E-commerce picture beautifying method and device, equipment and medium

Abstract

The invention relates to an e-commerce picture beautifying method and device, equipment and medium in the technical field of e-commerce, and the method comprises the steps: responding to a picture beautifying event triggered by a user through employing a target e-commerce picture, and employing a preset interest segmentation model to segment a buyer interest image and a non-buyer interest image corresponding to the target e-commerce picture; a preset image super-division model is adopted to improve the image resolution of the buyer interest image, and a corresponding super-division buyer interest image is obtained; determining the smoothness of a splicing edge in the to-be-verified e-commerce picture obtained by splicing the super buyer interest image and the non-buyer interest image; and when the smoothness does not reach the standard, performing smooth transition processing on the splicing edge in the to-be-verified e-commerce picture, and pushing the beautified e-commerce picture obtained after the processing is completed to the user. According to the method, the low-resolution e-commerce picture can be beautified, so that the image content interested by the buyer is highlighted, and the whole is clear and natural.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of e-commerce, and in particular to an e-commerce picture beautification method and a corresponding device, computer equipment and computer readable storage medium. BACKGROUND

[0002] In the e-commerce platform, many operation links have frequent and large use requirements for picture materials. For example, in actual operation, the merchant often faces the dilemma of insufficient resolution of the original picture. The causes of this problem are various. It may be that the limitation of the shooting device causes the picture to not achieve the ideal clarity in the initial generation. It may also be that the historical picture materials accumulated at that time due to the limitation of the shooting and technical conditions at that time are of poor quality. Another case is that the picture itself obtained from the supplier has quality problems. When these low-resolution pictures are displayed on the e-commerce platform, the problem will be highlighted. There will be obvious blurring on the high-definition screen, and the details of the goods will be severely lost, which greatly affects the presentation effect of the goods and has a negative effect on the purchasing experience of consumers.

[0003] In the current market environment where high-definition display screens have become mainstream and users' requirements for picture quality are constantly improving, the picture quality of the displayed picture is closely related to the click rate and conversion rate of the buyer. Low-quality pictures directly lead to a downward trend in these two aspects. At present, although some traditional picture processing technologies are applied to solve this problem, these technologies have certain limitations. For example, some traditional image enlargement technologies can increase the size of the picture to a certain extent, but the effect on improving the resolution is limited. They cannot effectively restore the lost details in the picture, and the processed picture still cannot meet the demand for high-definition pictures. In addition, some technologies may change the original content of the picture during processing, causing deviations in the color, contrast and other aspects of the picture, affecting the authenticity and aesthetics of the picture. It can be seen that in actual application, the problem of insufficient picture resolution faced by the e-commerce platform cannot be well solved. SUMMARY

[0004] The primary purpose of the present application is to solve at least one of the above problems and provide an e-commerce picture beautification method and a corresponding device, computer equipment and computer program product.

[0005] To meet the various purposes of the present application, the present application adopts the following technical solutions:

[0006] An e-commerce picture beautification method provided to adapt to one of the purposes of the present application includes the following steps:

[0007] In response to a picture beautification event triggered by a user applying a target e-commerce picture, a preset interest segmentation model is used to segment the corresponding buyer interest image and non-buyer interest image of the target e-commerce picture.

[0008] A preset image super-resolution model is used to enhance the image resolution of the buyer's interest image, thereby obtaining a corresponding super-resolution buyer interest image;

[0009] Determine the smoothness of the stitching edges in the e-commerce image to be verified, obtained by stitching together the super-resolution buyer interest image and the non-buyer interest image.

[0010] When the smoothness does not meet the standard, the splicing edges in the e-commerce image to be verified are smoothed, and the beautified e-commerce image obtained after processing is pushed to the user.

[0011] On the other hand, an e-commerce image enhancement device provided to meet one of the purposes of this application includes an event response module, an image super-resolution module, a smoothing determination module, and an enhancement push module. The event response module is used to respond to an image enhancement event triggered by a user's application of a target e-commerce image, and uses a preset interest segmentation model to segment the target e-commerce image into a buyer interest image and a non-buyer interest image. The image super-resolution module is used to use a preset image super-resolution model to increase the image resolution of the buyer interest image, obtaining a corresponding super-resolution buyer interest image. The smoothing determination module is used to determine the smoothness of the stitching edges in the e-commerce image to be verified obtained by stitching the super-resolution buyer interest image and the non-buyer interest image. The enhancement push module is used to perform smoothing transition processing on the stitching edges in the e-commerce image to be verified when the smoothness is insufficient, and push the processed enhanced e-commerce image to the user.

[0012] On another front, a computer device provided for one of the purposes of this application includes a central processing unit and a memory, wherein the central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the e-commerce image enhancement method described in this application.

[0013] In another aspect, a computer program product provided for another purpose of this application includes a computer program / instructions that, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0014] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0015] This application first employs an interest segmentation model to divide target e-commerce images into images of buyer interest and images of non-buyer interest. Then, it performs super-resolution enhancement only on areas that buyers are truly interested in. This avoids the computational waste of indiscriminately enlarging the entire image while achieving maximum detail restoration in the core display areas. For e-commerce platforms, this means that even with limited server computing power, they can still obtain product main images sufficient for high-definition screens with the most efficient and less resource-intensive method, directly improving the clarity of the product display on the page and significantly enhancing the user's first visual experience.

[0016] Secondly, when stitching the over-resolution buyer interest image with the non-buyer interest image (which maintains its original resolution) at the edges, if there is a visually jarring seam, the smoothness will be automatically detected, and transition processing will be applied to areas that do not meet the standards. This preserves the original authenticity of the primary and secondary areas of buyer interest while eliminating abrupt stitching marks, resulting in a final output image that appears natural, coherent, and without any artificial artifacts. This effectively avoids problems such as color shift and contrast distortion that are easily caused by traditional full-image enlargement, ensuring accurate color reproduction of the product.

[0017] Furthermore, because the areas of interest to buyers are enhanced with targeted high resolution, the image content that buyers are interested in is clearly presented, directly attracting their attention and enabling them to quickly and intuitively obtain sufficient decision-making information. This allows for faster buyer acquisition and facilitates the progress of product transactions. For merchants, this means obtaining high-quality image materials without the need for reshoots or high-end shooting equipment, greatly reducing operating costs. For the platform, a unified algorithm service improves the image quality across the entire site, ensuring a better platform service experience for both merchants and buyers. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0019] Figure 1 The network architecture of the e-commerce platform is an example of that used in this application;

[0020] Figure 2 This is a flowchart illustrating a typical embodiment of the e-commerce image enhancement method of this application;

[0021] Figure 3 This is a schematic diagram of the e-commerce image enhancement device of this application;

[0022] Figure 4 This is a schematic diagram of the structure of a computer device used in this application. Detailed Implementation

[0023] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0024] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0025] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0026] like Figure 1 In the network architecture shown, the e-commerce platform 82 is deployed on the Internet to provide corresponding services to its users. Similarly, the devices 80 of the merchant users and the devices 81 of the consumer users of the e-commerce platform 82 are also connected to the Internet to use the services provided by the e-commerce platform.

