Advertisement design assisting method and system based on image style migration

Abstract

The invention relates to the technical field of style migration, in particular to an advertisement design assisting method and system based on image style migration, and the method comprises the following steps: based on an input style image sample, analyzing the edge intensity characteristics of the image through calculating the brightness difference between a plurality of pixel points and neighborhood pixel points in the sample image, and obtaining the edge intensity characteristics of the image; in combination with analysis of texture modes and color distribution of images, a style feature data set is obtained by matching importance weights for various features. According to the method and the device, the style of the image sample is extracted, and the content of the processed image is extracted, so that the accuracy of style migration is improved, the style and the content of the target advertisement image are finely controlled, the definition and the conveying effect of the advertisement content are kept, and key visual features of continuous frame images are continuously monitored and adjusted; according to the method, the consistency of image styles among continuous frames is improved, and the consistency of the overall styles and the advertisement expressive force are ensured through analysis and adjustment of the style matching degree among a plurality of design elements.

Description

Technical Field

[0001] The present invention relates to the technical field of style transfer, and in particular to an auxiliary method and system for advertisement design based on image style transfer. Background Art

[0002] The field of style transfer technology focuses on applying the style of one image to another image while keeping the latter's content structure unchanged. It is driven by neural networks and applied to digital media, art creation, image editing, and advertising production. It draws images with a specific artistic style and makes the target image gradually approach the style features of the source style image and the content features of the target content image. This includes imitating the painting styles of famous painters and creating visual elements that conform to specific design themes, aiming to enhance the visual effect and add additional emotional or visual impact while maintaining the semantic content of the original image.

[0003] Among them, the advertising design assistance method based on image style transfer aims to use style transfer technology to assist the advertising creativity and design process. By automatically applying a certain artistic style to the images in the advertisement, it can quickly generate artistic advertising designs, help improve the visual appeal of advertisements, enhance brand recognition and enhance consumers' visual experience. It helps advertising designers quickly realize creative ideas, transform conventional advertising images into advertising works with high personalization and aesthetic value, optimize the creative display of advertisements, and effectively improve the communication effect of advertisements and the emotional resonance of the audience.

[0004] The application of traditional advertising design auxiliary methods in advertising design is usually limited to simply transferring artistic styles to commercial images. It cannot fully consider the deep integration between the content and style of the advertisement, resulting in unclear advertising information, making it difficult to convey advertising information and attract target groups, and weakening the communication effect. When dealing with advertisements with complex dynamic elements, it is difficult to maintain style consistency between consecutive frames, which leads to visual incoherence in user experience and affects the overall perceived quality of the advertisement. There is a lack of effective style and visual feature coordination mechanism, making it difficult to enhance the appeal of the advertisement. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and to propose an advertising design auxiliary method and system based on image style transfer.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: an advertisement design auxiliary method based on image style transfer, comprising the following steps:

[0007] S1: Based on the input style image sample, by calculating the brightness difference between multiple pixels and neighboring pixels in the sample image, analyzing the edge intensity features in the image, combining the analysis of the image's texture pattern and color distribution, and matching importance weights for multiple features, a style feature dataset is obtained;

[0008] S2: Based on the style feature dataset, multiple design elements in the advertising image to be processed are identified, content features of the target image are extracted, the color and texture of the target advertising image are adjusted, and a style-transferred image is generated;

[0009] S3: Based on the style-transferred image, visual features of the processed image are identified and marked, and processing parameters of the style transfer are adjusted by monitoring changes in key visual features in consecutive frame images, including brightness, texture, and color processing parameters of multiple visual feature points, to obtain consecutive frame processing results;

[0010] S4: Based on the continuous frame processing results, by analyzing the style matching between multiple design elements in the advertising image, performing style consistency detection on the image, adjusting the style feature weights of multiple elements, and obtaining a consistency calibration result.

[0011] As a further solution of the present invention, the step of acquiring the edge strength feature is specifically:

[0012] S111: based on the input style image sample, extract the brightness values ​​of multiple pixels in the image to generate a brightness value matrix;

[0013] S112: Based on the brightness value matrix, by formula:

[0014]

[0015] Calculate the edge strength around each pixel to obtain the edge strength matrix;

[0016] Among them, E i,j Represents the edge strength of the pixel located in the i-th row and j-th column, P i,j represents the brightness value of the pixel in the i-th row and j-th column, P i+1,j Indicates that it is located at P i,j The brightness value of the pixel below, P i-1,j Indicates that it is located at P i,j The brightness value of the pixel above, P i,j+1 Indicates that it is located at P i,j The brightness value of the pixel on the right, P i,j-1 Indicates that it is located at P i,j The brightness value of the pixel on the left, i is the row index, j is the column index;

[0017] S113: Based on the edge intensity matrix, identify and extract pixel points with edge intensity higher than the average level to obtain edge intensity features.

[0018] As a further solution of the present invention, the steps of acquiring the style feature data set are specifically as follows:

[0019] S121: Analyze the sample image based on the edge strength feature, extract the statistical characteristics of the texture and the frequency distribution of the color, and generate texture pattern and color distribution data;

[0020] S122: Based on the texture pattern and color distribution data, by formula:

[0021] w k =α·S k +β·F k +γ·C k

[0022] Calculate feature weight values;

[0023] Among them, w k is the weight of the kth feature, α is the significance S k The weight adjustment coefficient, S k is the significance score of the kth feature, and β is the frequency F k The weight adjustment coefficient, F k is the frequency of occurrence of the kth feature, γ is the correlation C k The weight adjustment coefficient, C k is the correlation between the kth feature and the target style;

[0024] S123: Based on the feature weight values, a style feature data set is generated by integrating and recording multiple features of the target sample image.

