E-commerce Image Dynamic Optimization and Adjustment Method Based on Market Feedback

Abstract

The present invention relates to the technical field of image processing, and specifically discloses an e-commerce image dynamic optimization and adjustment method based on market feedback. Color noise is removed and interference in edge detection is suppressed through gray conversion and smoothing filtering techniques. Canny edge detection and wavelet transform are used to extract and analyze edge curvature features, accurately identify serrated regions. Shape context descriptors and the Hungarian algorithm are used to align the boundaries of pictures of the same commodity from different perspectives, and the shape difference coefficient is calculated to evaluate its consistency. For further optimization, the shape difference coefficient and edge curvature features are transformed into a comprehensive feature vector, which is used as the input of the graph convolutional network model. Global features are extracted through multi-layer graph convolutional operations to identify regions with severe serration, and hard thresholding is used for denoising, and cubic spline interpolation is used to accurately repair the serrated regions.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and specifically relates to an e-commerce image dynamic optimization and adjustment method based on market feedback. Background Art

[0002] With the popularization of online shopping, consumers increasingly rely on high-quality product pictures to make purchase decisions. However, due to reasons such as shooting conditions, equipment differences, and improper post-processing, many product pictures have problems such as jaggedness, blurriness, or inconsistency, which seriously affect the shopping experience of consumers and their trust in products.

[0003] The prior art has the following deficiencies:

[0004] In e-commerce images with market feedback, due to the complex shapes, irregular edges, and diversity of multi-view displays in product pictures, problems such as jaggedness, discontinuity, or positioning deviation often occur during edge detection. These problems will seriously affect the subsequent image optimization effects, such as the accuracy of background replacement and copy layout, and thus reduce the attractiveness of product pictures and the user conversion rate. Summary of the Invention

[0005] The purpose of the present invention is to provide an e-commerce image dynamic optimization and adjustment method based on market feedback to solve the problems in the above background.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An e-commerce image dynamic optimization and adjustment method based on market feedback includes the following steps:

[0008] S1: Perform image preprocessing on the product picture and convert it into a grayscale image to remove color noise that interferes with edge detection;

[0009] S2: Extract the edge curvature features in the product picture, calculate the edge curvature change, calibrate the jagged area according to the threshold of the edge curvature change, and identify the specific position of the boundary jaggedness;

[0010] S3: Based on the jagged area, align the boundaries of pictures of the same product from multiple perspectives, calculate the difference in boundary shapes between different perspectives, and evaluate the consistency of the product images according to the degree of difference;

[0011] S4: Convert the shape difference coefficient and edge curvature features of the product image into a comprehensive feature vector as the input item of the machine learning model, and identify the area with severe jaggedness according to the model output result;

[0012] S5: Repair the edge curve for the area with severe jaggedness.

[0013] As a further solution of the present invention: The edge curvature feature is extracted as follows:

[0014] The input product image is converted into a grayscale image, and Canny edge detection is used to obtain a set of all edge points in the image;

[0015] All the edge points are connected into a continuous edge curve, and wavelet transform is performed on the edge curve;

[0016] Through wavelet transform, the curve is converted from the time domain to the frequency domain, so as to obtain the multi-scale features of the edge curve, denoted as the edge curvature feature.

[0017] As a further solution of the present invention: According to the edge curvature feature, the edge curvature is calculated, specifically including:

[0018] According to the multi-scale wavelet coefficients of the curve, the following formula is used to calculate the curvature: ;

[0019] Where represents the change rate of the wavelet coefficient, represents the scale parameter, represents the translation parameter, represents the wavelet transform coefficient in the x-direction, represents the wavelet transform coefficient in the y-direction;

[0020] By calculating the derivative of the wavelet coefficient, the edge curvature of each edge point is obtained.

[0021] As a further solution of the present invention: According to the edge curvature of each edge point, the specific position of the boundary serration is identified, specifically including:

[0022] Obtain the edge curvature of each edge point, compare the edge curvature of each edge point with a preset threshold, and judge whether the edge curvature of each edge point is greater than or equal to the preset threshold. If so, it is marked as a serrated area, and if not, it is marked as a non-serrated area.

