112] Alternatively, the discrete derivative of one-dimensional barcode feature edge detection can be enhanced by using the Sobel operator, which emphasizes edges in the image by calculating the first derivative of the image. The Sobel operator contains two sets of convolutional kernels, one for detecting horizontal and one for detecting vertical edges. The Sobel operator considers both the brightness and gradient direction of the image, making it very effective in edge detection and providing directional information about the edges.

[0113] Sobel convolution kernel for horizontal edge detection:

[0114]

[0115] Sobel convolution kernel for vertical edge detection:

[0116]

[0117] Optionally, the Sobel operator computes the gradient magnitudes in the horizontal and vertical directions by applying these convolution kernels to each pixel in the image. The direction of an edge can be determined by the arctangent of the gradient direction, while the intensity of the edge can be determined by the gradient magnitude.

[0118] Optionally, the Laplacian operator is used in this embodiment to detect significant changes in the image, improving the super-resolution algorithm's capabilities in image sharpening, edge detection, and frequency domain analysis. For one-dimensional barcode images, the Laplacian convolution kernel can help identify and enhance corners, lines, and textures in the image. It is a 4-neighborhood or 8-neighborhood convolution kernel, which can be represented as:

[0119] 4-neighbor Laplace convolution kernel:

[0120]

[0121] 8-neighbor Laplace convolution kernel:

[0122]

[0123] In this embodiment, the above four types of convolution kernels are used to extract first-order and second-order edge information from the image. This helps to restore and enhance details when the image is enlarged, thereby improving the quality of the final image.

[0124] Optionally, since one-dimensional barcodes are mainly composed of lines and edges, the Sobel operator can effectively detect these edges, enhance line details in the image, and improve the readability and recognition accuracy of the one-dimensional barcode image. In addition, the Sobel operator can provide edge direction information, which helps determine the reading direction of the one-dimensional barcode, and can also suppress image noise to a certain extent, thereby improving the recognition quality of the one-dimensional barcode.

[0125] Optionally, the Laplacian convolution kernel effectively enhances the clarity and sharpness of lines in the 1D code image by emphasizing rapidly changing areas, such as the start and end points of lines, thereby enhancing the features of the 1D code and maintaining high decoding accuracy even when the image is damaged or blurred.

[0126] Reference Figure 7 , Figure 7 This is a schematic diagram illustrating the principle of structural reparameterization in the embodiments of this application, such as... Figure 7 As shown, the five branches in the diagram include two ordinary convolutions, an ordinary convolution + Sobel operator convolution x, an ordinary convolution + Sobel operator convolution y, an ordinary convolution + Laplacian convolution -4, and an ordinary convolution + Laplacian convolution -8. These can all be converted into equivalent 3×3 convolutions. Therefore, structural reparameterization can be used to merge them, resulting in a simple linear network. The specific operations for each branch are as follows:

[0127] ① Two ordinary convolutions

[0128] like Figure 7 As shown, two 3×3 ordinary convolutions can be expressed by the formula:

[0129] Y n1 =W k1 ×X1+B k1

[0130] Y n2 =W k2 ×Y n1 +B k2

[0131] Among them, Y n1 W is the output of the first ordinary convolution. k1 Here is the weight matrix for the first layer, X1 is the input feature, and B... k1 Y is the bias vector for the first layer; n2 W is the output of the second ordinary convolution. k2 B is the weight matrix for the second layer. k2 This is the bias vector for the second layer.

[0132] These two convolutions can be combined into one convolution, as shown in the formula:

[0133] Y n =W k2 ×(W k1 ×X1+B k1 )+B k2

[0134] =W k ×X1+B k

[0135] Among them, Y n W is the output of two ordinary convolutions. k1 Here is the weight matrix for the first layer, X1 is the input feature, and B... k1 W is the bias vector for the first layer; k2 B is the weight matrix for the second layer. k2 W is the bias vector for the second layer. k Let B be the weight matrix of the k-th layer. k Let be the bias vector of the k-th layer, so two ordinary 3×3 convolutions can be equivalently represented by a single 3×3 convolution.

[0136] ②Ordinary convolution+ so Bel operator convolution

[0137] like Figure 7 As shown, ordinary convolution + so The bel convolution can be expressed by the formula:

[0138] Y n =W k ×X1+B k

[0139] Y sobel-x =G x ×Y n +B x

[0140] Among them, Y n W is the output of two ordinary convolutions. k Let X1 be the weight matrix of the k-th layer, and B be the input feature. k Y is the bias vector of the k-th layer. soble-x G is the output of the Sobel operator convolution. x For the horizontal Sobel operator, B x Let x be the bias vector of the x-th layer.