[0025] As a further solution of the present invention, the step of acquiring the content features of the target image is specifically as follows:

[0026] S211: Based on the style feature data set, identify multiple design elements in the target advertisement image to be processed, including graphics, texts, and layout structures, and generate a design element list;

[0027] S212: Using the design element list, use the formula:

[0028] V=A×T v +L×L v +Y×Y v

[0029] Calculate visual importance scores for multiple design elements;

[0030] Where V is the visual importance score of the target element, A is the relative area of ​​the element, and T v is the weight coefficient of the relative area of ​​the element, L is the positional advantage of the element, and L v is the weight coefficient of the position, Y is the color contrast of the element, and Y v is the weight coefficient of color contrast;

[0031] S213: Based on the visual importance scores, the visual influence of multiple design elements is evaluated, and the content features of the target image are extracted by combining the position and type information of the multiple design elements.

[0032] As a further solution of the present invention, the step of acquiring the style-transferred image is specifically as follows:

[0033] S221: Based on the content features of the target image, by formula:

[0034] CY′ x,y =δ·CY x,y +(1-δ)·CY s,x,y

[0035] Calculate the adjusted color value;

[0036] Among them, CY′ x,y is the adjusted color value, δ is the color fusion coefficient, CY x,y is the original color value of the target advertising image at position (x, y), CY s,x,y is the color value at position (x, y) in the style feature data set, where x represents the horizontal coordinate in the image and y represents the vertical coordinate in the image;

[0037] S222: Based on the adjusted color value, by formula:

[0038] TW′ x,y =∈·TW x,y +(1-∈)·TW s,x,y

[0039] Calculate the adjusted texture value to obtain the color and texture adjustment value;

[0040] Among them, TW′ x,y is the adjusted texture value, ∈ is the texture fusion coefficient, TW x,y is the original texture value of the target advertising image at position (x, y), TW s,x,y is the texture value at position (x, y) in the style feature data set;

[0041] S223: Based on the color and texture adjustment values, multiple pixels of the target advertisement image are adjusted to obtain a style-transferred image.

[0042] As a further solution of the present invention, the step of obtaining the continuous frame processing results is specifically:

[0043] S311: Based on the style-transferred image, identify and mark visual feature points, including edges and corners in the image, to generate a visual feature point map;

[0044] S312: Based on the visual feature point map, analyze the marked multiple visual feature points, extract image feature data of the multiple feature points in continuous frames, including color, texture, and brightness, to obtain a feature point data set;

[0045] S313: Based on the feature point data set, by formula:

[0046]

[0047] Calculate the color, texture and brightness consistency scores of each visual feature point in consecutive frames to obtain the consecutive frame consistency score;

[0048] Among them, U p is the consistency score of the visual feature point p in multiple frames, YS n,p is the color value of feature point p in the nth frame, WL n,p is the texture value of feature point p in the nth frame, LD n,p is the brightness value of feature point p in the nth frame, is the average color value of feature point p in all frames, is the average texture value of feature point p in all frames, is the average brightness value of feature point p in all frames, N is the total number of frames, n is the frame index, and p is the feature point index;

[0049] S314: Based on the continuous frame consistency score, adjust the image processing parameters in the continuous frame images, including the brightness, texture, and color processing parameters of multiple visual feature points, optimize the visual consistency of the continuous frame images, and obtain the continuous frame processing results.

[0050] As a further solution of the present invention, the step of obtaining the consistency calibration result is specifically:

[0051] S411: extracting information of multiple design elements in the processed advertising image based on the continuous frame processing results, including brightness information, color information, and texture information, and generating design element feature data;

[0052] S412: Based on the design element characteristic data, by formula:

[0053]

[0054] Calculate the structural similarity index between multiple design elements to obtain the style matching score;

[0055] Among them, S represents the structural similarity index, m o is the average brightness of element o, m l is the average brightness of element l, σ o is the standard deviation of the brightness of element o, σ l is the standard deviation of the brightness of element l, σ ol is the brightness covariance between elements o and l, k1 and k2 are small constants added to ensure calculation stability;

[0056] S413: Based on the style matching score, by analyzing the style consistency of the target advertisement image, adjusting the colors, textures and shapes of multiple design elements, optimizing the style coordination of the visual effect, and obtaining a consistency calibration result.

[0057] An advertisement design assistance system based on image style transfer, the advertisement design assistance system based on image style transfer is used to execute the above-mentioned advertisement design assistance method based on image style transfer, the system comprises:

[0058] The sample feature analysis module analyzes the edge strength features of the image based on the input style image sample, combines the color information and texture information, and performs quantitative analysis on the texture pattern and color distribution of the image. It generates a style feature dataset by matching importance weights for multiple features in the image.