[0027] An exemplary e-commerce platform 82 provides supply and demand matching of products and / or services to the general public through the Internet infrastructure. In e-commerce platform 82, products and / or services are provided as commodity information. For the sake of simplicity, the concepts of commodity and product are used in this application to refer to the products and / or services in e-commerce platform 82. Specifically, these may be physical products, digital products, tickets, service subscriptions, other offline services, etc.

[0028] In reality, various entities can access e-commerce platform 82 as users and utilize its online services to participate in the business activities facilitated by the platform. These entities can be natural persons, legal persons, or social organizations. Corresponding to the two types of entities in business activities—merchants and consumers—e-commerce platform 82 has two corresponding categories of users: merchant users and consumer users. Entities involved in the product distribution chain in business activities, including manufacturers, sellers, retailers, and logistics providers, can all use online services on e-commerce platform 82 as merchant users. Similarly, consumers in business activities, including actual or potential consumers, can use online services on e-commerce platform 82 as consumer users. In actual business activities, the same entity can operate as both a merchant user and a consumer user; this should be interpreted flexibly.

[0029] The infrastructure used to deploy the e-commerce platform 82 mainly includes the backend architecture and frontend devices. The backend architecture runs various online services through a service cluster, including middleware or frontend services for the platform, services for consumers, and services for merchants, to enrich and improve its service functions. The frontend devices mainly cover the terminal devices used by users as clients to access the e-commerce platform 82, including but not limited to various mobile terminals, personal computers, and point-of-sale devices. For example, merchant users can use their terminal device 80 to enter product information for their online stores or use the interfaces opened by the e-commerce platform to generate their product information; consumer users can use their terminal device 81 to access the webpage of the online store implemented by the e-commerce platform 82, trigger the shopping process by clicking the shopping button provided on the webpage, and call various online services provided by the e-commerce platform 82 during the shopping process to achieve the purpose of placing an order.

[0030] In some embodiments, the e-commerce platform 82 may be implemented via a processing facility including a processor and memory, which stores a set of instructions that, when executed, cause the e-commerce platform 82 to perform the e-commerce and support functions as described in this application. The processing facility may be part of a server, client, network infrastructure, mobile computing platform, cloud computing platform, fixed computing platform, or other computing platform, and may provide electronic components, merchant devices, payment gateways, application developers, marketing channels, transportation providers, customer devices, point-of-sale devices, etc., for the e-commerce platform 82.

[0031] E-commerce platform 82 can provide online services such as cloud computing services, Software as a Service (SaaS), Infrastructure as a Service (IaaS), Platform as a Service (PaaS), Desktop as a Service (DaaS), Hosted Software as a Service, Mobile Backend as a Service (MBaaS), and Information Technology Management as a Service (ITMaaS). In some embodiments, the various functional components of e-commerce platform 82 can be implemented to operate on various platforms and operating systems. For example, for an online store, its administrator user enjoys the same or similar functions regardless of whether it is on iOS, Android, HomonyOS, or a web page.

[0032] E-commerce platform 82 enables merchants to create their own independent websites to run their online stores. It provides merchants with corresponding business management engine instances, allowing them to establish, maintain, and operate one or more online stores across these independent websites. The business management engine instance can be used for content management, task automation, and data management for one or more online stores. It can be configured through interfaces or built-in components to support various specific business processes in the online store, supporting business activities. Independent websites are the infrastructure of e-commerce platform 82, which offers cross-border services. Merchants can maintain their online stores relatively independently and centrally based on these independent websites. Independent websites typically have dedicated domain names and storage space, and different independent websites are relatively independent. E-commerce platform 82 can provide standardized or customized technical support for a large number of independent websites, allowing merchants to customize a business management engine instance that suits their needs and use it to maintain one or more online stores.

[0033] Online stores can be configured and maintained in the backend by merchant users logging into their Business Management Engine instance as administrators. Supported by the various online services provided by the e-commerce platform 82's infrastructure, merchant users can configure various functions within their online stores and view various data as administrators. For example, merchant users can manage various aspects of their online stores, such as viewing recent online store activities, updating the online store's product catalog, managing orders, recent visit activity, and total order activity. Merchant users can also view more detailed information about their business and visitors to their online store by obtaining reports or metrics, such as displaying a sales summary of the merchant's overall business, specific sales and engagement data from promotional sales and marketing channels, etc.

[0034] E-commerce platforms 82 can provide communication facilities and associated merchant interfaces for electronic communication and marketing. For example, they can utilize electronic messaging aggregation facilities to collect and analyze communication interactions between merchants, consumers, merchant devices, customer devices, point-of-sale devices, etc., aggregating and analyzing communications to increase the potential for product sales. For instance, a consumer may have product-related questions, which could lead to a dialogue between the consumer and the merchant (or an automated processor-based agent representing the merchant), where the communication facilities handle the interaction and provide the merchant with analysis on how to increase the probability of a sale.

[0035] In some embodiments, applications suitable for installation on terminal devices can be provided to serve the access needs of different users, enabling various users to access the e-commerce platform 82 by running the application on their terminal devices. Examples include the merchant backend module of online stores within the e-commerce platform 82. During the process of conducting business activities through these functions, the e-commerce platform 82 can implement various functions related to business activities as middleware or online services and expose corresponding interfaces. Then, toolkits corresponding to the interface access functions are embedded into the application to achieve functional expansion and task completion. The business management engine can include a series of basic functions and expose these functions to online services and / or applications via APIs. Online services and applications use the corresponding functions by remotely calling the corresponding APIs.

[0036] With the support of various components of the Business Management Engine instance, E-commerce Platform 82 can provide online shopping functionality, enabling merchants to connect with customers in a flexible and transparent manner. Consumers can select items online, create orders, provide delivery addresses in the orders, and complete payment confirmation. Merchants can then review and complete or cancel orders.

[0037] The e-commerce image enhancement method of this application can be programmed into a computer program product and deployed on a client or server for execution. For example, in an exemplary application scenario of this application, it can be deployed on the server of an e-commerce customer service platform. In this way, the method can be executed by human-computer interaction with the process of the computer program product through a graphical user interface by accessing the interface opened after the computer program product is running.

[0038] Please see Figure 2 The e-commerce image enhancement method of this application, in its typical embodiment, includes the following steps:

[0039] Step S1100: Respond to the image enhancement event triggered by the user application's target e-commerce image, and use a preset interest segmentation model to segment the target e-commerce image into buyer interest images and non-buyer interest images.

[0040] E-commerce platforms have both buyers and merchants, and both utilize e-commerce images. It's easy to understand that e-commerce images are both the "first impression" for buyers making purchasing decisions and the "primary productivity" for merchants to showcase their products. Therefore, the triggering scenarios for "image enhancement events" are diverse and frequent. For buyers, typical scenarios include, but are not limited to: after becoming interested in a product recommended in a live stream or short video playback area on an e-commerce platform, taking a screenshot of the product displayed in the video, resulting in a blurry image due to insufficient resolution; and uploading real-life product images to buyer show communities or buyer review areas on e-commerce platforms, where the imaging resolution supported by the camera equipment and / or software is limited. For merchants, typical scenarios include, but are not limited to: providing product listing images for display on the product details page during the product listing stage; and providing product promotion images for display on advertising channels during the advertising campaign stage.