[0023] As a further solution of the present invention: The calculation process of the difference in boundary shape is as follows:

[0024] Obtain product images of the same product from multiple perspectives , represents the number of perspectives, obtain the edge curve , represents the th point on the boundary, represents the total number of sampling points;

[0025] For each point in the boundary point set, construct a shape context descriptor , the calculation expression is: ;

[0026] Among them, represents the th sampling point, represents the point and the point the Euclidean distance between, and represents a point in the boundary point set, represents the point and the point polar angle of, represents the discrete partition index, represents the distance interval, represents the angle interval, represents the indicator function, which is 1 when the condition is satisfied and 0 otherwise, represents the point shape context descriptor of;

[0027] For any two perspective images and boundary point sets of and , calculate the shape context descriptors and of each point respectively, construct a cost matrix, and each matrix element represents the shape context difference between the point and the point , and the calculation formula is: ;

[0028] In the formula, represents the number of distance partitions, represents the number of angle partitions, represents a small positive value to avoid a zero denominator, represents each matrix element;

[0029] Use the Hungarian algorithm to solve the optimal matching of the cost matrix, establish a one-to-one correspondence between the boundary point sets and , and according to the point set matching result, calculate the overall shape difference coefficient of the two boundary curves. The calculation expression is: ;

[0030] In the formula, represents the shape difference coefficient.

[0031] As a further solution of the present invention: Evaluating the consistency of commodity images specifically includes:

[0032] Determine whether the difference degree coefficient of the commodity image is greater than or equal to a preset threshold. If so, the images of the corresponding commodity from multiple perspectives are inconsistent. If not, the images of the corresponding commodity from multiple perspectives are consistent.

[0033] As a further solution of the present invention: The process of constructing the comprehensive feature vector is as follows:

[0034] Obtain the shape difference degree coefficient and edge curvature feature of the commodity image, and standardize the shape difference degree coefficient and edge curvature feature; Use a clustering algorithm to cluster the standardized shape difference degree coefficient and edge curvature feature vectors to obtain several clusters, and each cluster corresponds to a category of commodity images; Calculate the center point of each cluster, and the center point of the cluster is used as the comprehensive feature vector of the commodity images within the corresponding cluster, and the comprehensive feature vector represents the common features of the images within the cluster; Assign each commodity image to the corresponding cluster, and use the center point of the corresponding cluster as the comprehensive feature vector of the commodity image;

[0035] The clustering algorithm is the K-means clustering algorithm, and the specific process is as follows: Randomly select M initial clustering centers; Calculate the distance between the feature vector of each image and all the center points, and assign the image to the nearest clustering center; Update the center point of each cluster to the mean value of the feature vectors of all the images within the cluster; Repeat the clustering process until the clustering centers no longer change, and the clustering process ends.

[0036] As a further solution of the present invention: The process of constructing the machine learning model is as follows:

[0037] Replace the shape difference degree coefficient and edge curvature feature of the commodity image with the comprehensive feature vector as the input item of the machine learning model, and the machine learning model is a graph neural network model; Use graph convolution operations to gradually extract local and global features, and finally generate a global graph representation; Through the processing of multiple layers of graph convolution, the graph neural network can identify each group of input comprehensive feature vectors, and adjust the model parameters by minimizing the error between the prediction result and the actual annotation, and train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training, and finally generate a serration degree score for identifying severely serrated areas.

[0038] As a further solution of the present invention: Identifying severely serrated areas specifically includes:

[0039] According to the recognition result of the machine learning model, output a serration degree score for each serrated area;

[0040] Determine whether the serration degree score of each serrated area is greater than or equal to a preset threshold. If so, the corresponding area is a severely serrated area. If not, the corresponding area is a non-severely serrated area.

[0041] As a further solution of the present invention: for severely serrated regions, the edge curve is repaired, specifically including:

[0042] Use the hard threshold method to denoise the wavelet coefficients and remove the high-frequency noise in the image;

[0043] For the denoised wavelet coefficients and perform local curve repair using the cubic spline interpolation method, calculate the curve fitting function for each serrated region, and the calculation expression is: ;

[0044] wherein, represents the arc length parameter of the curve, represents the curve fitting function, represents the spline, represents the total number of splines, represents the th fitting coefficient of the spline, represents the th spline basis function; obtain the edge curvature of the serrated region, optimize the edge curvature, and use the least squares method to adjust the fitting curve to obtain the image after removing the serrations.