[0059] The design element adjustment module identifies multiple design elements in the target advertisement image based on the style feature data set, including graphics, text, and layout structure, and adjusts the color and texture of the target advertisement image in combination with the style features of the sample image to generate a style-transferred image;

[0060] The visual feature marking module identifies and marks visual features of the processed image based on the style-transferred image, including edges and corners, and adjusts the processing parameters of the style transfer, including the brightness, texture, and color processing parameters of multiple visual feature points, by monitoring the changes in key visual features in continuous frame images, to generate continuous frame processing results;

[0061] The design element analysis module analyzes the style matching between multiple design elements in the advertisement image based on the continuous frame processing results through structural similarity evaluation, performs style consistency detection on the image, and generates a style coordination score;

[0062] The image feature adjustment module adjusts the color, texture and shape of the advertisement image based on the style coordination score, optimizes the coordination of the image style, and generates a consistency calibration result.

[0063] Compared with the prior art, the advantages and positive effects of the present invention are:

[0064] In the present invention, by performing style extraction on image samples and combining it with content extraction of processed images, the accuracy of style transfer is improved, the visual impact of images is enhanced, and fine control of the style and content of target advertising images is achieved, while retaining the clarity and communication effect of advertising content. By continuously monitoring and adjusting key visual features of continuous frame images, the consistency of images between continuous frames is improved, the overall aesthetics of dynamic advertisements is enhanced, and the coherence of advertising narratives is enhanced. By analyzing and adjusting the style matching between multiple design elements, the consistency of styles between advertising design elements is ensured, the visual performance of advertisements is optimized, and the communication effect of advertisements is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0066] Figure 2 It is a flow chart of obtaining edge strength features of the present invention;

[0067] Figure 3 A flow chart of generating a style feature data set of the present invention;

[0068] Figure 4 It is a flow chart of obtaining content features of a target image according to the present invention;

[0069] Figure 5 A flow chart of obtaining an image after style transfer according to the present invention;

[0070] Figure 6 A flow chart of generating continuous frame processing results of the present invention;

[0071] Figure 7 This is a flow chart of obtaining consistency calibration results of the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0073] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0074] See also Figure 1 The present invention provides a technical solution, an advertisement design auxiliary method based on image style transfer, comprising the following steps:

[0075] S1: Based on the input style image sample, by calculating the brightness difference between multiple pixels and neighboring pixels in the sample image, the edge intensity features in the image are analyzed. Combined with the color information and texture information in the image, the texture pattern and color distribution of the image are analyzed. By matching importance weights for multiple features, a style feature dataset is obtained.

[0076] S2: Based on the style feature dataset, by identifying multiple design elements in the advertising image to be processed, including graphics, text, and layout structure, the content features of the target image are extracted. According to the style features of the sample image, the color and texture of the target advertising image are adjusted to generate a style-transferred image.

[0077] S3: Based on the style-transferred image, the visual features of the processed image are identified and marked, including edges and corners. By monitoring the changes in key visual features in consecutive frame images, the processing parameters of the style transfer are adjusted, including the brightness, texture, and color processing parameters of multiple visual feature points, to obtain the consecutive frame processing results.

[0078] S4: Based on the results of continuous frame processing, the style consistency of the image is detected by analyzing the style matching between multiple design elements in the advertising image. Through structural similarity evaluation, the style feature weights of multiple elements are adjusted to correct the style coordination of color, texture and shape to obtain consistency calibration results.

[0079] The style feature data set specifically includes edge intensity feature information, color distribution feature information, and texture pattern feature data set. The image after style transfer includes content feature extraction records, texture feature and color feature migration records, and design element layout information. The continuous frame processing results include edge tracking accuracy information, corner point stability data, and inter-frame adjustment parameters. The consistency calibration results include style weight optimization data, color consistency index, and texture and shape matching data.

[0080] See also Figure 2 , the steps for obtaining edge strength features are as follows:

[0081] S111: based on the input style image sample, extract the brightness values ​​of multiple pixels in the image to generate a brightness value matrix;

[0082] In sub-step S111, based on the input style image sample, each pixel in the image is traversed, and an image processing library such as OpenCV is used to perform grayscale conversion on the image, extract the brightness value of each pixel, use a two-dimensional array to store the target brightness value, and generate a brightness value matrix. The matrix dimension is the same as the original image, and each matrix element represents the brightness level of the corresponding pixel. By normalizing the matrix, the data range is standardized, and the data stability and efficiency of subsequent processing are improved.

[0083] S112: Based on the brightness value matrix, by formula:

[0084]

[0085] Calculate the edge strength around each pixel to obtain the edge strength matrix;

[0086] Among them, E i,j Represents the edge strength of the pixel located in the i-th row and j-th column, P i,j represents the brightness value of the pixel in the i-th row and j-th column, P i+1,j Indicates that it is located at P i,j The brightness value of the pixel below is used to calculate the brightness of the pixel below. i,j The brightness difference in the vertical direction, P i-1,j Indicates that it is located at P i,j The brightness value of the pixel above is used to calculate the brightness of the pixel with P i,j The brightness difference in the vertical direction, P i,j+1 Indicates that it is located at P i,j The brightness value of the pixel on the right is used to calculate the brightness of the pixel i,j The brightness difference on the right side in the horizontal direction, P i,j-1 Indicates that it is located at P i,j The brightness value of the pixel on the left is used to calculate the brightness of the pixel with P i,j The brightness difference on the left side in the horizontal direction, i is the index of the row, which is used to indicate the position of the pixel in the image, and j is the index of the column, which is used to indicate the position of the pixel in the image;

[0087] formula:

[0088]

[0089] Parameter meaning and acquisition method:

[0090] P i,j : is the brightness value of the pixel at the i-th row and j-th column, which is directly obtained from the sample image through the image acquisition device.

