Image compression method and device, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an image compression method and device, computer equipment and a storage medium. According to the scheme, an original image can be obtained through a client; performing feature extraction on the original image by using a pre-trained convolutional neural network model to obtain a key subject; performing image segmentation on the key main body to obtain a key region and a non-key region; respectively carrying out differential compression on the key region and the non-key region to obtain a key region compressed image and a non-key region compressed image; image reconstruction is carried out on the key region compressed image and the non-key region compressed image, detail features of the key region are recovered, and an optimized compressed image is generated, the definition of key details can be kept, meanwhile, the size of an image file is remarkably reduced, the storage and transmission cost is effectively reduced, and the image quality is improved. Not only is data processing efficiency improved, but also accurate transmission of image information required in the claim settlement process is ensured.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence technology and financial technology, and in particular to a graphics compression method, device, equipment and medium. Background Art

[0002] With the rapid development of modern information technology, image compression technology has become an important means of storing and transmitting image data. However, existing image compression methods often face a thorny problem: how to achieve efficient data compression while ensuring image quality. Especially in the insurance claims process, the clarity of key details in the image is extremely high, because these details are often the key evidence for judging accident responsibility and loss extent.

[0003] At present, many image compression technologies, such as JPEG and PNG, can reduce the size of image files to a certain extent, but they often sacrifice image quality during the compression process. Especially at high compression ratios, some subtle features and key details in the image may become blurred, which is undoubtedly a huge challenge for insurance claims work that requires precise analysis of image content.

[0004] To solve this problem, researchers and engineers are constantly exploring new image compression technologies. For example, by improving the JPEG standard, more advanced image formats such as JPEG 2000 and JPEG XR have been developed. These formats can better preserve image details and quality while providing higher compression efficiency. However, despite these advances, there is currently no image compression technology that can perfectly address the high requirements for image quality in the insurance claims process. Summary of the invention

[0005] The present invention provides an image compression method, device, computer equipment and medium to meet the high requirements for image quality in the insurance claim process.

[0006] In a first aspect, an image compression method is provided, comprising:

[0007] Get the original image;

[0008] Using a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects;

[0009] Performing image segmentation on the key subject to obtain a key area and a non-key area;

[0010] Differentially compressing the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image;

[0011] Image reconstruction is performed on the key area compressed image and the non-key area compressed image to restore the detail features of the key area and generate an optimized compressed image.

[0012] In a second aspect, an image compression device based on artificial intelligence is provided, comprising:

[0013] An acquisition module, used for acquiring the original image;

[0014] An extraction module, used for extracting features from the original image using a pre-trained convolutional neural network model to obtain key subjects;

[0015] A segmentation module, used for performing image segmentation on the key subject to obtain a key area and a non-key area;

[0016] A compression module, used for differentially compressing the key area and the non-key area respectively to obtain a key area compressed image and a non-key area compressed image;

[0017] The reconstruction module is used to reconstruct the compressed image of the key area and the compressed image of the non-key area, restore the detail features of the key area, and generate an optimized compressed image.

[0018] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned image compression method when executing the computer program.

[0019] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned image compression method are implemented.

[0020] In the scheme implemented by the above-mentioned artificial intelligence-based image compression method, device, computer equipment and storage medium, the original image can be obtained through the client; the pre-trained convolutional neural network model is used to extract features of the original image to obtain the key subject; the key subject is segmented to obtain the key area and the non-key area; the key area and the non-key area are differentially compressed to obtain a compressed image; the compressed image is reconstructed to restore the detail features of the key area and generate an optimized compressed image. In the present invention, a particularly suitable method is developed for the compression processing of high-resolution photos in insurance claims scenarios. This method uses advanced image processing technology, which can significantly reduce the image file size while maintaining the clarity of key details, effectively reducing storage and transmission costs. This innovation not only improves data processing efficiency, but also ensures the accurate transmission of image information required in the claims process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0022] Figure 1 is a schematic diagram of an application environment of an image compression method in one embodiment of the present invention;

[0023] Figure 2 is a flow chart of an image compression method according to an embodiment of the present invention;

[0024] Figure 3 yes Figure 2 A schematic flow chart of a specific implementation of step S200;

[0025] Figure 4 yes Figure 2 A schematic flow chart of a specific implementation of step S300;

[0026] Figure 5 yes Figure 2 A schematic flow chart of a specific implementation of step S400;

[0027] Figure 6 yes Figure 5 A schematic flow chart of a specific implementation of step S420;

[0028] Figure 7 yes Figure 2 A schematic flow chart of a specific implementation of step S500;

[0029] Figure 8 is a structural schematic diagram of an image compression device in one embodiment of the present invention;