[0041] Whether serving buyers' immediate needs for "clear visibility and highlighting key information" or merchants' operational needs for "increasing conversion rates and clicks while reducing bounce rates," any user on an e-commerce platform can actively use the platform's "one-click enhancement" or "intelligent enhancement" functions when using e-commerce images in specific application scenarios. Alternatively, the platform can automatically detect if the used e-commerce image does not match the corresponding resolution requirements of the application scenario and trigger an image enhancement event to beautify the used e-commerce image as the target e-commerce image. The platform's "one-click enhancement" or "intelligent enhancement" functions can manifest as controls in the graphical user interface or components triggered by specified user voice or actions.

[0042] The interest segmentation model is pre-trained to convergence, learning to segment images representing buyer interests and non-buyer interests from input e-commerce images. When an image enhancement event is triggered, the e-commerce platform's server responds by calling the interest segmentation model with the target e-commerce image as input. First, it extracts color, texture, edge, and high-order semantic features through multiple convolutional layers to form a multi-scale feature map. Then, based on these features, it densely generates candidate boxes globally. After performing non-maximum suppression on overlapping regions, it retains anchor boxes representing buyer interests and non-buyer interests. Subsequently, pixel-level classification is performed again within each anchor box, assigning a label for each pixel representing buyer interest or non-buyer interest, constructing continuous pixel-level masks. Finally, all masks are merged to output a complete buyer interest image and its corresponding non-buyer interest image in one go.

[0043] For interest segmentation models, YOLO series models such as YOLOV8 are recommended. Other models that are also suitable for object detection and image instance segmentation, such as U-Net, DeepLab series, BiSeNet, etc., can also be used. Those skilled in the art can choose to implement them as needed.

[0044] For the pre-prepared training set used to train the interest segmentation model, specifically, multiple original e-commerce images can be crawled from the offline image library of e-commerce platforms as training samples. These images cover diverse application scenarios, including various product images on product detail pages, product header images, product images captured during live streams and short videos, and product images from buyer reviews and ratings. This ensures that any one or more of the key image content elements that buyers are interested in, such as the product itself, its selling points, the model, the usage scenario, and the functional demonstration, appear fully in each training sample. For the supervision labels of each training sample, annotators can use polygon or brush tools to precisely delineate the key image content elements that buyers are interested in at the pixel level. The remaining image content elements represent those that buyers are not interested in. These are then labeled with a binary mask, where 1 represents the set of pixels of interest to the buyer, and 0 represents the set of other image content elements. Finally, all training samples and their supervision labels are compiled into a training set. The training process of the interest segmentation model can be flexibly implemented by those skilled in the art.

[0045] Step S1200: Use a preset image super-resolution model to improve the image resolution of the buyer interest image to obtain a corresponding super-resolution buyer interest image;

[0046] The image super-resolution model is pre-trained to convergence, learning the ability to enhance the resolution of input e-commerce images. Therefore, the model can be used with an image representing the buyer's interest as input, upscaling it from a low resolution to a high resolution, resulting in a highly detailed and textured super-resolution image representing the buyer's interest. Real-ESRGAN is recommended as the image super-resolution model, but other equally suitable models such as EDSR, ESRGAN, SRCNN, FSRCNN, and SRGAN can also be used.

[0047] For a pre-prepared training set used to train the image super-resolution model, in one embodiment, an interest segmentation model trained to convergence can be applied to accurately segment buyer interest images from each training sample within the training set. Then, each buyer interest image undergoes resolution upscaling and resolution downscaling respectively. For each buyer interest image, the image obtained after resolution upscaling is used as a supervision label, and the image obtained after resolution downscaling is used as a training sample corresponding to that supervision label. Finally, all training samples and their supervision labels are aggregated into a training set. The training process of the image super-resolution model can be implemented by those skilled in the art through modifications.

[0048] Step S1300: Determine the smoothness of the stitching edges in the e-commerce image to be verified obtained by stitching the super-resolution buyer interest image and the non-buyer interest image.

[0049] Determining the smoothness of the stitching edge involves objectively quantifying the boundary region after the over-resolution buyer interest image and the non-buyer interest image are re-fused into a complete e-commerce image to be verified. This boundary region refers to the stitching edge, which is the set of all pixels directly adjacent to the pixel regions of the over-resolution buyer interest image and the non-buyer interest image on a two-dimensional plane. First, the pixel matrices of the two images are aligned according to spatial coordinates, ensuring complete geometric overlap. Then, only the overlapping transition region is sampled and analyzed. To measure whether sharp breaks, brightness jumps, or texture misalignments exist in this region, a repeatable smoothness index can be used to quantify visual differences numerically. The quantified smoothness is compared with a preset threshold; if it exceeds the preset threshold, the smoothness is considered substandard; if it does not exceed the preset threshold, the smoothness is considered satisfactory.

[0050] Commonly used smoothness metrics include gradient magnitude difference, Laplacian energy, edge strength variance, SSIM local values, and edge density based on the Sobel operator. Those skilled in the art can choose one or combine multiple metrics, and can also assign weights to different metrics and calculate weighted scores. In one embodiment, the gradient magnitude difference metric is used. First, the Sobel operator is used to calculate the horizontal gradient Gx and vertical gradient Gy for each pixel in the stitching edge and its neighborhood. Then, the gradient magnitude G of each pixel is calculated as Gx. 2 With Gy 2 The sum of the values ​​is then squared. Next, a fixed-width (e.g., 8 pixels) buffer is taken on both sides of the stitching edge, and the average gradient magnitude μ and standard deviation σ of all pixels within the buffer are calculated. If the average μ is higher than the empirical threshold Tμ (e.g., 30), or the standard deviation σ is higher than the empirical threshold Tσ (e.g., 15), then the edge is considered to have a significant abrupt change, and the smoothness does not meet the standard; otherwise, it meets the standard. Each empirical threshold can be determined through feedback from the MOS (Subjective Mean Score) of an online A / B experiment.

[0051] Step S1400: When the smoothness does not meet the standard, perform smooth transition processing on the splicing edges in the e-commerce image to be verified, and push the beautified e-commerce image obtained after processing to the user.

[0052] At this point, a smooth transition is applied to the splicing edges. The splicing edges appear as one or more transition bands in a closed or open form on a two-dimensional plane. In one embodiment, firstly, a buffer zone with a width of N pixels is defined on both sides of the splicing edge. The value of N typically ranges from 4 to 16 pixels, and the specific value can be dynamically adjusted according to the image resolution; the higher the resolution, the larger N becomes. Those skilled in the art can configure it as needed. Subsequently, a bilateral filtering algorithm is used to perform nonlinear filtering on the pixels within the buffer zone. This algorithm reduces local differences in brightness and color while maintaining the edge structure. The spatial domain standard deviation σs of the filtering kernel typically ranges from 3 to 5 pixels, and the pixel value domain standard deviation σr typically ranges from 0.1 to 0.3 times the dynamic range of pixel values. If significant step-like artifacts are still detected after bilateral filtering, the Laplacian pyramid fusion method is further applied. The super-resolution buyer interest image and the non-buyer interest image are decomposed into multiple (e.g., 5) Laplacian pyramids. At each layer, the coefficients at corresponding positions are linearly weighted and fused. The weights smoothly transition from 0 to 1 at the stitching edges, with the length of the transition curve equal to the buffer width. Multiple (e.g., 3rd order) Hermitian interpolation is used to ensure the continuity of the first derivative. After fusion, the synthesized image undergoes inverse pyramid reconstruction to obtain a beautified e-commerce image with naturally transitioning edges. The resulting beautified e-commerce image can then be pushed to users, completing the beautification of the target e-commerce image.