[0091] P i+1,j : is located at P i,j The brightness values ​​of the pixels below are directly obtained from the sample image through the image acquisition device;

[0092] P i-1,j : is located at P i,j The brightness value of the pixel above is directly obtained from the sample image through the image acquisition device;

[0093] P i,j+1 : is located at P i,j The brightness value of the pixel on the right is directly obtained from the sample image through the image acquisition device;

[0094] P i,j-1 : is located at P i,j The brightness value of the pixel on the left is directly obtained from the sample image through the image acquisition device;

[0095] Calculation example:

[0096] Set the brightness value of the target pixel to 150, and the brightness values ​​of its neighboring pixels to P i+1,j =160, P i-1,j =145,P i,j+1 =152,P i,j-1 =148, calculate edge strength E i,j :

[0097] Calculate P i,j The square of the brightness difference with its four surrounding neighboring pixels:

[0098] (150-160) 2 =100;

[0099] (150-145) 2 =25;

[0100] (150-152) 2 =4;

[0101] (150-148) 2 =4;

[0102] Compute the sum of squared differences:

[0103] 100+25+4+4=133;

[0104] Calculate E i,j :

[0105]

[0106] E i,j The result is 11.53, which indicates that the target pixel has a relatively high edge strength in the image, reflecting the contrast change of the pixel in its neighborhood. The calculation process is used to identify key edge features in the image, which is especially important for feature enhancement in style migration.

[0107] S113: Based on the edge intensity matrix, identify and extract pixel points with edge intensity higher than the average level to obtain edge intensity features;

[0108] In sub-step S113, based on the generated brightness value matrix, the Canny algorithm is used for edge detection. The edge strength is determined by calculating the gradient of the image brightness value. By setting a threshold, pixels with edge strength higher than the average level are identified. The position and brightness value of the target pixel are recorded to obtain the edge strength feature. The threshold used is 1.5 times the average value of the brightness gradient to ensure that only obvious edges are extracted. The results are stored in a list form, including the coordinates of each edge point and the corresponding brightness value.

[0109] See also Figure 3 , the steps to obtain the style feature dataset are as follows:

[0110] S121: Analyze the sample image based on the edge intensity feature, extract the statistical characteristics of the texture and the frequency distribution of the color, and generate texture pattern and color distribution data;

[0111] In sub-step S121, based on the edge intensity features, the gray-level co-occurrence matrix method is used to analyze the texture pattern in the sample image, evaluate the repetitive pattern and spatial relationship of pixel pairs, calculate the statistical characteristics of the texture, such as contrast, uniformity and entropy, and use the color histogram technology to analyze the color frequency distribution. The frequency of occurrence of different color values ​​in the image is counted to obtain the color frequency distribution data. The generated texture pattern and color distribution data are stored as two independent structures, which contain texture statistical characteristics and color frequency distribution values, providing basic data for subsequent analysis.

[0112] S122: Based on texture pattern and color distribution data, by formula:

[0113] w k =α·S k +β·F k +γ·C k

[0114] Calculate feature weight values;

[0115] Among them, w kis the weight of the kth feature, reflecting the importance of the feature in the overall style description, and α is the significance S k The weight adjustment coefficient, S k is the significance score of the kth feature, indicating the significance of the feature in the image, and β is the frequency F k The weight adjustment coefficient is used to adjust the influence of feature frequency on weight calculation, F k is the frequency of occurrence of the kth feature, indicating the prevalence of the feature in the image set, and γ is the correlation C k The weight adjustment coefficient is used to measure the direct impact of features on the style transfer effect, C k is the correlation between the kth feature and the target style, indicating the strength of association between the feature and a specific style;

[0116] formula:

[0117] w k =α·S k +β·F k +γ·C k ;

[0118] Parameter meaning and acquisition method:

[0119] α: is the weight adjustment coefficient of significance, which is set by experts;

[0120] S k : is the significance score of the k-th feature, the score obtained by edge detection;

[0121] β: is the frequency weight adjustment coefficient, obtained based on statistical analysis;

[0122] F k : is the frequency of occurrence of the kth feature, obtained by counting the frequency of multiple image samples;

[0123] γ: is the weight adjustment coefficient of correlation, which is obtained by calculating the correlation coefficient between the feature and the target style;

[0124] C k : is the correlation between the kth feature and the target style, calculated by the Pearson correlation coefficient;

[0125] Calculation example:

[0126] Set α = 0.3, S k =0.8,β=0.4,F k =0.5,γ=0.3,C k =0.7, calculate the feature weight value:

[0127] w k=0.3 0.8 + 0.4 0.5 + 0.3 0.7;

[0128] w k =0.24+0.2+0.21;

[0129] w k =0.65;

[0130] The result of 0.65 indicates that the comprehensive weight of this feature is high, reflecting its significant impact on the image style. The result is used to guide the processing of this feature during style transfer to ensure that the style transfer accurately reflects the style characteristics of the original image.