[0030] Fig. 9 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;

[0031] Fig.10 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0033] The image compression method provided by the embodiment of the present invention can be applied in Figure 1 In the application environment, the client communicates with the server through the network. The server can obtain the original image through the client, use the pre-trained convolutional neural network model to extract the features of the original image, and obtain the key subject; perform image segmentation on the key subject to obtain the key area and the non-key area; perform differential compression on the key area and the non-key area to obtain a compressed image; perform image reconstruction on the compressed image, restore the detail features of the key area, and generate an optimized compressed image. In the present invention, for the compression processing of high-resolution photos in the insurance claim scenario, the method adopts advanced image processing technology, which can significantly reduce the image file size while maintaining the clarity of key details, and effectively reduce storage and transmission costs. This innovation not only improves the data processing efficiency, but also ensures the accurate transmission of the image information required in the claims process. Among them, the client can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0034] See also Figure 2 As shown, Figure 2 A schematic flow chart of an image compression method provided by an embodiment of the present invention includes the following steps:

[0035] S100: Acquire an original image.

[0036] The image compression method provided by the present invention can be applied to intelligent compression platforms such as photo processing platforms in various application scenarios. The photo processing platform is usually implemented through a server, which can obtain the original image. These images usually come from various sensors or devices, such as cameras, drones, etc. In order to extract useful information from these original images, we can use a pre-trained convolutional neural network (CNN) model for feature extraction. For example, in the field of insurance applications, in the insurance claims scenario, it is first necessary to obtain high-resolution original images of the accident scene or damaged items. These images are usually taken with professional cameras or smart phones to ensure that enough detailed information can be captured for subsequent analysis and processing.

[0037] For example, in the process of handling traffic accident claims, a comprehensive record of the accident vehicle is a crucial link. This not only involves the overall condition of the vehicle, but also includes detailed photography of each key part of the vehicle. In particular, valuable parts of the vehicle, such as lights and engines, are often severely damaged in accidents, so they need special attention.

[0038] By photographing these parts from different angles, we can ensure that all damages are accurately recorded. Such meticulous work helps to assess the extent of the loss and provide strong evidence support for subsequent claims. At the same time, it also avoids claims disputes caused by incomplete information and protects the legitimate rights and interests of both parties.

[0039] In addition, during the shooting process, you also need to pay attention to the impact of factors such as light and angle on image quality. Appropriate light and angle can make the image clearer and more realistic, which is conducive to more accurate judgment of the damage. Therefore, when shooting on site, you should try to choose a location with sufficient light and a suitable angle for shooting.

[0040] In short, at the traffic accident claim scene, it is very important to fully record the accident vehicle and take detailed photos of key parts. This not only helps to accurately assess the extent of the loss, but also provides strong evidence support for subsequent claims. Therefore, relevant personnel should pay full attention to this link when they are at the scene and ensure that the photography work is done well.

[0041] Of course, the source of the original image is not limited to the photos taken on-site by the insurance claims adjuster, but may also come from photos taken by other people on-site, as long as the photo is crucial to the insurance claims process, it can be used.

[0042] S200: Use a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects.

[0043] It is a common practice to use a pre-trained convolutional neural network (CNN) model to extract features from raw images. CNN automatically learns hierarchical features of images through its convolutional layers, which can capture the key subjects of the image. For example, models such as VGG and ResNet can be used for feature extraction.

[0044] The VGG model is an early classic CNN structure that uses a deeper network structure to extract image features through multiple convolutional layers and pooling layers. The VGG model achieved good results in the ImageNet competition, proving its effectiveness in image classification tasks. However, due to its deeper network structure, the amount of computation is large, so it is necessary to balance computing resources and performance in practical applications.

[0045] The ResNet model is a deeper CNN structure that solves the gradient vanishing problem by introducing residual connections, allowing the network to better learn deep features. The ResNet model has shown superior performance in multiple visual tasks and has become one of the mainstream models in the current deep learning field.

[0046] In addition to VGG and ResNet, CNN models have been proposed and applied in different scenarios. For example, the Inception model extracts multi-scale features by using convolution kernels of different sizes; the DenseNet model connects each layer to all previous layers to enhance feature propagation and reuse. Pre-trained CNN models provide powerful tools for the field of image processing, which can help us better understand and analyze image content.

[0047] For example, in the process of car insurance claims, using a pre-trained convolutional neural network (CNN) model to extract features from raw images is an effective method. These raw images usually contain elements such as cars and environmental objects, among which the car is the key subject. Through the CNN model, the hierarchical features in the image can be automatically learned, thereby accurately identifying the location and status of the car.