[0045] Advantages of the present invention:

[0046] (1) Through a series of advanced image processing technologies, including grayscale conversion, edge detection, wavelet transform, and the application of shape context descriptors, the present invention can accurately identify and repair the serrated regions in commodity pictures, improving the quality and visual effect of a single picture. In addition, by aligning the boundaries of pictures of the same commodity from multiple perspectives and calculating the shape difference coefficient to evaluate its consistency, the high consistency and coordination of multi-perspective display are ensured. This high-quality and consistent commodity display not only enhances the visual appeal of the commodity but also effectively reduces consumers' doubts due to image quality problems, enhancing user trust. More importantly, by optimizing the overall presentation of commodity images, the shopping experience and satisfaction of users can be significantly improved, thereby increasing the sales conversion rate. Specifically, the present invention uses a graph neural network model to extract global and local features from the comprehensive feature vector and combines advanced image repair technologies (such as cubic spline interpolation method) to achieve precise repair of severely serrated regions, thus providing an efficient, automated, and highly practical commodity image optimization solution for e-commerce platforms.

[0047] (2)The present invention proposes a fully automated image optimization method that covers all aspects from preprocessing, feature extraction, consistency evaluation to image repair, and is used for the efficient processing of a large number of commodity images on e-commerce platforms. By adopting advanced gray conversion, edge detection, and wavelet transform technologies for preprocessing, and using shape context descriptors to accurately identify serrated regions, this method ensures high-quality single-image output. Further, a machine learning model based on graph convolutional network (GCN) performs boundary alignment and consistency evaluation on commodity images from multiple perspectives, and can intelligently identify serrated regions that need to be repaired. Combining optimization algorithms (such as the Hungarian algorithm) and automated repair technologies (such as cubic spline interpolation), this method achieves precise repair of serrated regions, significantly reducing the need for manual intervention, and remarkably improving the processing efficiency and the stability of the results. This highly automated process not only saves time and costs, but also enables e-commerce platforms to batch update commodity images in a short time. In addition, by improving the quality and consistency of commodity displays, the present invention effectively enhances the user experience and trust. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below with reference to the accompanying drawings.

[0049] Figure 1 It is a specific step flowchart of the e-commerce image dynamic optimization and adjustment method based on market feedback of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without any creative work shall fall within the protection scope of the present invention.

[0051] Please refer to Figure 1 As shown, the present invention is an e-commerce image dynamic optimization and adjustment method based on market feedback, including the following steps:

[0052] S1: Perform image preprocessing on commodity images, convert them into grayscale images to remove color noise that interferes with edge detection;

[0053] S2: Extract the edge curvature features in commodity images, calculate the edge curvature change, and calibrate the serrated regions according to the threshold of the edge curvature change to identify the specific positions of boundary serrations;

[0054] S3: Based on the serrated regions, perform boundary alignment on images of the same commodity from multiple perspectives, calculate the differences in boundary shapes between different perspectives, and evaluate the consistency of commodity images according to the degree of difference;

[0055] S4: Convert the shape difference coefficient and edge curvature feature of the commodity image into a comprehensive feature vector, which is used as an input item of the machine learning model, and identify the severely serrated area according to the model output result;

[0056] S5: Repair the edge curve for the severely serrated area.

[0057] In S1, perform image preprocessing on the commodity picture and convert it into a grayscale image to remove color noise that interferes with edge detection. Specifically, it includes:

[0058] During the image preprocessing of the commodity picture, first convert the color picture from the RGB color space to a grayscale image. This operation is completed by weighted averaging the red, green, and blue components of each pixel point to generate a single grayscale value, thereby retaining the brightness information of the image and eliminating the interference of color differences on subsequent processing. This conversion not only simplifies data processing but also improves the computational efficiency of the image, providing a cleaner image basis for subsequent edge detection.