[0131] S123: Based on the feature weight values, a style feature dataset is generated by integrating and recording multiple features of the target sample image;

[0132] In sub-step S123, based on the aforementioned texture and color feature data, a weighted method is adopted to integrate multiple image features, and feature weight values ​​are used. The target weight is determined by pre-set algorithm parameters, including principal component analysis, which reduces data dimensions while maximizing data variables. Texture statistical features and color distribution features are integrated to generate a style feature data set. The data set contains key visual elements of image samples for further style migration or image classification work.

[0133] See also Figure 4 , the steps for obtaining the content features of the target image are as follows:

[0134] S211: Based on the style feature data set, identify multiple design elements in the target advertising image that need to be processed, including graphics, texts, and layout structures, and generate a design element list;

[0135] In sub-step S211, based on the style feature dataset, an image segmentation algorithm, such as the region growing method, is used to identify multiple design elements in the target advertising image. The target elements include graphics, texts, and layout structures. The identification of each element is based on pixel similarity and geometric continuity. A detailed list of all key design elements and attributes is generated. The list is stored in JSON format to ensure that each design element is accurately recorded and can be used for subsequent processing.

[0136] S212: Through the list of design elements, use the formula:

[0137] V=A×T v +L×L v +Y×Y v

[0138] Calculate visual importance scores for multiple design elements;

[0139] Among them, V is the visual importance score of the target element, A is the relative area of ​​the element, which reflects the space proportion of the element in the visual presentation, and T v is the weight coefficient of the relative area of ​​the element, L is the positional advantage of the element, and L v is the weight coefficient of the position, Y is the color contrast of the element, and Y v is the weight coefficient of color contrast, which is used to adjust the contribution of color contrast in visual importance scoring;

[0140] formula:

[0141] V=A×T v +L×L v +Y×Y v ;

[0142] Parameter meaning and acquisition method:

[0143] A: The relative area of ​​an element, which indicates the proportion of the element in the image. It is obtained by measuring the ratio of the number of pixels occupied by the element to the total number of pixels in the image using image analysis software.

[0144] T v : Weight coefficient of element relative area;

[0145] L: The positional dominance of the element, obtained by analyzing the specific position of the element in the image;

[0146] L v : The weight coefficient of the position is used to adjust the influence of the position factor in the visual importance score, which is preset according to the strategic importance of the element and the visual marketing principles;

[0147] Y: The color contrast of an element, which is obtained by quantitatively analyzing the color difference between the element and its surrounding area, and reflects the visual prominence of the element;

[0148] Y v : The weight coefficient of color contrast, which is used to adjust the contribution of color contrast in the visual importance score and is set based on the visual appeal and psychological impact of color;

[0149] Calculation example:

[0150] Set A = 0.05, T v =1.5,L=0.8,L v =2.0,Y=0.7,Y v =1.0, calculate the visual importance of the target design element:

[0151] V=0.05×1.5+0.8×2.0+0.7×1.0;

[0152] V = 0.075 + 1.6 + 0.7;

[0153] V = 2.375;

[0154] The result of 2.375 indicates that this element has a relatively high importance in the visual design. The result is used to guide the style transfer and optimization of the image, ensuring that the design decisions make full use of the visual impact of each element.

[0155] S213: Based on the visual importance scores, the visual influence of multiple design elements is evaluated, and the content features of the target image are extracted by combining the position and type information of the multiple design elements;

[0156] In sub-step S213, visual saliency detection technology, such as graph-based saliency model, is used to evaluate the visual influence of each design element in the image. The color and contrast features of the elements are combined with position and type information to calculate the visual importance score of each element. The content features of the target image are extracted and recorded. The content features are expressed in vector form, including the spatial distribution and visual weight of the elements.

[0157] See also Figure 5 , the steps to obtain the image after style transfer are as follows:

[0158] S221: Based on the content features of the target image, the formula is:

[0159] CY′ x,y =δ·CY x,y +(1-δ)·CY s,x,y

[0160] Calculate the adjusted color value;

[0161] Among them, CY′ x,y is the adjusted color value, δ is the color fusion coefficient, which is used to control the degree of fusion between the original color and the style color, CY x,y is the original color value of the target advertising image at position (x, y), CY s,x,y is the color value at position (x, y) in the style feature data set, where x represents the horizontal coordinate in the image, used to specify the horizontal position of the pixel, and y represents the vertical coordinate in the image, used to specify the vertical position of the pixel;

[0162] formula:

[0163] CY′ x,y =δ·CY x,y +(1-δ)·CY s,x,y ;

[0164] Parameter meaning and acquisition method:

[0165] δ: color fusion coefficient;

[0166] CY x,y : The original color value of the target advertising image at position (x, y);

[0167] CY s,x,y : The color value at position (x, y) in the style feature data set;

[0168] Calculation example:

[0169] Set δ = 0.7, CY x,y =150, CY s,x,y =100, calculate CY′ x,y :

[0170] CY′ x,y =0.7·150+0.3·100=105+30=135;

[0171] ResultsCY′ x,y =135 indicates that the color at position (x, y) is adjusted to 135, showing the effect of style transfer, the calculation process is used to adjust the image color to match a given style.