[0048] In practical applications, some classic CNN models, such as VGG, ResNet, etc., can be selected to extract features from the original image. These models have been pre-trained on a large number of data sets and have good generalization capabilities. By forward propagating the original image, the probability distribution of each pixel belonging to a certain category can be obtained, thereby achieving accurate positioning of the car.

[0049] In addition to cars, environmental objects in the original image can also be used as auxiliary information to help determine accident responsibility. For example, traffic signs, pedestrians, buildings, etc. may all be related to the accident. By extracting and analyzing the features of these environmental objects, we can have a more comprehensive understanding of the situation at the accident scene and provide strong support for claims decision-making.

[0050] Among them, Figure 3 As shown, in step S200, that is, using a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects, the following steps are included:

[0051] S210, identifying the core area and the secondary area of ​​the original image;

[0052] S220: Segment the core area and the secondary area, extract the core area, and obtain the key body.

[0053] For steps S210-S220, in the field of image processing and computer vision, identifying the core area and secondary area of ​​the original image is an important step. The core area usually refers to the most important part of the image, such as the face area in face recognition, or a specific object in object detection. The secondary area refers to those areas that have little impact on the current task, such as the background or other irrelevant objects.

[0054] In order to segment the core area from the secondary area and extract the key subject, we can use a variety of methods. One of the common methods is to use image segmentation techniques, such as threshold segmentation, region growing, etc. These methods can divide different regions according to the similarity or difference between pixels. Another common method is to use deep learning models, such as convolutional neural networks (CNN) or fully convolutional networks (FCN), to learn how to extract the key subject from the original image by training the model.

[0055] In practical applications, we usually combine multiple methods to achieve better results. For example, we can first use image segmentation technology to divide the image into multiple regions, and then use deep learning models to classify and identify these regions to accurately extract the key subjects. In addition, we can also use some prior knowledge or rules to guide the segmentation and extraction process to improve the accuracy and robustness of the algorithm.

[0056] For example, in the application field of auto insurance claims, accident photos are identified, that is, the core area and the secondary area of ​​the photos containing the car are identified, wherein when assessing the damage to the car, the car parts are defined as the core rest, and the parts of the original image that are not related to the car are defined as the secondary area, such as the people in the image, environmental weeds, etc., and then the core area and the secondary area are segmented, that is, the car image is segmented, wherein the damaged parts of the car are extracted to obtain the key subject.

[0057] S300: performing image segmentation on the key subject to obtain key areas and non-key areas.

[0058] In this embodiment, after obtaining the key subject, we need to perform image segmentation on the key subject to distinguish the key area from the non-key area. Image segmentation is the process of dividing an image into multiple areas, each area representing a specific object or scene. By segmenting the key subject from the non-key area, we can better understand the structural and semantic information in the image.

[0059] Image segmentation is an important technique in computer vision and image processing, which can help us decompose complex images into simpler parts, making it easier to analyze and process these parts. Image segmentation plays a vital role in many applications, such as medical imaging, remote sensing image analysis, and autonomous driving.

[0060] When performing image segmentation, we usually need to select the appropriate algorithm based on the specific features of the image. These features may include color, texture, shape, spatial relationship, etc. For example, in medical imaging, we may need to segment different organs based on the density or contrast of the tissue; in remote sensing image analysis, we may need to segment different terrain types based on the reflectivity or temperature of the surface; in autonomous driving, we may need to segment roads and obstacles based on the position and posture of the vehicle.

[0061] For example, in the vehicle damage assessment stage in the field of insurance claims, the key subject is the car, and the key area is the damaged part of the car. Since the damage assessment process requires an assessment of the damage level of the damaged part of the car before a corresponding claims plan can be issued, it is necessary to have a clear understanding of the key area, that is, the damaged part of the car, in order to issue an accurate claims plan.

[0062] In order to achieve effective image segmentation, we need to use some advanced techniques and methods. One of the commonly used methods is threshold segmentation, which divides the image into two parts by setting a threshold: one part is the pixels above the threshold, and the other part is the pixels below the threshold. Another common method is region growing, which starts from a seed point and gradually merges adjacent pixels until a continuous region is formed. In addition, there are some deep learning-based methods, such as fully convolutional networks (FCN) and U-Net, which can automatically learn the features of images and achieve high-precision segmentation.

[0063] like Figure 4 As shown, step S300, performing image segmentation on the key subject to obtain key areas and non-key areas, includes:

[0064] S310: Acquire a data set with the key area and the non-key area marked.

[0065] S320: Perform image segmentation on the key subject using a semantic algorithm according to the data set to obtain the key area and the non-key area.