[0053] As can be seen from the typical embodiments of this application, the technical solution of this application has many advantages, including but not limited to the following aspects:

[0054] This application first employs an interest segmentation model to divide target e-commerce images into images of buyer interest and images of non-buyer interest. Then, it performs super-resolution enhancement only on areas that buyers are truly interested in. This avoids the computational waste of indiscriminately enlarging the entire image while achieving maximum detail restoration in the core display areas. For e-commerce platforms, this means that even with limited server computing power, they can still obtain product main images sufficient for high-definition screens with the most efficient and less resource-intensive method, directly improving the clarity of the product display on the page and significantly enhancing the user's first visual experience.

[0055] Secondly, when stitching the over-resolution buyer interest image with the non-buyer interest image (which maintains its original resolution) at the edges, if there is a visually jarring seam, the smoothness will be automatically detected, and transition processing will be applied to areas that do not meet the standards. This preserves the original authenticity of the primary and secondary areas of buyer interest while eliminating abrupt stitching marks, resulting in a final output image that appears natural, coherent, and without any artificial artifacts. This effectively avoids problems such as color shift and contrast distortion that are easily caused by traditional full-image enlargement, ensuring accurate color reproduction of the product.

[0056] Furthermore, because the areas of interest to buyers are enhanced with targeted high resolution, the image content that buyers are interested in is clearly presented, directly attracting their attention and enabling them to quickly and intuitively obtain sufficient decision-making information. This allows for faster buyer acquisition and facilitates the progress of product transactions. For merchants, this means obtaining high-quality image materials without the need for reshoots or high-end shooting equipment, greatly reducing operating costs. For the platform, a unified algorithm service improves the image quality across the entire site, ensuring a better platform service experience for both merchants and buyers.

[0057] In a further embodiment, before step S1100, which responds to the image enhancement event triggered by the user application's target e-commerce image, the following steps are included:

[0058] Step S2100: Respond to the user's product listing event and obtain the corresponding product listing image;

[0059] Typically, merchants on e-commerce platforms list their products in their online stores. This usually requires the user to provide relevant product information, including product images and descriptions. Product images are typically used to visually showcase the product's selling points and / or appearance. Product descriptions refer to all textual information describing the product, including the product title, attributes, inventory, and price.

[0060] When a merchant submits product images for listing on an e-commerce platform by enabling an image upload control or API interface, a product listing event is triggered. The e-commerce platform's server responds to this event and obtains the product images for listing.

[0061] Step S2110: Use a preset interest assessment model to determine the buyer interest rating corresponding to the product images on the shelves;

[0062] The server can invoke a pre-trained, converged interest assessment model to analyze product images and determine corresponding buyer interest scores. This score is a quantifiable value representing the potential of a given e-commerce image to attract buyer attention or increase their purchase intention; a higher score indicates greater buyer interest. This interest assessment model has been trained to predict the degree of buyer interest in input e-commerce images.

[0063] The VGG model is recommended for interest evaluation. Other models suitable for binary classification tasks in the CV (computer vision) field, such as MobileNet and ResNet, can also be used. Those skilled in the art can flexibly choose and implement them.

[0064] For the pre-prepared training set used to train the interest assessment model, in one embodiment, images of multiple products listed on an e-commerce platform can be collected, with each image serving as a training sample. For the supervision labels of each training sample, annotators can check whether the image resolution of the key image elements that buyers focus on in each training sample is sufficient at the pixel level. This allows buyers to quickly understand the product's selling points and appearance, helping them make a quick purchase decision. If the resolution is sufficient, the supervision label for that training sample is set to 1; otherwise, it is set to 0. Finally, all training samples and their supervision labels are aggregated into a training set. The training process of the interest assessment model can be flexibly adapted by those skilled in the art.

[0065] Step S2120: When the buyer's interest rating does not meet the exposure effectiveness conditions, confirm that the listed product image is the target e-commerce image to trigger the corresponding image beautification event.

[0066] Exposure effectiveness criteria can be based on pre-set quantitative thresholds to measure whether product images are large enough on product detail pages, search results pages, or advertising placements to achieve the expected visual appeal and conversion effect for buyers. For example, the success criterion for exposure effectiveness could be set as "buyer interest rating ≥ 0.75" as the benchmark. A higher rating indicates clearer image content and a more prominent subject, increasing the likelihood of stimulating buyer clicks; conversely, a rating below 75 is considered unsatisfactory.

[0067] When the calculated buyer interest rating fails to meet the exposure effectiveness criteria, it means that the currently uploaded product image has deficiencies in dimensions such as pixel clarity and background interference that attract buyer attention, failing to effectively showcase the product's selling points and appearance on a high-definition screen. In this case, the uploaded product image is identified as a target e-commerce image requiring enhancement, immediately triggering an image enhancement event. Through subsequent interest segmentation, local super-resolution, and edge smoothing processes, the clarity of the image content that buyers are interested in is specifically improved, thereby ensuring that the image can quickly capture buyer attention after exposure and reduce bounce rate.

[0068] If the buyer's interest rating already meets the exposure criteria, it indicates that the product image already has sufficient visual impact and can be used directly for product display without additional processing. In this case, there's no need to consume computing power for image enhancement; the image can be uploaded as is, saving server resources and avoiding the risk of distortion caused by over-processing.

[0069] In this embodiment, by introducing a "merchant listing products—interest assessment—trigger enhancement" chain, an alternative initiation point is provided besides passively waiting for users to actively enhance the images. This allows the platform backend to automatically complete "clarity verification" during the product listing process. When the buyer interest score given by the interest assessment model is lower than the preset exposure performance condition, the listed product image is identified as a target to be processed, thus ensuring that any main image, header image, or advertising image that will be presented to buyers meets the high-definition, sellable threshold before going live. For merchants, this mechanism eliminates the cost of repeated trial and error, manual selection, or reshooting. For the platform, a single investment of computing power results in a uniform improvement in the quality of product display across the entire site, reducing subsequent bounces due to blurry images and preventing low-quality images from wasting traffic in display positions. More importantly, it quantifies the originally subjective aesthetic judgment of "whether it can attract buyers" into a reproducible and auditable scoring system, giving platform governance an interpretable technical tool.

[0070] In a further embodiment, before step S1100, which responds to the image enhancement event triggered by the user application's target e-commerce image, the following steps are included:

[0071] Step S3100: Respond to the user's real product review event and obtain the product image corresponding to the event;

[0072] When a user's current product review on an e-commerce platform is determined to be a genuine review, a genuine review event is triggered. The server responds to this event and initiates subsequent processes. A genuine product review is characterized by the user completing a transaction on their device, then using the product review portal provided by the e-commerce platform to complete the review form, including editing the review text and uploading product images. The review is then confirmed to match the purchased product through image-text matching and product image comparison. Therefore, the product review is deemed valid and credible.

[0073] The server can obtain the product image from the user's submitted review form.