[0172] S222: Based on the adjusted color value, by the formula:

[0173] TW′ x,y =∈·TW x,y +(1-∈)·TW s,x,y

[0174] Calculate the adjusted texture value to obtain the color and texture adjustment value;

[0175] Among them, TW′ x,y is the adjusted texture value, ∈ is the texture fusion coefficient, which is used to control the fusion degree of the original texture and the style texture, TW x,y is the original texture value of the target advertising image at position (x, y), TW s,x,y is the texture value at position (x, y) in the style feature data set;

[0176] formula:

[0177] TW′ x,y =∈·TW x,y +(1-ε)TW s,x,y ;

[0178] Parameter meaning and acquisition method:

[0179] ε: texture fusion coefficient;

[0180] TW x,y : The original texture value of the target advertising image at position (x, y);

[0181] TW s,x,y : The texture value at position (x, y) in the style feature data set;

[0182] Calculation example:

[0183] Set ∈ = 0.5, TW x,y =0.6, TW s,x,y = 0.4, calculate TW′ x,y :

[0184] TW′ x,y =0.5 0.6 + 0.5 0.4 = 0.3 + 0.2 = 0.5;

[0185] Calculation result TW′ x,y =0.5 indicates that the texture at position (x, y) is adjusted to 0.5. The formula is used to adjust the image texture and integrate the target style.

[0186] S223: adjusting multiple pixels of the target advertisement image based on the color and texture adjustment values ​​to obtain a style-transferred image;

[0187] In sub-step S223, based on the color and texture adjustment values, image processing technology is used to adjust multiple pixel points of the target advertising image, and a hierarchical adjustment method, such as multi-scale fusion, is adopted. The adjustment includes color balance and texture smoothing. The parameters include color temperature and saturation setting values ​​for color adjustment, and blurriness and sharpness setting values ​​for texture adjustment. Through the fine adjustment of the target parameters, the generation of the image after style transfer is achieved, and the resulting image shows a more coordinated and attractive visual style.

[0188] See also Figure 6 , the specific steps for obtaining the continuous frame processing results are:

[0189] S311: Based on the style-transferred image, identify and mark visual feature points, including edges and corners in the image, and generate a visual feature point map;

[0190] In sub-step S311, the Harris corner detection algorithm is used to perform visual feature point recognition on the image after style transfer. Based on the sensitivity of motion changes of pixels in a local window of the image, the edges and corners in the image are identified, and coordinates and intensity values ​​are generated for each feature point. The results are stored in a visual feature point map, which is represented in the form of a two-dimensional array, and each array element is marked as a feature point.

[0191] S312: Based on the visual feature point map, analyze the marked multiple visual feature points, extract image feature data of the multiple feature points in continuous frames, including color, texture, and brightness, and obtain a feature point data set;

[0192] In sub-step S312, feature analysis methods based on visual feature point maps, such as grayscale co-occurrence matrix and color histogram, are used to extract the texture and color characteristics of each marked feature point in continuous frames, respectively. Brightness information is obtained through local area brightness analysis, and the target information is integrated to generate a feature point data set. The data set contains image feature data such as color, texture, brightness, etc. for each feature point. The target data is in vector form for subsequent image analysis or machine learning training.

[0193] S313: Based on the feature point data set, the formula is:

[0194]

[0195] Calculate the color, texture and brightness consistency scores of each visual feature point in consecutive frames to obtain the consecutive frame consistency score;

[0196] Among them, U p is the consistency score of the visual feature point p in multiple frames, YS n,p is the color value of feature point p in the nth frame, WL n,p is the texture value of feature point p in the nth frame, LD n,p is the brightness value of feature point p in the nth frame, is the average color value of feature point p in all frames, is the average texture value of feature point p in all frames, is the average brightness value of feature point p in all frames, N is the total number of frames, n is the frame index, and p is the feature point index;

[0197] formula:

[0198]

[0199] Parameter meaning and acquisition method:

[0200] U p : is the consistency score of feature point p, which is calculated by the root mean square of the color, texture, and brightness differences in multiple frames of data;

[0201] YS n,p : The color value of feature point p in the nth frame;

[0202] WL n,p : Texture value of feature point p in the nth frame;

[0203] LD n,p : The brightness value of feature point p in the nth frame;

[0204] The average color value of feature point p in all frames;

[0205] The average texture value of feature point p in all frames;

[0206] The average brightness value of feature point p in all frames;

[0207] N: the total number of frames involved in the calculation;

[0208] Calculation example:

[0209] Set specific values ​​to perform example calculations:

[0210] Assume that there are 3 frames of image data, the color value YS of feature point p in consecutive frames = [100, 105, 95], the texture value WL of feature point p = [0.5, 0.55, 0.45], the brightness value LD of feature point p = [200, 205, 195], and calculate the consistency score:

[0211] Calculate the average:

[0212]

[0213]

[0214] Calculate U p :

[0215]

[0216] Results p =4.714 represents the average difference in color, texture, and brightness changes of feature point p in three frames, indicating that the feature point has high visual consistency in multiple frames. The calculation process is used to evaluate the stability of image feature points in video sequences, and the numerical value reflects the degree of change of visual features.

[0217] S314: Based on the continuous frame consistency score, adjust the image processing parameters in the continuous frame images, including the brightness, texture, and color processing parameters of the multiple visual feature points, optimize the visual consistency of the continuous frame images, and obtain the continuous frame processing results;

[0218] In sub-step S314, the image processing parameters in the continuous frame images are adjusted according to the continuous frame consistency score, and dynamic parameter adjustment technology, such as adaptive filtering, is used to dynamically adjust the brightness, contrast and saturation according to the visual continuous frame consistency score between continuous frames to optimize image continuity and obtain continuous frame processing results. The result ensures that the image sequence maintains consistency in visual performance.