[0066] In order to obtain a dataset with annotated key and non-key areas, we need to perform a series of steps. First, we need to collect a large amount of raw image data, which can come from various sources, such as public datasets, image resources on the Internet, or images collected by specific devices. Then, we need to preprocess these images, including denoising, contrast enhancement, cropping, etc., to improve image quality and reduce the difficulty of subsequent processing.

[0067] Next, we need to annotate the key subjects in the image. This can be done manually or with some automated tools. During the annotation process, we need to ensure that each key subject is accurately marked and clearly distinguished from non-key areas. At the same time, we also need to record the location information of each key subject for subsequent processing and analysis.

[0068] Once we have the dataset with labeled key and non-key regions, we can use semantic algorithms to segment images for key subjects based on this dataset. Semantic algorithms are a deep learning-based approach that automatically learns semantic information in images and divides images into different regions based on that information. By training a deep learning model, we can teach it how to extract features of key subjects from images and segment images into key and non-key regions based on those features.

[0069] During the training process, we need to use a large amount of annotated data to train the model to ensure that it can accurately identify the key subjects and correctly divide the regions. In addition, we also need to adjust the parameters and structure of the model to optimize its performance and improve the accuracy of segmentation. Through continuous iteration and optimization, we can eventually get a model that can effectively perform image segmentation on key subjects.

[0070] For example, in the vehicle damage assessment in the field of automobile claims, the key areas and non-key areas with annotations refer to the specific information of the damaged parts of the automobile and the specific information of the undamaged parts of the automobile, respectively. For example, the key areas with annotations refer to the areas where valuable parts of the damaged parts of the automobile are damaged, such as the engine and gearbox of the automobile, and the non-key areas with annotations, such as the undamaged car body and wheels, can be segmented by using the data set to segment the key subjects, so that the important factors related to the insurance claims process can be segmented more accurately. The insurance personnel can issue an accurate claims plan based on the segmented important factors. The important factors, in the field of automobile damage assessment, refer to the damaged areas of the automobile: engine, gearbox, etc.

[0071] S400: differentially compress the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image.

[0072] In image processing, differential compression is performed on the key area and the non-key area. By adjusting the compression parameters, different degrees of compression can be achieved for different areas, thereby reducing the amount of data while ensuring image quality.

[0073] As for the key areas, since they contain important information and details, these details need to be retained as much as possible. Therefore, a lower compression rate can be used to compress these areas. A lower compression rate means that more pixel values ​​will be retained during the compression process, thereby reducing information loss. This ensures that after decompression, the details of the key areas can be restored and the image quality can be guaranteed.

[0074] For non-critical areas, since the information they contain is relatively small or unimportant, a higher compression rate can be used to compress these areas. A higher compression rate means that more pixel values ​​will be discarded during the compression process, thereby further reducing the amount of data. Although this may result in some loss of information, this loss is acceptable for non-critical areas.

[0075] In order to achieve differential compression, it is necessary to select appropriate compression algorithms and technologies. Common compression algorithms include JPEG, PNG, GIF, etc., which all have their own advantages and disadvantages and applicable scenarios. When selecting an algorithm, we need to consider factors such as the characteristics of the image, the application scenario, and performance requirements. In addition, we can also achieve differential compression by adjusting the parameters of the algorithm. For example, in JPEG compression, we can control the compression rate by adjusting the quality factor (QF); in PNG compression, we can control the compression effect by adjusting the filter type and bit depth.

[0076] In summary, by differentially compressing the key area and the non-key area, the amount of data can be reduced while ensuring image quality. This method is very useful in many application scenarios, such as network transmission, storage optimization, etc. By selecting appropriate algorithms and techniques and adjusting corresponding parameters, we can achieve efficient and accurate differential compression.

[0077] like Figure 5 As shown, wherein S400, the key area and the non-key area are differentially compressed respectively to obtain a key area compressed image and a non-key area compressed image, including:

[0078] S410, compressing the key area using a lossless or low-loss compression algorithm to obtain a key area compressed image;

[0079] S420: Lossy compress the non-critical area and adjust the compression level of the non-critical area to obtain a compressed image of the non-critical area.

[0080] In image processing, in order to effectively reduce storage space and improve transmission efficiency while maintaining image quality in key areas, this application adopts a differentiated compression strategy, which involves applying different compression techniques to key areas and non-key areas of the image.

[0081] For the key areas, since they contain important information or details, such as lesions in medical images, important landmarks in satellite images, etc., we need to retain these details as much as possible. Therefore, lossless or low-loss compression of key areas is necessary. Lossless compression algorithms, such as PNG, FLIF or the lossless mode of JPEG 2000, can reduce file size without losing any information. Low-loss compression slightly sacrifices some image quality in exchange for a higher compression ratio, such as using JPEG with a high quality setting or lossy WebP. These methods ensure that the details of the key areas are fully restored after decompression, thereby ensuring the diagnostic value or analysis accuracy of the image.