[0059] Next, perform smoothing processing on the converted grayscale image to suppress the influence of noise. Use a filtering method to remove small noises in the grayscale image, smooth the edges while avoiding the influence of interference information on edge detection. This process can effectively reduce the interference of random noise and detailed information on subsequent analysis, ensuring that the extracted edge curvature features are more accurate and stable, laying a reliable foundation for the subsequent identification of serrated areas.

[0060] In S2, extract the edge curvature features of the commodity picture, calculate the change in edge curvature, and calibrate the serrated area according to the threshold of the edge curvature change to identify the specific position of the boundary serration. Specifically, it includes:

[0061] The extraction of edge curvature features is as follows:

[0062] Convert the input commodity picture into a grayscale image and use Canny edge detection to obtain a set of all edge points in the image;

[0063] Connect all the edge points into a continuous edge curve and perform wavelet transform on the edge curve;

[0064] Through wavelet transform, convert the curve from the time domain to the frequency domain, thereby obtaining the multi-scale features of the edge curve, denoted as edge curvature features.

[0065] According to the edge curvature features, calculate the edge curvature, specifically including:

[0066] According to the multi-scale wavelet coefficients of the curve, use the following formula to calculate the curvature: ;

[0067] Among them, represents the change rate of wavelet coefficients, represents the scale parameter, represents the translation parameter, represents the wavelet transform coefficient in the x - direction, represents the wavelet transform coefficient in the y - direction;

[0068] By calculating the derivative of the wavelet coefficients, the edge curvature of each edge point is obtained;

[0069] According to the edge curvature of each edge point, the specific position of boundary serration is identified, specifically including:

[0070] Obtain the edge curvature of each edge point, compare the edge curvature of each edge point with a preset threshold, and determine whether the edge curvature of each edge point is greater than or equal to the preset threshold. If so, it is marked as a serrated area; if not, it is marked as a non - serrated area.

[0071] It should be noted that: by adopting wavelet transform technology to perform multi - scale analysis on the edge curves in commodity pictures, the edge curvature features can be accurately extracted, and the position of the serrated area can be quickly calibrated in combination with the edge curvature threshold; wavelet transform is the core step of edge curvature feature extraction, including the type of wavelet function adopted (such as Haar wavelet) and parameter settings (such as scale parameter and translation parameter).

[0072] In S3, based on the serrated area, the pictures of the same commodity from multiple perspectives are aligned at the boundary, the difference in boundary shapes between different perspectives is calculated, and the consistency of the commodity images is evaluated according to the degree of difference, specifically including:

[0073] According to the degree of difference in boundary shapes between different perspectives, a shape difference coefficient is calculated to evaluate the consistency of the commodity images; the calculation process of the shape difference coefficient is as follows:

[0074] Obtain the commodity pictures of the same commodity from multiple perspectives , represents the number of perspectives, obtain the edge curve , represents the th point on the boundary, represents the total number of sampling points;

[0075] For each point in the boundary point set, a shape context descriptor is constructed, and the calculation expression is: ;

[0076] Among them, represents the th sampling point, Represents a point and the point The Euclidean distance between them, and represents a point in the boundary point set, Represents the point and the point The polar angle of, Represents the discrete partition index, Represents the distance interval, Represents the angle interval, Represents the indicator function, which is 1 when the condition is satisfied and 0 otherwise, Represents the point The shape context descriptor of;

[0077] For any two perspective images and The boundary point sets of and respectively calculate the shape context descriptors of each point and to construct a cost matrix, and each matrix element represents the shape context difference between the point and the point The calculation formula is: ;

[0078] In the formula, represents the number of distance partitions, represents the number of angle partitions, represents a small positive value to avoid a zero denominator, represents each matrix element;

[0079] Use the Hungarian algorithm to solve the optimal matching of the cost matrix and establish a one-to-one correspondence between the boundary point sets and According to the point set matching result, calculate the overall shape difference coefficient of the two boundary curves. The calculation expression is: ;

[0080] In the formula, represents the shape difference coefficient;

[0081] Judge whether the difference coefficient of the commodity image is greater than or equal to the preset threshold. If so, the images of the corresponding commodity in multiple perspectives are inconsistent. If not, the images of the corresponding commodity in multiple perspectives are consistent;

[0082] It should be noted that: by calculating the shape difference coefficient to evaluate the consistency of pictures of the same commodity from different perspectives, and using the shape context descriptor and the Hungarian algorithm to achieve the optimal matching of the boundary point sets, the inconsistency between images can be effectively identified. It can not only quantify the similarity degree of the commodity image boundaries from different perspectives, but also effectively identify the inconsistency caused by the jagged area, thus ensuring the accurate evaluation of the consistency of the commodity images.