[0141] These two convolutions can be combined into one convolution, as shown in the formula:

[0142] Y Sobel-x =G x ×(W k ×X1+B k )+B x

[0143] =W k ×X1+B k

[0144] Where Gx is the Sobel convolution kernel for the aforementioned horizontal edge detection, it can be seen that ordinary convolution + soThe bel convolution can also be equivalently replaced by a 3×3 ordinary convolution. Similarly, Ysobel-y is equivalently replaced by a 3×3 ordinary convolution.

[0145] ③ Ordinary convolution + Laplace convolution kernel

[0146] like Figure 7 As shown, a regular convolution with a Laplacian kernel can be expressed by the following formula:

[0147] Y n =W k ×X1+B k

[0148] Y lap-4 =L4×Y n +B lap-4

[0149] Among them, Y n W is the output of two ordinary convolutions. k Let X1 be the weight matrix of the k-th layer, Bk be the input features, and Y be the bias vector of the k-th layer. lap-4 B is the output of Laplacian convolution -4. lap-4 Given the bias vector of the Laplacian convolution layer -4, these two convolutions can be merged into one convolution, expressed by the formula:

[0150] Y lap-4 =L4×(W k ×X1+B k )+B lap-4

[0151] =W l4 ×X1+B l4

[0152] Among them, Y lap-4 W is the output of Laplacian convolution -4. k Let X1 be the weight matrix of the k-th layer, and B be the input feature. k Let B be the bias vector of the k-th layer. lap-4 W is the bias vector for the Laplacian convolution - 4 layers. 14 This is the weight matrix.

[0153] Here, L4 is the aforementioned 4-neighborhood Laplacian convolution kernel. It can be seen that a regular convolution plus a Laplacian convolution kernel can also be equivalently replaced by a 3×3 regular convolution. Furthermore, Ylap-8 uses the same method, with an equivalent replacement of a 3×3 regular convolution.

[0154] ④ Merge the above 5 branches

[0155] All five branches mentioned above can be converted into equivalent 3×3 convolutions, so structural reparameterization can be used to merge them, resulting in a simple linear network, namely:

[0156] Y n +Y sob_x +Y sob_y +Y lap_4 +Y lap_8 =Y rep

[0157] Among them, Y n The outputs of two ordinary convolutions, Y sob-x The output of the Sobel operator x convolution, Y sob-y The output of the Sobel operator y convolution, Y lap-4 The output of Laplacian convolution-4, Y lap-8 The output of Laplacian convolution -8, Y rep This is the output of a linear network.

[0158] Reference Figure 8 , Figure 8 This is a schematic diagram of the network structure in an embodiment of this application, such as... Figure 8 As shown, by stacking multiple edge-enhancing super-resolution operators, the model can focus on recovering important visual features in the image, such as edges and textures, which is crucial for generating high-quality super-resolution images. At the same time, the edge-enhancing super-resolution operator module is designed with inference speed and model complexity in mind, making it suitable for deployment on resource-constrained devices.

[0159] Optionally, this embodiment details a dedicated network structure and its reparameterization process designed for 1D barcode super-resolution tasks. This network significantly improves the processing speed and efficiency of 1D barcode image super-resolution tasks by dynamically adjusting the network structure between the training and inference phases. However, despite the speed breakthrough achieved by this network, further improvements in feature extraction accuracy and detail restoration quality are still needed when processing 1D barcode images with complex backgrounds or low-contrast lines. To address this issue and fully utilize the efficiency of the reparameterized network, this embodiment proposes an innovative global edge feature-aware attention mechanism. This mechanism aims to enhance the network's ability to perceive key information in 1D barcode images. By adaptively adjusting the degree of attention to different regions, it ensures accurate extraction and enhancement of high-frequency detail features such as line edges and textures, even in cases of poor image quality. Specific solutions can be found in the foregoing embodiments, and will not be elaborated upon here.

[0160] This embodiment, through the above scheme, specifically obtains several convolutional branches, including ordinary convolution, Sobel convolution, and Laplacian convolution; performs structural reparameterization on the several convolutional branches to obtain at least one edge enhancement super-resolution operator module, which specifically extracts and enhances key edges and high-frequency details in the one-dimensional barcode image, ensuring that important features can still be accurately recovered and extracted under low-contrast and complex background image conditions.