[0074] Step S3110: When the image resolution of the evaluated product image is lower than a preset threshold, the evaluated product image is confirmed as a target e-commerce image to trigger a corresponding image beautification event.

[0075] After obtaining the product images for review, the server first reads the width and height fields from their EXIF ​​information, multiplies these two values ​​to get the total number of pixels, and then calculates the square root of the total number of pixels. The result is the image resolution. If the EXIF ​​information is missing, the server directly parses the width and height data from the image file header; if it still cannot be read, it obtains the resolution by decoding the image pixel matrix and counting the number of rows and columns. There are two methods for setting the threshold: the first is a fixed value method, for example, setting the threshold to 1280×720 corresponds to a total number of pixels square root of approximately 985; the second is a dynamic scaling method, using the physical resolution of the mainstream terminal screen of the e-commerce platform as a benchmark, and taking 0.75 times that resolution as the threshold. For example, when the mainstream screen is 1920×1080, setting the threshold to 1440×810 corresponds to a total number of pixels square root of approximately 1354. When the resolution of a product image is lower than the set threshold, it means that the image will appear jagged, blurry, or distorted when magnified on a high-definition screen. Buyers will find it difficult to clearly identify the product features through the image, directly affecting the efficiency of their purchase decision. Therefore, the system automatically identifies it as a target e-commerce image that needs to be beautified, triggering subsequent image beautification events. If the resolution is not lower than the threshold, it means that the image already has sufficient clarity and can be used directly for product review display without additional processing, avoiding waste of computing power.

[0076] In this embodiment, the trigger point is shifted to the scenario of "genuine buyer reviews." Only after confirming that the user has indeed received the goods and that the uploaded review images match the product description text and official listing images within a reasonable range of "neither stolen nor irrelevant," is the low-resolution real-shot image beautification process initiated. This avoids unnecessarily processing merchant materials or stolen images as buyer photos and prevents irrelevant images from consuming computing power. Furthermore, since images in review scenarios are often taken under rudimentary conditions and have insufficient resolution, targeted over-resolution and edge smoothing significantly improve the efficiency of information retrieval for subsequent consumers browsing reviews, reducing hesitation or refunds due to "not being able to see details clearly." Thus, the platform, in the high-conversion arena of product reviews, maintains authentic product image content while using technical means to compensate for imaging defects caused by limitations in shooting equipment.

[0077] In a further embodiment, before step S3100, which responds to a user's genuine product review event, the following steps are included:

[0078] Step S4100: Respond to the user's product review event, obtain the product image corresponding to the event, as well as the product description text and the product image of the reviewed product;

[0079] It's understandable that users only have the authority to rate a product after completing a transaction on their device, i.e., when the corresponding order status is "received" or "completed." Subsequently, users can tap the "Rate" control on their device's order page to access the e-commerce platform's product rating portal, redirecting them to the rating form or displaying a pop-up window. To facilitate editing the rating form, editable text and image upload controls are typically provided. Users can edit text reviews and select locally stored photos, or take photos on-site and tap the "Submit" or "Upload" control to upload the photos to the rating form. Once the rating form is complete, users can tap the "Submit" control to submit it to the e-commerce platform's server for review. After approval, it is published in the buyer review section for that product. It's easy to understand that the rating form usually carries a product identifier indicating the user's rating of the purchased product. This product identifier uniquely identifies the corresponding product to distinguish it from other products; for example, a product ID, which can be configured as needed by those skilled in the art.

[0080] When a review form is submitted, a user review event is triggered. Upon receiving the review form, the server responds to this event, retrieving the product image and identifier from the form. Based on the identifier, the server identifies the product as the one being reviewed and retrieves its description and image from the product database. E-commerce platforms typically maintain a product database to store the descriptions and images of all products listed in their online stores.

[0081] Step S4110: Determine the matching degree between the product image content in the product image and the product being evaluated, based on the product description text and the product image uploaded to the product store.

[0082] In one embodiment, a pre-trained, converged image-text matching model is used. The product description text and product image are taken as input. The text encoding sub-network in the model extracts deep semantic information representing the product from the product description text and encodes this deep semantic information into a text vector. Similarly, the image encoding sub-network in the model extracts semantic information representing the product from the product image and encodes this deep semantic information into an image vector. It is easy to understand that both text encoding and image encoding map the corresponding input information to the same semantic space. Furthermore, a vector distance algorithm is used to calculate the vector distance between the text vector and the image vector as the image-text matching degree.

[0083] A pre-trained, converged dual-tower image matching model is employed. The first image encoding sub-network extracts deep semantic information representing the product from the product evaluation image and encodes this deep semantic information into a first image vector. The second image encoding sub-network extracts deep semantic information representing the product from the product listing image and encodes this deep semantic information into a second image vector. It can be understood that the first and second image encoding sub-networks have the same network architecture and share weights. Furthermore, a vector distance algorithm is used to calculate the vector distance between the first and second image vectors as the image matching degree.

[0084] The text-image matching score and the image matching score are multiplied by their respective weights and then summed to obtain the matching score between the product image content in the product evaluation image and the product being evaluated. Each weight can be configured as needed by those skilled in the art; for example, the weight of the text-image matching score can be 0.4, and the weight of the image matching score can be 0.6.

[0085] Vector distance algorithms include, but are not limited to, any of the following: cosine similarity algorithm, vector dot product algorithm, Manhattan distance, Euclidean distance algorithm, Pearson correlation coefficient, etc., which can be selected and implemented as needed by those skilled in the art. For the image-text matching model, the Cl ip model is recommended; however, other equivalent models may also be used. In the dual-tower image matching model, the first and second image coding sub-networks have the same network architecture and share weights; the ResNet model is recommended for network selection, but other equivalent models may also be used. Those skilled in the art can flexibly adapt the training process for each model based on the relevant disclosures regarding the application of the image-text matching model and the dual-tower image matching model in this embodiment.

[0086] Step S4120: When the matching degree belongs to the preset value range, confirm that the user has genuinely rated the product, so as to trigger the user's genuine product rating event.

[0087] The preset value range defines the matching degree range of the product image content displayed in the product review image. If the matching degree is less than the lower limit of the preset value range, it means that the matching degree is too small, indicating that the product image content of the product being reviewed differs too much from the product image on the platform, and the text description is almost unrelated to the product image content. It can be determined that the user did not take a picture of the product they actually purchased, thus confirming that the user is not genuinely reviewing the product. If the matching degree exceeds the upper limit of the preset threshold, it means that the matching degree is too large, indicating that the product image content of the product being reviewed is almost identical to the product image on the platform, and the product description text is almost entirely describing the product image content. It can be determined that the user directly used or cropped the product image on the platform as the product review image, rather than actually taking a picture of the product they actually purchased, thus confirming that the user is not genuinely reviewing the product.

[0088] When the matching degree falls within the preset range (i.e., greater than the lower limit and less than the upper limit), it can be determined that the user has actually photographed and purchased the product. This confirms that the user has genuinely reviewed the product and triggers the "user genuine product review" event. Those skilled in the art can further set the preset range as needed according to the description herein, for example, the preset range is [0.5, 0.9].