[0083] In S4, the shape difference coefficient and the edge curvature feature of the commodity image are replaced with a comprehensive feature vector as the input item of the machine learning model, and the areas with serious jagging are identified according to the model output result, which specifically includes:

[0084] Obtain the shape difference coefficient and the edge curvature feature of the commodity image, and standardize the shape difference coefficient and the edge curvature feature; use the clustering algorithm to cluster the standardized shape difference coefficient and the edge curvature feature vectors to obtain several clusters, and each cluster corresponds to a type of commodity image; calculate the center point of each cluster, and the center point of the cluster is used as the comprehensive feature vector of the commodity images within the corresponding cluster, and the comprehensive feature vector represents the common features of the images within the cluster; assign each commodity image to the corresponding cluster, and use the center point of the corresponding cluster as the comprehensive feature vector of the commodity image;

[0085] The clustering algorithm is the K-means clustering algorithm, and the specific process is: randomly select M initial clustering centers; calculate the distances between the feature vectors of each image and all the center points, and assign the image to the clustering center with the closest distance; update the center point of each cluster to the mean value of the feature vectors of all the images within the cluster; repeat the clustering process until the clustering center no longer changes, and the clustering process ends;

[0086] Extract the shape difference coefficient and the edge curvature feature from the commodity images from different perspectives, and combine these features into a comprehensive feature vector. The shape difference coefficient reflects the inconsistency of the boundary shapes from different perspectives, while the edge curvature feature captures the bending degree and change trend of the boundary line.

[0087] Regard each commodity view as a node in the graph, and the comprehensive feature vector as the node attribute. If there is a significant shape or position relationship between two views, an edge is established between these two nodes, and the weight of the edge can be defined according to their similarity or distance.

[0088] Use the graph convolutional network (GCN) to process the graph structure. Through multi-layer graph convolutional operations, the model can gradually extract local and global feature information. Each layer of graph convolution enhances the understanding of the node about the feature information of its neighboring nodes, so as to better capture complex patterns and associations.

[0089] After multiple layers of graph convolution, a global graph representation is finally generated, which integrates the information of all nodes and their interrelationships. This global graph representation can effectively reflect the overall characteristics of the entire set of product images.

[0090] The model is trained using a labeled dataset, and the model parameters are adjusted by minimizing the error between the prediction results and the actual labels. During the training process, an appropriate loss function (such as mean squared error) is used to measure the gap between the predicted value and the true value, and the model parameters are updated through the backpropagation algorithm until the sum of the prediction errors reaches the convergence criterion.

[0091] Once the model training is completed, it can be used to evaluate a new set of product images and generate a jaggedness degree score. This score can help identify those product images with severely jagged regions, thereby providing guidance for image optimization.

[0092] In S5, for severely jagged regions, the edge curves are repaired, specifically including:

[0093] The wavelet coefficients are denoised using the hard threshold method to remove high-frequency noise in the image;

[0094] For the denoised wavelet coefficients and cubic spline interpolation is used for local curve repair, calculating the curve fitting function for each jagged region, and the calculation expression is: ;

[0095] where represents the arc length parameter of the curve, represents the curve fitting function, represents the spline, represents the total number of splines, represents the th fitting coefficient of the spline, represents the th spline basis function; obtaining the edge curvature of the jagged region, optimizing the edge curvature, and using the least squares method to adjust the fitting curve to obtain an image with jaggedness removed.