[0161] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the image processing method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0162] This application also provides an image processing apparatus, please refer to... Figure 9 The image processing apparatus includes:

[0163] The acquisition module is used to acquire the feature map to be processed;

[0164] The processing module is used to process the feature map to be processed based on a preset super-resolution model to obtain a target feature map; wherein, the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module is used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multi-layer features from the feature map to be processed.

[0165] In this embodiment, the EGO module is used to specifically extract and enhance key edges and high-frequency details in the 1D barcode image, ensuring accurate recovery and extraction of important features even in low-contrast and complex background images. Through multi-layer feature extraction and fusion, the network can adaptively perceive global features in the 1D barcode image and ensure precise weighting of key features through an attention mechanism, improving the image super-resolution processing effect. By combining the edge enhancement attention mechanism, specifically through the Sigmoid function combined with the EGO module, adaptive weights are generated not only on the feature map but also adaptive enhancement is specifically applied to edges and high-frequency features. This innovative combination improves the processing accuracy of the attention mechanism for 1D barcode images in complex scenes. The feature map generated through multi-level convolutional operations and PReLU nonlinear processing allows the attention mechanism to consider both local and global features when generating weights, ensuring precise adjustment of the importance of different image regions. The global edge feature perception attention mechanism in this embodiment complements the reparameterized network, fully utilizing the dynamic structural adjustment capability of reparameterization technology during the training and inference stages, significantly improving inference speed while ensuring the accuracy of feature extraction and detail recovery. This not only optimizes super-resolution performance but also takes into account the computing power of edge devices, enabling efficient operation on resource-constrained devices. This innovative design provides a practical solution for 1D barcode image super-resolution processing in edge computing. The various modules in this embodiment (BN layer, convolutional layer, EGO, PReLU, Sigmoid, and fusion module) work closely together, forming a complete and efficient super-resolution processing framework from global feature perception to edge detail enhancement and feature fusion. Under this architecture, the network not only possesses powerful feature extraction capabilities but also maintains efficient inference, making it particularly suitable for 1D barcode image processing in complex backgrounds.

[0166] This application achieves efficient and accurate super-resolution processing of 1D barcode images in complex scenarios by combining global edge feature perception with an attention mechanism, accurately extracting and adaptively weighting edge enhancement features, and efficiently integrating with a structurally reparameterized network. This not only improves the quality of super-resolution but also demonstrates significant practicality in edge device applications. The super-resolution algorithm in this application significantly improves the recognition accuracy of low-resolution or blurry 1D barcode images, maintaining high accuracy even under adverse scanning or printing conditions. To address the resource limitations of edge devices, a dedicated super-resolution network is designed for 1D barcode features, incorporating techniques such as EGO and GEFSA to reduce model complexity and the number of parameters, thereby reducing computation time and resource consumption and ensuring the algorithm can run quickly on edge devices. This provides a super-resolution algorithm capable of fast image processing to meet the real-time 1D barcode image recognition needs of retail, logistics, and other fields. It also provides a super-resolution model adaptable to 1D barcode images with different characteristics, ensuring the algorithm remains efficient and accurate in various application scenarios.

[0167] The image processing apparatus provided in this application, employing the image processing method described in the above embodiments, can solve the technical problems of image processing. Compared with the prior art, the beneficial effects of the image processing apparatus provided in this application are the same as those of the image processing method described in the above embodiments, and other technical features in the image processing apparatus are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0168] This application provides an image processing apparatus, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the image processing method in Embodiment 1 above.

[0169] The following is for reference. Figure 10 This document illustrates a structural schematic diagram of an image processing device suitable for implementing embodiments of this application. The image processing device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 10 The image processing device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0170] like Figure 10 As shown, the image processing device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the image processing device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the image processing device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows an image processing device with various systems, it should be understood that it is not required to implement or possess all of the systems shown. More or fewer systems may be implemented alternatively.

[0171] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0172] The image processing apparatus provided in this application, employing the image processing method described in the above embodiments, can solve the technical problems of image processing. Compared with the prior art, the beneficial effects of the image processing apparatus provided in this application are the same as those of the image processing method described in the above embodiments, and other technical features of the image processing apparatus are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0173] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0174] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0175] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the image processing method described in the above embodiments.

[0176] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, 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 devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0177] The aforementioned computer-readable storage medium may be included in an image processing device or may exist independently without being assembled into an image processing device.