[0089] In this embodiment, the prerequisite of "confirming genuine reviews" is broken down into a double-insurance verification of "text-image matching + image matching," thus implementing insurance at the review entry point. The text-image matching model and the dual-tower image matching model map text semantics and image semantics to the same vector space, respectively. Only when text-real-shot image-official... Figure Three Only when the vector distance between samples is neither too far apart nor too overlapping will they proceed to the next stage of beautification. This design, on the one hand, intercepts malicious order boosting, image theft, and irrelevant posts before consuming computing power, and on the other hand, ensures that genuinely valuable real-life reviews are reproduced in high definition, thereby maintaining the credibility of the review system. The computing power saved by the platform can then be precisely allocated to genuine reviews that are most persuasive in influencing purchasing decisions, ultimately creating a positive cycle among merchants, buyers, and the platform: merchants gain credible reputations, buyers obtain clear references, and the platform improves its overall operational level.

[0090] In a further embodiment, step S1200, using a preset image super-resolution model to improve the image resolution of the buyer interest image to obtain a corresponding super-resolution buyer interest image, includes the following steps:

[0091] Step S1210: Extract the shallow feature map corresponding to the buyer's interest image from the shallow representation sub-network in the image super-resolution model;

[0092] The shallow representation subnetwork is used to extract initial features. Its structure consists of a unit combination layer consisting of a convolutional layer followed by a ReLU activation layer. When the buyer's interest image is input into this subnetwork, local features are calculated by sliding the convolutional kernels in the convolutional layers across the image to extract basic visual elements (such as edges, corners, and basic textures). Each convolutional operation outputs one channel of the feature map. The ReLU function (defined as max(0,x)) performs a non-linear transformation on the convolution result, enhancing the model's expressive power.

[0093] Step S1220: Extract the first salient feature map and the second salient feature map corresponding to the shallow feature map from the two deep explicit subnetworks that follow the shallow representation subnetwork;

[0094] Two deep explicit subnetworks are used to extract different salient features respectively. The two subnetworks are structurally identical but do not share network parameters. The structure is represented as a deep extraction combination layer followed by a salient extraction combination layer. The deep extraction combination layer consists of N sequentially concatenated unit combination layers, each of which is a convolutional layer followed by a ReLU activation layer. The salient extraction combination layer consists of a convolutional layer, a Consin convolutional layer, and a ReLU activation layer connected sequentially. N can be configured as needed by those skilled in the art, for example, 7.

[0095] It is understandable that, due to the lack of parameter sharing, the two deep explicit sub-networks can extract different salient features. This is manifested in the N-layer stacked convolutional activation network, which can progressively deepen the extraction of higher-level image semantic features. For example, shallow layers (such as layers 1-2) can extract low-order geometric features such as relative edges and corners; middle layers (such as layers 3-5) can extract textures (such as wrinkles and material stripes on the surface of a product); and deep layers (such as layers 6-N) further extract higher-order semantic features (such as the discriminative representation of "product outline" and "background interference"). Further salient extraction layers, with their convolution followed by Consin convolution operations, can deeply mine feature similarity. Then, ReLU activation further introduces non-linear sparsity, enhancing high-frequency details (such as texture) in salient regions while filtering low-frequency information (such as a uniform background) in non-salient regions.

[0096] Step S1230: Extract the first deep feature map and the second deep feature map corresponding to the shallow feature map from the two deep representation subnetworks that follow the shallow representation subnetwork.

[0097] Two deep explicit subnetworks are used to extract different deep high-level features respectively. The two subnetworks are structurally identical but do not share network parameters. The structure is represented as a deep extraction combination layer. The deep extraction combination layer consists of N unit combination layers spliced ​​sequentially. Each unit combination layer consists of a convolutional layer followed by a ReLU activation layer. N can be configured as needed by those skilled in the art, for example, 7.

[0098] It is understandable that deep explicit subnetworks can extract deep, high-level features, which is manifested in the fact that N-layer stacked convolutional activation networks can progressively deepen the extraction of more advanced image semantic features.

[0099] Step S1240: After fusing the first salient feature map, the second salient feature map, the first deep feature map, and the second deep feature map, pixel recombination processing is performed to obtain a super-resolution buyer interest image.

[0100] First, the first and second salient feature maps are concatenated and input into the unit combination layer to output a first fused feature map. Then, the resulting first fused feature map is concatenated with the first deep feature map and input into the unit combination layer to output a second fused feature map. Next, the second fused feature map is concatenated with the second deep feature map to obtain a third fused feature map. Finally, the third fused feature map is input into a pixel reassembly layer for pixel reassembly processing to obtain a super-resolution buyer interest image. The pixel reassembly layer structure is represented as a pixel shuffle layer or a convolutional layer concatenated before it.

[0101] It's understandable that concatenating feature maps followed by convolutional activation allows for more refined feature map fusion and enables the learning of optimal weights for refined feature points, thus improving the accuracy of the super-resolution image. Further pixel recombination rearranges the channel dimensions of the low-resolution feature maps into the spatial dimensions of the high-resolution feature maps, outputting the final super-resolution image.

[0102] This embodiment presents an engineering paradigm that balances computational efficiency and visual quality by implementing a model super-resolution process of "shallow representation - double saliency - double deep layer - fusion and recombination". The shallow network uses minimal convolutional ReLU units to quickly capture edge textures, the saliency sub-network uses a network structure that deepens feature extraction and saliency feature extraction to enhance high-frequency details, and the deep sub-network uses multi-layer convolutional ReLU to progressively abstract semantic contours. Finally, the pixel-shuffle layer "folds" the channel dimension into space to improve image resolution. The intuitive logic of "coarse to fine, local to global" allows for the accurate and natural restoration of low-resolution buyer interest images; simultaneously, it ensures that the server maintains high efficiency and quality even when batch processing millions of product images.

[0103] In a further embodiment, after step S1400, which involves pushing the processed and beautified e-commerce image to the user, the following steps are included:

[0104] Step S1500: Respond to the thumbnail product image preparation event and obtain the thumbnail image specifications corresponding to the event;

[0105] E-commerce platforms can also provide merchants with the ability to generate thumbnail product images based on beautified e-commerce images. Users only need to select the application scenario for the thumbnail product image when enabling the function, such as displaying the image in a product list, search results page, or product recommendation results, to trigger the thumbnail product image generation event. The function can be presented in various forms, usually in the form of controls or pop-ups in a visual operation interface for user operation. Those skilled in the art can flexibly adapt and implement it.

[0106] The server responds to the thumbnail product image preparation event. By parsing the parameters returned by the above-mentioned functional interface, the application scenario of the thumbnail product image selected by the user can be obtained. Subsequently, the pre-configured thumbnail image specifications required to display the thumbnail product image in this application scenario can be determined.

[0107] Step S1510: According to the thumbnail image specifications, the target image content of the super-score buyer interest image in the beautified e-commerce image is thumbnailed to obtain a thumbnail product image that contains the corresponding complete target image content.

[0108] It is understandable that the server can save the pixel location information of the image content of the super-target buyer interest image overlaid on the beautified e-commerce image during the process of beautifying the target e-commerce image, so as to be retrieved during the aforementioned event response. Therefore, by calling the pre-stored pixel location information, the image content of the super-target buyer interest image can be located from the beautified e-commerce image as the target image content.