[0082] For the non-critical areas, since the information they contain is relatively small or not so important, we can use lossy compression to further reduce the amount of data. Lossy compression algorithms, such as standard JPEG, achieve high compression ratios by removing high-frequency information (i.e. details) in the image. In this step, adjusting the compression level of non-critical areas is key. A higher compression level will significantly reduce the file size, but may introduce obvious compression artifacts (such as blocky images, blur, etc.), affecting the overall look and feel of the image. Therefore, it is necessary to balance the compression level based on actual needs and the acceptable degree of image quality loss.

[0083] Through the above steps, a differentially compressed image can be obtained, wherein the critical area maintains a high image quality, obtaining the critical area compressed image, and the non-critical area is compressed to different degrees according to the set compression level, obtaining the non-critical area compressed image. This method not only optimizes storage and transmission efficiency, but also meets the needs of key information protection. In practical applications, this strategy is particularly suitable for scenarios that require the transmission of optimized compressed images under limited bandwidth or storage resources, such as telemedicine, satellite remote sensing and other fields.

[0084] like Figure 6 As shown, in step S420, adjusting the compression level of the non-critical area to obtain the compressed image of the non-critical area includes:

[0085] S421, for the non-critical area, adjusting the compression parameters to obtain a preliminary compressed image;

[0086] S422: When it is determined that the storage value of the preliminary compressed image and the visual quality of the preliminary compressed image are at a balance point, output the non-critical area compressed image.

[0087] In the field of image processing and storage, it is crucial to effectively compress images to reduce storage space and improve transmission efficiency. For the non-critical areas, this goal can be achieved by adjusting the compression parameters. First, the image needs to be initially compressed, which usually involves selecting a suitable compression algorithm, such as JPEG or PNG, and setting an initial compression level. This compression level determines the compression ratio of the image and the final file size.

[0088] After obtaining the preliminary compressed image, it is necessary to evaluate the relationship between its storage value and visual quality. Storage value mainly refers to the storage space occupied by the compressed image file, while visual quality refers to the image details and overall aesthetics that can be perceived by the human eye. The ideal situation is to reduce the storage value as much as possible while maintaining acceptable visual quality. In order to achieve this balance, it may be necessary to adjust the compression parameters multiple times and conduct repeated testing and evaluation.

[0089] In this process, some objective image quality assessment indicators, such as peak signal-to-noise ratio (PSNR) or structural similarity index (SSIM), can be used to quantify the change in visual quality. At the same time, subjective visual inspection can also be performed to ensure that there is no visually obvious distortion or degradation in the compressed image.

[0090] When we find a suitable compression parameter that makes the storage value and visual quality in a reasonable balance, we can output the compressed image. This balance point may vary depending on the application scenario and user needs. For example, in web design, in order to speed up page loading, a higher compression ratio may be selected; in medical imaging, because as much detail information as possible needs to be retained, a lower compression ratio may be selected.

[0091] In summary, by adjusting the compression parameters of the non-critical area and finding a balance between storage value and visual quality, the image can be effectively compressed to obtain the non-critical area compressed image while ensuring its usability in practical applications. This method not only helps save storage resources, but also improves the transmission efficiency of the image and meets the needs of different scenarios.

[0092] S500: reconstruct the key area compressed image and the non-key area compressed image to restore the detail features of the key area and generate an optimized compressed image.

[0093] In the image processing process, lossless or low-loss compression is performed on the key areas, while lossy compression is used on the non-key areas, which is a common strategy for achieving efficient data transmission and storage. However, in order to accurately analyze and use image information at the receiving end, it is a crucial step to reconstruct the compressed image, restore the detailed features of the key areas, and generate an optimized compressed image.

[0094] The process of image reconstruction involves the use of mathematical and computational models to estimate the original image signal. For differentially compressed images, the focus of reconstruction is to restore as much detail as possible in key areas that were preserved during the initial compression process but may have been affected by the high compression ratio in non-key areas. By applying advanced image reconstruction algorithms such as iterative back projection (IBP), maximum a posteriori probability estimation (MAP), or deep learning methods, compression noise can be effectively reduced and image quality can be improved.

[0095] In the reconstruction process, the compressed image needs to be analyzed first to identify the key areas and non-key areas. Then, the appropriate reconstruction technology is selected according to the compression characteristics of different areas. For example, for key areas, edge-preserving reconstruction algorithms can be used to enhance details; while for non-key areas, smoothing can be used to reduce visual artifacts. In addition, machine learning technology can be used to train a model to predict and compensate for the information lost during the compression process.