[0096] Working principle of the present invention: The present invention converts the original product picture into a grayscale image to remove color noise, and performs smoothing processing through filtering technology to ensure the accuracy of subsequent edge detection. Extract and analyze the edge curvature features in the product picture, use Canny edge detection and wavelet transform to identify the specific positions of boundary serrations, so as to accurately calibrate the serrated area. Next, perform boundary alignment on the pictures of the same product from different perspectives, calculate the shape difference coefficient to evaluate its consistency, and use the shape context descriptor and the Hungarian algorithm to achieve the optimal match. Combine the shape difference coefficient with the edge curvature features, transform them into a comprehensive feature vector as the input of a machine learning model (such as a graph neural network), extract global features through multi-layer graph convolution operations, and train the model to identify severely serrated areas. Finally, adopt the hard threshold method to denoise, and use the cubic spline interpolation method to repair the severely serrated edge curve, eliminating the serration phenomenon in the image. This method not only improves the overall quality of the product image, but also improves the user browsing experience, and is especially suitable for large-scale product image processing on e-commerce platforms, which helps to enhance the product display effect and improve the sales conversion rate. The entire process emphasizes the comprehensive optimization from preprocessing, feature extraction, consistency evaluation to image repair, demonstrating high technological innovation and application value.

[0097] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0098] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server or data center to another website, computer, server or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center containing one or more available medium sets. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0099] It should be understood that the term "and / or" in this text is merely a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. Here, A and B can be singular or plural. In addition, the character " / " in this text generally represents an "or" relationship between the associated objects before and after, but it may also represent an "and / or" relationship, and the specific meaning can be understood by referring to the context before and after.

[0100] It should be understood that in various embodiments of the present application, the magnitude of the serial numbers of the above processes does not mean the sequence of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0101] The above has described in detail one embodiment of the present invention, but the content described is only a preferred embodiment of the present invention and cannot be considered as being used to limit the scope of implementation of the present invention. Any equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. An e-commerce image dynamic optimization and adjustment method based on market feedback, characterized in that It includes the following steps: S1: Perform image preprocessing on the commodity image and convert it into a grayscale image to remove color noise that interferes with edge detection; S2: Extract the edge curvature features in the commodity image, calculate the change in edge curvature, calibrate the serrated area according to the threshold of the change in edge curvature, and identify the specific position of the boundary serration; S3: Based on the serrated area, align the boundaries of the images of the same commodity from multiple perspectives, calculate the difference in boundary shapes between different perspectives, and evaluate the consistency of the commodity images according to the degree of difference; S4: Convert the shape difference coefficient and edge curvature features of the commodity image into a comprehensive feature vector, which is used as an input item of the machine learning model, and identify the area with severe serration according to the model output result; S5: For the area with severe serration, repair the edge curve, specifically including: For the area with severe serration, repair the edge curve, specifically including: Use the hard threshold method to denoise the wavelet coefficients and remove the high-frequency noise in the image; On the denoised wavelet coefficients and the cubic spline interpolation method is used for local curve repair, and the curve fitting function of each serrated area is calculated. The calculation expression is: ; where represents the arc length parameter of the curve, represents the curve fitting function, represents the spline, represents the total number of splines, represents the th fitting coefficient of the spline, represents the th spline basis function; obtain the edge curvature of the serrated area, optimize the edge curvature, and use the least squares method to adjust the fitting curve to obtain an image without serrations; Among them, represents the scale parameter, represents the translation parameter, represents the wavelet transform coefficient in the x-direction, represents the wavelet transform coefficient in the y-direction.

2. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 1, wherein, The edge curvature features are extracted as follows: For the grayscale image after removing color noise, use Canny edge detection to obtain a set of all edge points in the image; Connect all the edge points into a continuous edge curve and perform wavelet transform on the edge curve; Through wavelet transform, convert the curve from the time domain to the frequency domain, so as to obtain the multi-scale features of the edge curve, denoted as edge curvature features.

3. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 2, wherein According to the edge curvature features, calculate the edge curvature, specifically including: Based on the multi-scale wavelet coefficients of the curve, the following formula is used to calculate the curvature: ; Among them, represents the change rate of wavelet coefficients, represents the scale parameter, represents the translation parameter, represents the wavelet transform coefficient in the x-direction, represents the wavelet transform coefficient in the y-direction; Obtain the edge curvature of each edge point by calculating the derivative of the wavelet coefficients.

4. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 3, characterized in that, According to the edge curvature of each edge point, identify the specific position of the boundary serration, specifically including: Obtain the edge curvature of each edge point, compare the edge curvature of each edge point with a preset threshold, and judge whether the edge curvature of each edge point is greater than or equal to the preset threshold. If so, mark it as a serrated area; if not, mark it as a non-serrated area.

5. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 1, wherein The calculation process of the difference in boundary shapes is: Obtain product images of the same product from multiple perspectives , represents the number of perspectives, and obtain the edge curve , represents the th point on the boundary, represents the total number of sampling points; For each point in the boundary point set , construct a shape context descriptor , and the calculation formula is: ; Among them, represents the th sampling point, represents the Euclidean distance between point and point and represent a point in the boundary point set, represents the polar angle between point and point represents the discrete partition index, represents the distance interval, represents the angle interval, represents the indicator function, which is 1 when the condition is satisfied and 0 otherwise, represents the shape context descriptor of point For any two perspective images and of the boundary point sets and , calculate the shape context descriptors of each point and respectively, construct a cost matrix, and each matrix element represents the shape context difference between point and point . The calculation formula is: ; In the formula, represents the number of distance partitions, represents the number of angle partitions, represents a small positive value to avoid a zero denominator, represents each matrix element; Use the Hungarian algorithm to solve the optimal matching of the cost matrix and establish a one-to-one correspondence between the boundary point sets and . According to the point set matching result, calculate the overall shape difference coefficient of the two boundary curves. The calculation expression is as follows: ; In the formula, represents the shape difference coefficient.

6. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 1, characterized in that Evaluating the consistency of commodity images specifically includes: Judge whether the difference coefficient of the commodity image is greater than or equal to the preset threshold. If so, the images of the corresponding commodity from multiple perspectives are inconsistent; if not, the images of the corresponding commodity from multiple perspectives are consistent.

7. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 1, characterized in that, The construction process of the comprehensive feature vector is: Obtain the shape difference coefficient and edge curvature features of the commodity image, and standardize the shape difference coefficient and edge curvature features; use the clustering algorithm to cluster the standardized shape difference coefficient and edge curvature feature vectors to obtain several clusters, and each cluster corresponds to a type of commodity image; calculate the center point of each cluster, and the center point of the cluster is used as the comprehensive feature vector of the commodity images within the corresponding cluster, and the comprehensive feature vector represents the common features of the images within the cluster; Assign each commodity image to the corresponding cluster, and use the center point of the corresponding cluster as the comprehensive feature vector of the commodity image; The clustering algorithm is the K-means clustering algorithm, and the specific process is as follows: randomly select M initial cluster centers; calculate the distances between the feature vectors of each image and all the center points, and assign the image to the cluster center with the closest distance; update the center point of each cluster to the mean value of all the image feature vectors within the cluster; repeat the clustering process until the cluster centers no longer change, and the clustering process ends.

8. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 1, wherein, The construction process of the machine learning model is as follows: Replace the shape difference degree coefficient and edge curvature features of the commodity image with a comprehensive feature vector as the input item of the machine learning model, and the machine learning model is a graph neural network model; use graph convolution operations to gradually extract local and global features, and finally generate a global graph representation; through the processing of multiple layers of graph convolution, the graph neural network can identify each group of input comprehensive feature vectors, adjust the model parameters by minimizing the error between the prediction result and the actual annotation, train the machine learning model until the sum of the prediction errors reaches convergence, and then stop the model training. Finally, a sawtooth degree score is generated to identify the severely sawtoothed area.

9. The method for dynamically optimizing and adjusting e-commerce images based on market feedback according to claim 8, characterized in that, Identifying the severely sawtoothed area specifically includes: According to the recognition result of the machine learning model, output the sawtooth degree score for each sawtoothed area; Judge whether the sawtooth degree score of each sawtoothed area is greater than or equal to the preset threshold. If so, the corresponding area is a severely sawtoothed area; if not, the corresponding area is a non-severely sawtoothed area.

Images