[0109] When the width and height values ​​of the thumbnail image specification are larger than the actual size of the target image content, a rectangular area containing the complete content of interest can be cropped from the beautified e-commerce image based on the geometric center of the target image content and the aspect ratio of the thumbnail image specification. In this case, the cropped thumbnail product image will completely retain the target image content. When the minimum value of the width and height values ​​of the thumbnail image specification is smaller than the target image content, the center of the target image content can be used as the center of the desired thumbnail product image. The scaling ratio between the minimum value (which may be the minimum width or the minimum height) and the corresponding width or height of the target image content is calculated. Bilinear interpolation or Lanzos resampling algorithm is used to scale the target image content proportionally to ensure that the scaled thumbnail product image meets the specification requirements, completely presents the target image content, and does not suffer from stretching, distortion, or content loss.

[0110] In this embodiment, the enhanced high-definition images are extended downstream to practical application scenarios. After pushing enhanced e-commerce images, the platform can fulfill merchants' requests for thumbnail sizes and precisely crop or scale the image content of images that capture buyer interests within the enhanced e-commerce images. This ensures that even in thumbnail displays with lower image sizes, such as search lists and "You May Also Like," the image content elements that buyers are interested in are fully preserved. This eliminates the need for merchants to prepare multiple sets of image materials for different booths and avoids issues like "critical information being cropped" or "stretching and distortion" caused by blind compression, which negatively impact the buyer's viewing experience. The intuitive effect of "high-definition large images and clear small images" demonstrates how to achieve end-to-end visual consistency with minimal storage and bandwidth costs, potentially enabling efficient conversion from "clear viewing—clicking—purchase" when users browse and / or search for products.

[0111] In a further embodiment, before step S1200, which involves using a preset image super-resolution model to improve the image resolution of the buyer interest image and obtain the corresponding super-resolution buyer interest image, the following steps are included:

[0112] Step S2200: Obtain a prepared training set, which includes training samples and their supervision labels. The training samples are low-resolution e-commerce images, and the supervision labels are high-resolution e-commerce images corresponding to the training samples.

[0113] When preparing the training set, data was first collected from the entire image library of e-commerce platforms according to high-resolution standards. This covered all visible online scenarios, including product images on product detail pages, product header images, screenshots from live streams and short videos, buyer reviews, customer photos, advertising materials, brand posters, and event pages. This ensured coverage of all primary categories such as apparel, 3C products, beauty products, home furnishings, baby products, and fresh produce, while also incorporating various shooting methods such as studio photography, street photography, handheld shooting, flat lay photography, model shots, and scene combinations. This ensured statistical significance in terms of color distribution, lighting conditions, background complexity, subject size, and occlusion relationships. After collection, the original high-resolution images underwent compliance preprocessing such as deduplication and watermark removal to obtain a clean image pool. Subsequently, a low-resolution version was generated for each clean image in the clean pool. These low-resolution e-commerce images were used as training samples, and the corresponding original high-resolution e-commerce images were used as supervision labels. All training samples and their supervision labels were then compiled into a training set.

[0114] High-resolution standards can be expressed as follows: a pre-trained, converged interest evaluation model assesses whether the buyer's interest score for the input e-commerce image meets the exposure effectiveness criteria; or the image resolution of the e-commerce image meets high-resolution numerical requirements, etc. Those skilled in the art can configure these standards as needed. Furthermore, they can flexibly adapt these methods to reduce the image resolution, obtaining a lower-resolution version of the corresponding image.

[0115] Step S2210: Train the image super-resolution model using the training set until it converges, thereby learning the ability to improve the resolution of input e-commerce images.

[0116] The image super-resolution model uses a single training sample and its supervision label from the training set to upscale the training sample, outputting a corresponding predicted super-resolution image. Then, an objective function is used to calculate the target value between the predicted super-resolution image and the supervision label. The training objective is to increase the target value as close as possible to the maximum value of the objective function. Therefore, it can be determined whether the target value in this round exceeds a preset threshold based on the maximum value of the objective function. If it does, the network parameters of each layer in the image super-resolution model are adjusted accordingly based on the target value in this round. Iterative training is then performed using other training samples and their supervision labels until the target value in the corresponding round exceeds the preset threshold, at which point the iteration ends, confirming that the image super-resolution model has been trained to a convergent state. The preset threshold can be configured as needed by those skilled in the art based on the information disclosed herein; for example, if the objective function is SSIM, the preset threshold can be configured to 0.9.

[0117] The objective function can be PSNR (Peak Signal-to-Noise Ratio), where a higher PSNR value indicates better image quality; or SSIM (Structural Similarity), which is an index that measures the similarity between two images. Its value ranges from [0,1]. A larger SSIM value indicates less image distortion and better image quality.

[0118] This embodiment reveals the training process of the interest assessment model, ensuring the accuracy, reliability, and interpretability of its reasoning ability, thus laying a solid foundation for practical applications.

[0119] Please see Figure 3 This application provides an e-commerce image enhancement device, which is a functional embodiment of the e-commerce image enhancement method of this application. On another note, this e-commerce image enhancement device, also provided to fulfill one of the purposes of this application, includes an event response module 1100, an image super-resolution module 1200, a smoothing determination module 1300, and an enhancement push module 1400. The event response module 1100 is used to respond to an image enhancement event triggered by a user's application of a target e-commerce image, and uses a preset interest segmentation model to segment the target e-commerce image into corresponding buyer interest images and non-buyer interest images. The image super-resolution module 1200 is used to use a preset image super-resolution model to increase the image resolution of the buyer interest image, obtaining a corresponding super-resolution buyer interest image. The smoothing determination module 1300 is used to determine the smoothness of the stitching edges in the e-commerce image to be verified obtained by stitching the super-resolution buyer interest image and the non-buyer interest image. The enhancement push module 1400 is used to perform smoothing transition processing on the stitching edges in the e-commerce image to be verified when the smoothness is insufficient, and pushes the processed enhanced e-commerce image to the user.

[0120] In a further embodiment, before the event response module 1100, there are: a first event response submodule, used to respond to a user's product listing event and obtain the product image corresponding to the event; an interest evaluation submodule, used to determine the buyer interest rating corresponding to the product image using a preset interest evaluation model; and a first event triggering submodule, used to confirm that the product image is a target e-commerce image when the buyer interest rating does not meet the exposure effectiveness conditions, so as to trigger the corresponding image beautification event.

[0121] In a further embodiment, before the event response module 1100, there are: a second event response submodule, used to respond to a user's real product review event and obtain the product review image corresponding to the event; and a second event triggering submodule, used to confirm that the product review image is a target e-commerce image when the image resolution of the product review image is lower than a preset threshold, so as to trigger a corresponding image beautification event.

[0122] In a further embodiment, before the second event response submodule, there is a third event response submodule, used to respond to a user review product event, obtain the review product image corresponding to the event, as well as the product description text and the listed product image of the review product; a product matching submodule, used to determine the matching degree between the product image content in the review product image and the review product based on the product description text and the listed product image of the review product; and a third event triggering submodule, used to confirm that the user has actually reviewed the product when the matching degree belongs to a preset value range, so as to trigger a user's actual product review event.