[0178] The aforementioned computer-readable storage medium carries one or more programs that, when executed by an image processing device, cause the image processing device to: acquire a feature map to be processed; process the feature map to be processed based on a preset super-resolution model to obtain a target feature map; wherein the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module being used to extract and enhance features from the feature map to be processed, and the at least one global edge feature perception attention module being used to perform multi-layer processing on the feature map to be processed. Feature extraction and fusion are performed by processing the acquired feature map to obtain the target feature map through a super-resolution model. At least one edge enhancement super-resolution operator module in the super-resolution model is used to extract and enhance features from the feature map to ensure accurate recovery and extraction of important features even under low-contrast and complex background image conditions. At least one global edge feature awareness attention module is used to perform multi-layer feature extraction and fusion on the feature map to be processed, enabling the model to adaptively perceive global features in the feature map to be processed, thereby reducing the computational cost in the image processing process and improving image processing efficiency and processing effect.

[0179] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code 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 can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0180] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0181] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0182] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for performing the above-described image processing method, thereby solving the technical problem of image processing. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the image processing method provided in the above embodiments, and will not be repeated here.

[0183] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the image processing method described above.

[0184] The computer program product provided in this application can solve the technical problems of image processing. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the image processing method provided in the above embodiments, and will not be repeated here.

[0185] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. An image processing method, characterized in that: The method comprises: Get the feature map to be processed; The feature map to be processed is processed based on a preset super-resolution model to obtain a target feature map; wherein the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module is used to extract and enhance features of the feature map to be processed, and the at least one global edge feature perception attention module is used to perform multi-layer feature extraction and fusion on the feature map to be processed.

2. The method according to claim 1, characterized in that The global edge feature perception attention module includes a batch normalization layer, a first convolutional layer, a parameterized rectified linear unit, a second convolutional layer and an adaptive weighting module.

3. The method according to claim 2, characterized in that The step of processing the feature map to be processed based on a preset super-resolution model to obtain a target feature map comprises: Inputting the feature map to be processed into the batch normalization layer to obtain a normalized feature map; Performing a first convolution operation on the normalized feature map through the first convolution layer to obtain a first dimensional feature map; Performing feature extraction on the first dimensional feature map by using the at least one edge enhancement super-resolution operator module to obtain an edge enhancement feature map; Performing a nonlinear transformation on the edge enhancement feature map by using the parameterized corrected linear unit to obtain a nonlinear feature map; The second convolution layer performs a second convolution operation on the nonlinear feature map to obtain a second dimensional feature map; The second dimensional feature map is adaptively adjusted through the adaptive weighting module to obtain the target feature map.

4. The method according to claim 3, characterized in that The step of adaptively adjusting the second dimensional feature map by the adaptive weighting module to obtain the target feature map comprises: In the adaptive weighting module, the second dimensional feature map is normalized by an activation function to generate an attention weight; The attention weight is multiplied element-by-element by the second dimensional feature map to obtain the target feature map.

5. The method according to claim 3, characterized in that The step of processing the feature map to be processed based on a preset super-resolution model to obtain a target feature map also includes: Acquire a plurality of convolution branches, wherein the plurality of convolution branches include ordinary convolution, Sobel convolution and Laplace convolution; The plurality of convolution branches are structurally reparameterized to obtain the at least one edge enhancement super-resolution operator module.

6. The method according to claim 3, characterized in that The at least one edge enhancement super-resolution operator module includes a Sobel convolution kernel and a Laplace convolution kernel, and the step of extracting features from the first dimensional feature map by using the at least one edge enhancement super-resolution operator module to obtain an edge enhancement feature map includes: Extracting horizontal and vertical edges of the first dimensional feature map using the Sobel convolution kernel to obtain first-order edge information; Performing edge detection and feature enhancement on the first dimensional feature map through the Laplace convolution kernel to obtain second-order edge information; The edge enhancement feature map is obtained according to the first-order edge information and the second-order edge information.

7. An image processing device, characterized in that: The device comprises: An acquisition module, used to acquire a feature map to be processed; A processing module is used to process the feature map to be processed based on a preset super-resolution model to obtain a target feature map; wherein the super-resolution model includes at least one edge enhancement super-resolution operator module and at least one global edge feature perception attention module, the at least one edge enhancement super-resolution operator module is used to extract and enhance features of the feature map to be processed, and the at least one global edge feature perception attention module is used to extract and fuse multiple layers of features of the feature map to be processed.

8. An image processing device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the image processing method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the image processing method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the image processing method according to any one of claims 1 to 6 are implemented.

Images