[0096] After reconstruction, the image should be close to the original uncompressed image quality, especially in the details of key areas. Such optimized compressed images are of great significance for subsequent analysis, diagnosis or further processing. For example, in the field of medical imaging, accurate image reconstruction can help doctors better identify lesions; in satellite remote sensing, high-quality reconstructed images can provide more accurate environmental monitoring and resource assessment. Image reconstruction is a bridge connecting compressed images with practical applications. It not only restores the key information of the image, but also provides reliable data support for end users.

[0097] like Figure 7 As shown, in step S500, the compressed image of the key area and the compressed image of the non-key area are reconstructed to restore the detail features of the key area and generate an optimized compressed image, including:

[0098] S510, using the trained generative adversarial network model to restore the detailed features of the key area;

[0099] S520: Adjust the color difference between the key area and the non-key area to generate an optimized compressed image.

[0100] In the field of image processing, Generative Adversarial Networks (GANs) have become a powerful tool for restoring image details and improving image quality. In particular, when faced with differentially compressed images, GANs can effectively restore the detailed features of key areas while adjusting the color differences between key and non-key areas, thereby generating optimized compressed images.

[0101] Generative adversarial networks consist of two parts: a generator and a discriminator. The task of the generator is to create images that are as realistic as possible to deceive the discriminator, while the discriminator tries to distinguish between real images and images produced by the generator. Through this adversarial process, the generator gradually learns to capture complex patterns in the training data and is able to generate increasingly realistic images.

[0102] In the application of image reconstruction, a specially trained GAN model can be used to restore details in key areas. This model has learned how to extract and reconstruct detail features from similar key areas during the training phase. When a differentially compressed image is input, the generator will try to reconstruct details that may be lost or blurred during the compression process. Due to the powerful capabilities of GANs, they are able to infer reasonable details without explicit guidance, which is difficult to achieve with traditional image processing methods.

[0103] In addition to detail restoration, GANs can also be used to adjust the color distribution of an image. During differential compression, color distortion or unevenness may occur due to the high compression ratio of the non-critical areas. By training GANs to identify and correct these color deviations, the color of the entire image can be made more consistent and natural. This step is critical to ensuring the visual quality of the final image, especially in application scenarios that require accurate color information, such as medical imaging, satellite remote sensing, etc.

[0104] After completing the above steps, the image we obtained not only restores rich details in key areas, but also maintains color consistency and accuracy overall. Such optimized compressed images are of great value for subsequent analysis, diagnosis or further processing. For example, in the field of automobile damage assessment, accurate images can help insurance personnel make better claims solutions; in environmental monitoring, high-quality satellite images can provide more reliable data support.

[0105] In short, using the trained generative adversarial network model to restore the detailed features of the key areas of the image and adjust the color differences is an innovative and effective image reconstruction method. With the continuous advancement of deep learning technology, GANs will play a more important role in the field of image processing in the future, bringing us a higher quality image experience.

[0106] For example, in the process of adjusting the color difference between the key area and the non-key area, it is inevitable that there will be a color difference at the boundary between the key area and the non-key area during the image segmentation process. Adjusting the color difference between the key area and the non-key area can ensure that the optimized compressed image is clear and realistic enough.

[0107] The training step of generating the adversarial network model includes:

[0108] Initialize the generator and discriminator;

[0109] The generator generates a reconstructed image based on the compressed image;

[0110] The discriminator determines the similarity between the reconstructed image and the original image, and obtains a similar probability value;

[0111] The weights of the generator and the discriminator are updated respectively according to the probability value to repeatedly generate the reconstructed image until the generative adversarial network model converges, wherein the goal of the generator is to maximize the probability that the generated image is mistaken for the real one, and the goal of the discriminator is to maximize the ability to correctly distinguish between real and fake images.

[0112] Generative Adversarial Network (GAN) is a deep learning model that consists of two main parts: Generator and Discriminator. The training process of this model involves a dynamic "game" or adversarial process, in which the generator tries to create increasingly realistic fake images, while the discriminator strives to distinguish between real and fake images. The following are the detailed training steps:

[0113] Initialize the generator and discriminator: Before starting training, you need to initialize the parameters of the generator and discriminator. These parameters are usually the weights and biases in the neural network, which determine the behavior of each network. Initialization can be done randomly or based on prior knowledge.

[0114] Generator generates a reconstructed image based on a compressed image: The task of the generator is to receive input data (in this case, a compressed image) and output a reconstructed image. In the initial stage, the generator may only be able to produce low-quality images, but as training proceeds, it will gradually learn how to generate more and more realistic images.