[0123] In a further embodiment, the image super-resolution module 1200 includes: a first feature extraction submodule, used to extract a shallow feature map corresponding to the buyer interest image from a shallow representation subnetwork in the image super-resolution model; a second feature extraction submodule, used to extract a first salient feature map and a second salient feature map corresponding to the shallow feature map from two deep explicitation subnetworks connected to the shallow representation subnetwork; a third feature extraction submodule, used to extract a first deep feature map and a second deep feature map corresponding to the shallow feature map from two deep representation subnetworks connected to the shallow representation subnetwork; and a super-resolution imaging submodule, used to fuse the first salient feature map, the second salient feature map, the first deep feature map, and the second deep feature map and then perform pixel recombination processing to obtain a super-resolution buyer interest image.

[0124] In a further embodiment, after the beautification push module 1400, there are: a fourth event response submodule, used to respond to a thumbnail product image preparation event and obtain the thumbnail image specifications corresponding to the event; and a thumbnail processing submodule, used to perform thumbnail processing on the target image content of the super-score buyer interest image in the beautified e-commerce image according to the thumbnail image specifications to obtain a thumbnail product image, so that it contains the corresponding complete target image content.

[0125] In a further embodiment, before the image super-resolution module 1200, there are: a training set acquisition submodule, used to acquire a prepared training set, the training set including training samples and their supervision labels, the training samples being low-resolution e-commerce images, and the supervision labels being high-resolution e-commerce images corresponding to the training samples; and a model training submodule, used to train the image super-resolution model to a convergent state using the training set, thereby learning the ability to improve the resolution of the input e-commerce image.

[0126] To address the aforementioned technical problems, embodiments of this application also provide computer equipment. For example... Figure 4 The diagram shows the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store a sequence of control information. When the computer-readable instructions are executed by the processor, they enable the processor to implement an e-commerce image enhancement method. The processor of the computer device provides computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may store computer-readable instructions, which, when executed by the processor, enable the processor to execute the e-commerce image enhancement method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0127] In this embodiment, the processor is used to execute... Figure 3 The system contains the specific functions of each module and its sub-modules. The memory stores the program code and various data required to execute these modules or sub-modules. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules / sub-modules in the e-commerce image enhancement device of this application. The server can call the server's program code and data to execute the functions of all sub-modules.

[0128] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the e-commerce image enhancement method of any embodiment of this application.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0130] In summary, this application can enhance low-resolution e-commerce images, making the image content that buyers are interested in stand out while maintaining overall clarity and naturalness.

[0131] Those skilled in the art will understand that the steps, measures, and solutions in the various operations, methods, and processes discussed in this application can be alternated, modified, combined, or deleted. Furthermore, other steps, measures, and solutions in the various operations, methods, and processes discussed in this application can also be alternated, modified, rearranged, decomposed, combined, or deleted. Furthermore, steps, measures, and solutions in the prior art that are similar to those in the open-source operations, methods, and processes of this application can also be alternated, modified, rearranged, decomposed, combined, or deleted.

[0132] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for enhancing e-commerce images, characterized in that, Includes the following steps: In response to the image enhancement event triggered by the user's application of the target e-commerce image, a preset interest segmentation model is used to segment the target e-commerce image into buyer interest images and non-buyer interest images. A preset image super-resolution model is used to enhance the image resolution of the buyer's interest image, thereby obtaining a corresponding super-resolution buyer interest image; Determine the smoothness of the stitching edges in the e-commerce image to be verified, obtained by stitching together the super-resolution buyer interest image and the non-buyer interest image. When the smoothness does not meet the standard, the splicing edges in the e-commerce image to be verified are smoothed, and the beautified e-commerce image obtained after processing is pushed to the user.

2. The e-commerce image enhancement method according to claim 1, characterized in that, Before responding to an image enhancement event triggered by a user's target e-commerce image, the following steps are included: In response to a user's product listing event, retrieve the corresponding product image. A preset interest assessment model is used to determine the buyer interest rating corresponding to the images of the listed products. When the buyer's interest rating does not meet the exposure effectiveness conditions, the product image is confirmed as the target e-commerce image to trigger the corresponding image enhancement event.

3. The e-commerce image enhancement method according to claim 1, characterized in that, Before responding to an image enhancement event triggered by a user's target e-commerce image, the following steps are included: Respond to user reviews of products and obtain the corresponding product images. When the image resolution of the product being evaluated is lower than a preset threshold, the product being evaluated is confirmed as a target e-commerce image, thereby triggering a corresponding image enhancement event.

4. The e-commerce image enhancement method according to claim 3, characterized in that, Before responding to genuine user reviews of products, the following steps are included: In response to user review events, retrieve the corresponding product image, product description text, and product image of the reviewed product. Based on the product description text and the product images of the reviewed products, determine the degree of matching between the product image content in the product images of the reviewed products and the reviewed products; When the matching degree falls within a preset range, the user's genuine product review is confirmed, thereby triggering a genuine product review event.

5. The e-commerce image enhancement method according to claim 1, characterized in that, The image resolution of the buyer interest image is improved using a preset image super-resolution model to obtain a corresponding super-resolution buyer interest image, including the following steps: The shallow feature map corresponding to the buyer's interest image is extracted from the shallow representation subnetwork in the image super-resolution model; The first salient feature map and the second salient feature map corresponding to the shallow feature map are extracted from the two deep explicit subnetworks that follow the shallow representation subnetwork. The first deep feature map and the second deep feature map corresponding to the shallow feature map are extracted from the two deep representation subnetworks that follow the shallow representation subnetwork. After fusing the first salient feature map, the second salient feature map, the first deep feature map, and the second deep feature map, pixel recombination processing is performed to obtain a super-resolution buyer interest image.

6. The e-commerce image enhancement method according to claim 1, characterized in that, After the processed and beautified e-commerce image is pushed to the user, the following steps are included: In response to the thumbnail product image creation event, obtain the thumbnail image specifications corresponding to the event; According to the thumbnail image specifications, the target image content of the super-score buyer interest image in the beautified e-commerce image is thumbnailed to obtain a thumbnail product image that contains the corresponding complete target image content.

7. The e-commerce image enhancement method according to claim 1, characterized in that, Before using a preset image super-resolution model to enhance the image resolution of the buyer's interest image and obtain the corresponding super-resolution buyer's interest image, the following steps are included: Obtain a prepared training set, which includes training samples and their supervision labels. The training samples are low-resolution e-commerce images, and the supervision labels are high-resolution e-commerce images corresponding to the training samples. The image super-resolution model is trained to convergence using the training set, thereby learning the ability to improve the resolution of input e-commerce images.

8. An e-commerce image enhancement device, characterized in that, include: The event response module is used to respond to image enhancement events triggered by the user application's target e-commerce image, and uses a preset interest segmentation model to segment the target e-commerce image into buyer interest images and non-buyer interest images. The image super-resolution module is used to improve the image resolution of the buyer interest image using a preset image super-resolution model, so as to obtain the corresponding super-resolution buyer interest image. The smoothness determination module is used to determine the smoothness of the stitching edges in the e-commerce image to be verified, which is obtained by stitching the super-resolution buyer interest image and the non-buyer interest image. The beautification push module is used to smooth the splicing edges in the e-commerce image to be verified when the smoothness does not meet the standard, and push the beautified e-commerce image obtained after processing to the user.

9. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images