[0219] See also Figure 7 , the specific steps for obtaining the consistency calibration results are:

[0220] S411: extracting information of multiple design elements in the processed advertising image based on the continuous frame processing results, including brightness information, color information, and texture information, and generating design element feature data;

[0221] In sub-step S411, image analysis tools such as MATLAB or Python's image processing library are used to extract the brightness information, color information and texture information of the advertising image design elements in the continuous frame processing results, the gray level co-occurrence matrix is ​​used to evaluate the texture characteristics, the color histogram is used to analyze the color distribution, the local area brightness method is used to evaluate the brightness change, and the target data is integrated to generate a design element feature data set. The data set records the visual attributes of each design element and provides a basis for subsequent visual effect optimization.

[0222] S412: Based on the design element characteristic data, through the formula:

[0223]

[0224] Calculate the structural similarity index between multiple design elements to obtain the style matching score;

[0225] Among them, S represents the structural similarity index, m o is the average brightness of element o, reflecting the average brightness level of element o, m l is the average brightness of element l, reflecting the average brightness level of element l, σ o is the standard deviation of the brightness of element o, σ l is the standard deviation of the brightness of element l, σ ol is the brightness covariance between elements o and l, k1 and k2 are small constants added to ensure calculation stability;

[0226] formula:

[0227]

[0228] Parameter meaning and acquisition method:

[0229] m o : average brightness of element o;

[0230] m l : average brightness of element l;

[0231] σ o : The brightness standard deviation of element o, which measures the brightness variation;

[0232] σ l : The brightness standard deviation of element l, which measures the degree of brightness variation;

[0233] σ ol: The brightness covariance between elements o and l, which measures the relationship between the brightness changes of the two elements;

[0234] k1, k2: small constants set to maintain calculation stability;

[0235] Calculation example:

[0236] Set the average brightness of elements o and l to m o =120,m l =115, the brightness standard deviations are σ o =15,σ l =20, brightness covariance is σ ol =18, set k1 = 0.01 × 255 2 =650.25, k2=0.03×255 2 =1950.75, calculate the structural similarity score:

[0237]

[0238] The result of 0.787 shows that elements o and l have a high visual style consistency. The value reflects the similarity of brightness and texture between the two design elements. The calculation process is used to evaluate the visual coordination between different design elements in the advertising image.

[0239] S413: Based on the style matching score, by analyzing the style consistency of the target advertising image, adjusting the color, texture and shape of multiple design elements, optimizing the style coordination of the visual effect, and obtaining a consistency calibration result;

[0240] In sub-step S413, based on the style matching score, a style adjustment algorithm is used, such as k-means clustering analysis to optimize color grouping, texture matching technology to evaluate and adjust texture similarity, and shape adjustment using vector-based graphics conversion technology to adjust the color, texture and shape of multiple design elements in the advertising image. Through the comprehensive application of target technologies, the style consistency of the visual effect is optimized and improved, and a consistency calibration result is obtained, making the overall advertising vision more harmonious and unified.

[0241] An advertisement design assistance system based on image style transfer, the advertisement design assistance system based on image style transfer is used to execute the above-mentioned advertisement design assistance method based on image style transfer, and the system includes:

[0242] The sample feature analysis module analyzes the edge strength features of the image based on the input style image sample, combines the color information and texture information, and performs quantitative analysis on the texture pattern and color distribution of the image. It generates a style feature dataset by matching importance weights for multiple features in the image.

[0243] The design element adjustment module identifies multiple design elements in the target advertising image, including graphics, text, and layout structure, based on the style feature dataset, and adjusts the color and texture of the target advertising image in combination with the style features of the sample image to generate a style-transferred image.

[0244] The visual feature tagging module identifies and tags the visual features of the processed image, including edges and corners, based on the style-transferred image. By monitoring the changes in key visual features in consecutive frame images, the processing parameters of the style transfer are adjusted, including the brightness, texture, and color processing parameters of multiple visual feature points, to generate consecutive frame processing results.

[0245] The design element analysis module analyzes the style matching between multiple design elements in the advertising image based on the continuous frame processing results through structural similarity evaluation, performs style consistency detection on the image, and generates a style coordination score;

[0246] The image feature adjustment module adjusts the color, texture and shape of the advertising image based on the style coordination score, optimizes the coordination of the image style, and generates a consistent calibration result.

[0247] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An advertising design auxiliary method based on image style transfer, characterized in that: The following steps are involved: Based on the input style image sample, by calculating the brightness difference between multiple pixels and neighboring pixels in the sample image, analyzing the edge intensity features in the image, combining the analysis of the image's texture pattern and color distribution, and matching importance weights for multiple features, a style feature dataset is obtained; Based on the style feature data set, multiple design elements in the advertising image to be processed are identified, content features of the target image are extracted, the color and texture of the target advertising image are adjusted, and a style-transferred image is generated; Based on the style-transferred image, visual features of the processed image are identified and marked, and processing parameters of the style transfer are adjusted by monitoring changes in key visual features in consecutive frame images, including brightness, texture, and color processing parameters of multiple visual feature points, to obtain consecutive frame processing results; Based on the continuous frame processing results, by analyzing the style matching between multiple design elements in the advertisement image, the image is style-consistency detected, the style feature weights of the multiple elements are adjusted, and the consistency calibration result is obtained.