[0115] The discriminator determines the similarity of the reconstructed image to the original image: The discriminator receives two inputs: one is the real, unmodified original image, and the other is the reconstructed image produced by the generator. Its goal is to distinguish between the two sources of images, that is, to determine whether a given image is real or created by the generator.

[0116] Derive similar probability values: The discriminator outputs a probability value for each input image, indicating the degree to which the image is considered to be real. This probability value is usually between 0 and 1, where 1 means that it is completely sure that it is real and 0 means that it is sure that it is fake.

[0117] Update the weights of the generator and discriminator based on the probability values: This is a key step in the training process. Based on the output of the discriminator, we can calculate a loss function that measures the performance of the discriminator. Then, the backpropagation algorithm is used to calculate the gradients with respect to the parameters of the generator and discriminator. These gradients are used to update the weights of the network to minimize the loss function.

[0118] Repeat the above process: The above process is repeated over multiple training cycles. In each cycle, the generator and discriminator are updated based on new data samples. Over time, the generator should be able to produce fake images that are increasingly difficult to distinguish from real images, while the discriminator should become better at distinguishing real from fake images.

[0119] GAN model convergence: When the performance of the generator and the discriminator reaches a certain balance, the model is considered to have converged. This means that the generator has learned to generate high-quality images that the discriminator cannot reliably distinguish whether they are real or not. At this point, the generator has successfully fooled the discriminator into incorrectly classifying fake images as real images.

[0120] The goal of the generator: Throughout the training process, the goal of the generator is to maximize the probability that its output is mistaken for real. In other words, it tries to create a distribution such that samples drawn from this distribution are indistinguishable from the true data distribution.

[0121] Objective of the Discriminator: Meanwhile, the discriminator’s objective is to maximize its ability to correctly distinguish between real and fake images. It tries to learn a function that accurately maps an input to a label indicating whether the input is real or generated.

[0122] In this way, the Generative Adversarial Network establishes a competitive relationship between the generator and the discriminator, prompting both to continuously improve their performance. The end result is a powerful model that can generate high-quality images in a variety of tasks, such as artistic creation, data enhancement, image restoration, etc.

[0123] It can be seen that in the above scheme, for the insurance claim scenario, firstly, the original image is obtained, and the pre-trained convolutional neural network model is used to extract the features of the original image to obtain the key subject; the key subject is segmented to obtain the key area and the non-key area; the key area and the non-key area are differentially compressed to obtain a compressed image; the compressed image is reconstructed to restore the detail features of the key area and generate an optimized compressed image. The present invention adopts a differential compression method to significantly reduce the image file size while maintaining the clarity of key details, effectively reducing storage and transmission costs.

[0124] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0125] In one embodiment, an image compression device based on artificial intelligence is provided, and the image compression device based on artificial intelligence corresponds one-to-one to the image compression method in the above embodiment. Figure 8 As shown, the image compression device includes an acquisition module 101, an extraction module 102, a segmentation module 103, a compression module 104 and a reconstruction module 105. The functional modules are described in detail as follows:

[0126] An acquisition module 101 is used to acquire an original image;

[0127] An extraction module 102 is used to extract features from the original image using a pre-trained convolutional neural network model to obtain key subjects;

[0128] A segmentation module 103 is used to perform image segmentation on the key subject to obtain a key area and a non-key area;

[0129] A compression module 104, configured to differentially compress the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image;

[0130] The reconstruction module 105 is used to reconstruct the key area compressed image and the non-key area compressed image, restore the detail features of the key area, and generate an optimized compressed image.

[0131] In one embodiment, the extraction module 102 is specifically configured to:

[0132] identifying a core region and a secondary region of the original image;

[0133] The core area is segmented from the secondary area, the core area is extracted, and the key subject is obtained.

[0134] In one embodiment, the segmentation module 103 is specifically configured to:

[0135] Obtain a dataset with the key area and the non-key area marked

[0136] The key subject is segmented according to the data set using a semantic algorithm to obtain the key area and the non-key area.

[0137] In one embodiment, the compression module 104 is further configured to:

[0138] Compressing the key area using a lossless or low-loss compression algorithm to obtain a key area compressed image;

[0139] The non-critical area is lossily compressed, and the compression level of the non-critical area is adjusted to obtain a compressed image of the non-critical area.

[0140] In one embodiment, the reconstruction module 105 is specifically configured to:

[0141] Restoring detailed features of the key area using the trained generative adversarial network model;

[0142] The color difference between the key area and the non-key area is adjusted to generate an optimized compressed image.

[0143] In one embodiment, the compression module 104 is specifically configured to:

[0144] For the non-critical area, adjusting the compression parameters to obtain a preliminary compressed image;

[0145] When it is determined that the storage value of the preliminary compressed image and the visual quality of the preliminary compressed image are at a balance point, the non-critical area compressed image is output.

[0146] The present invention provides an image compression device, which first obtains the original image, uses a pre-trained convolutional neural network model to extract features from the original image, and obtains a key subject; performs image segmentation on the key subject to obtain a key area and a non-key area; differentially compresses the key area and the non-key area to obtain a compressed image; and reconstructs the compressed image to restore the detail features of the key area and generate an optimized compressed image. The present invention adopts a differential compression method to significantly reduce the image file size while maintaining the clarity of key details, effectively reducing storage and transmission costs.

[0147] For the specific definition of the image compression device, please refer to the definition of the image compression method above, which will not be repeated here. Each module in the above-mentioned image compression device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0148] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig. 9 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps of a server-side image compression method based on artificial intelligence.

[0149] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, it implements the functions or steps of the client side of an image compression method

[0150] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0151] Get the original image;

[0152] Using a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects;

[0153] Performing image segmentation on the key subject to obtain a key area and a non-key area;

[0154] Differentially compressing the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image;

[0155] Image reconstruction is performed on the key area compressed image and the non-key area compressed image to restore the detail features of the key area and generate an optimized compressed image.

[0156] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0157] Get the original image;

[0158] Using a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects;

[0159] Performing image segmentation on the key subject to obtain a key area and a non-key area;

[0160] Differentially compressing the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image;

[0161] Image reconstruction is performed on the key area compressed image and the non-key area compressed image to restore the detail features of the key area and generate an optimized compressed image.

[0162] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0163] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0164] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0165] The above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should be included in the protection scope of the present invention. It should be noted that if software tools or components other than those of the company appear in the embodiments of this application, they are only used for example introduction and do not represent actual use.

Claims

1. An image compression method, characterized in that: include: Get the original image; Using a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects; Performing image segmentation on the key subject to obtain a key area and a non-key area; Differentially compressing the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image; Image reconstruction is performed on the key area compressed image and the non-key area compressed image to restore the detail features of the key area and generate an optimized compressed image.

2. The image compression method according to claim 1, characterized in that: The use of a pre-trained convolutional neural network model to extract features from the original image to obtain key subjects includes: identifying a core region and a secondary region of the original image; The core area is segmented from the secondary area, the core area is extracted, and the key subject is obtained.

3. The image compression method according to claim 1, wherein: The image segmentation of the key subject to obtain the key area and the non-key area includes: Acquire a data set with the key area and the non-key area marked; The key subject is segmented according to the data set using a semantic algorithm to obtain the key area and the non-key area.

4. The image compression method according to claim 1, wherein: The differential compression of the key area and the non-key area to obtain a key area compressed image and a non-key area compressed image includes: Compressing the key area using a lossless or low-loss compression algorithm to obtain a key area compressed image; The non-critical area is lossily compressed, and the compression level of the non-critical area is adjusted to obtain a compressed image of the non-critical area.

5. The image compression method according to claim 1, wherein: Reconstructing the compressed image of the key area and the compressed image of the non-key area to restore the detail features of the key area and generate an optimized compressed image includes: Restoring detailed features of the key area using the trained generative adversarial network model; The color difference between the key area and the non-key area is adjusted to generate an optimized compressed image.

6. The image compression method according to claim 4, characterized in that: The step of adjusting the compression level of the non-critical area to obtain the compressed image of the non-critical area includes: For the non-critical area, adjusting the compression parameters to obtain a preliminary compressed image; When it is determined that the storage value of the preliminary compressed image and the visual quality of the preliminary compressed image are at a balance point, the non-critical area compressed image is output.

7. The image compression method according to claim 5, characterized in that: The training steps of the adversarial network model include: Initialize the generator and discriminator; The generator generates a reconstructed image based on the compressed image; The discriminator determines the similarity between the reconstructed image and the original image, and obtains a similar probability value; The weights of the generator and the discriminator are updated respectively according to the probability value to repeatedly generate the reconstructed image until the generative adversarial network model converges, wherein the goal of the generator is to maximize the probability that the generated image is mistaken for the real one, and the goal of the discriminator is to maximize the ability to correctly distinguish between real and fake images.

8. An image compression device, characterized in that: include: An acquisition module, used for acquiring the original image; An extraction module, used for extracting features from the original image using a pre-trained convolutional neural network model to obtain key subjects; A segmentation module, used for performing image segmentation on the key subject to obtain a key area and a non-key area; A compression module, used for differentially compressing the key area and the non-key area to obtain a compressed image; The reconstruction module is used to reconstruct the compressed image, restore the detail features of the key area, and generate an optimized compressed image.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the image compression method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the image compression method according to any one of claims 1 to 7 are implemented.

Images