2. The advertising design auxiliary method based on image style transfer according to claim 1, characterized in that: The step of obtaining the edge strength feature is specifically as follows: Based on the input style image sample, the brightness values ​​of multiple pixels in the image are extracted to generate a brightness value matrix; Based on the brightness value matrix, by the formula: Calculate the edge strength around each pixel to obtain the edge strength matrix; Among them, E i,j Represents the edge strength of the pixel located in the i-th row and j-th column, P i,j represents the brightness value of the pixel in the i-th row and j-th column, P i+1,j Indicates that it is located at P i,j The brightness value of the pixel below, P i-1,j Indicates that it is located at P i,j The brightness value of the pixel above, P i,j+1 Indicates that it is located at P i,j The brightness value of the pixel on the right, P i,j-1 Indicates that it is located at P i,j The brightness value of the pixel on the left, i is the row index, j is the column index; Based on the edge intensity matrix, pixel points with edge intensity higher than the average level are identified and extracted to obtain edge intensity features.

3. The advertising design auxiliary method based on image style transfer according to claim 2 is characterized in that: The steps for obtaining the style feature dataset are specifically as follows: Based on the edge intensity feature, the sample image is analyzed to extract the statistical characteristics of the texture and the frequency distribution of the color, and the texture pattern and color distribution data are generated; Based on the texture pattern and color distribution data, by the formula: w k =α·S k +β·F k +γ·C k Calculate feature weight values; Among them, w k is the weight of the kth feature, α is the significance S k The weight adjustment coefficient, S k is the significance score of the kth feature, and β is the frequency F k The weight adjustment coefficient, F k is the frequency of occurrence of the kth feature, γ is the correlation C k The weight adjustment coefficient, C k is the correlation between the kth feature and the target style; Based on the feature weight values, a style feature data set is generated by integrating and recording multiple features of the target sample image.

4. The advertising design auxiliary method based on image style transfer according to claim 1, characterized in that: The steps of acquiring the content features of the target image are specifically as follows: Based on the style feature data set, identifying multiple design elements in the target advertising image that need to be processed, including graphics, text, and layout structure, and generating a list of design elements; With the list of design elements, use the formula: V=A×T v +L×L v +Y×Y v Calculate visual importance scores for multiple design elements; Where V is the visual importance score of the target element, A is the relative area of ​​the element, and T v is the weight coefficient of the relative area of ​​the element, L is the positional advantage of the element, and L v is the weight coefficient of the position, Y is the color contrast of the element, and Y v is the weight coefficient of color contrast; Based on the visual importance scores, the visual influence of multiple design elements is evaluated, and the content features of the target image are extracted by combining the position and type information of the multiple design elements.

5. The advertising design auxiliary method based on image style transfer according to claim 4 is characterized in that: The steps of obtaining the style-transferred image are specifically as follows: Based on the content features of the target image, the formula: CY′ x,y =δ·CY x,y +(1-δ)·CY s,x,y Calculate the adjusted color value; Among them, CY′ x,y is the adjusted color value, δ is the color fusion coefficient, CY x,y is the original color value of the target advertising image at position (x, y), CY s,x,y is the color value at position (x, y) in the style feature data set, where x represents the horizontal coordinate in the image and y represents the vertical coordinate in the image; Based on the adjusted color value, by the formula: TW′ x,y =ε·TW x,y +(1-∈)·TW s,x,y Calculate the adjusted texture value to obtain the color and texture adjustment value; Among them, TW′ x,y is the adjusted texture value, ∈ is the texture fusion coefficient, TW x,y is the original texture value of the target advertising image at position (x, y), TW s,x,y is the texture value at position (x, y) in the style feature data set; Based on the color and texture adjustment values, multiple pixels of the target advertisement image are adjusted to obtain a style-transferred image.

6. The advertising design auxiliary method based on image style transfer according to claim 1, characterized in that: The steps of obtaining the continuous frame processing results are specifically as follows: Based on the style-transferred image, performing recognition and marking of visual feature points, including edges and corners in the image, to generate a visual feature point map; Based on the visual feature point map, analyzing the marked multiple visual feature points, extracting image feature data of the multiple feature points in continuous frames, including color, texture, and brightness, to obtain a feature point data set; Based on the feature point data set, through the formula: Calculate the color, texture and brightness consistency scores of each visual feature point in consecutive frames to obtain the consecutive frame consistency score; Among them, U p is the consistency score of the visual feature point p in multiple frames, YS n,p is the color value of feature point p in the nth frame, WL n,p is the texture value of feature point p in the nth frame, LD n,p is the brightness value of feature point p in the nth frame, is the average color value of feature point p in all frames, is the average texture value of feature point p in all frames, is the average brightness value of feature point p in all frames, N is the total number of frames, n is the frame index, and p is the feature point index; Based on the continuous frame consistency score, the image processing parameters in the continuous frame images are adjusted, including the brightness, texture, and color processing parameters of multiple visual feature points, the visual consistency of the continuous frame images is optimized, and the continuous frame processing results are obtained.

7. The advertising design auxiliary method based on image style transfer according to claim 1, characterized in that: The steps for obtaining the consistency calibration result are specifically as follows: Based on the continuous frame processing results, extracting information of multiple design elements in the processed advertising image, including brightness information, color information, and texture information, and generating design element feature data; Based on the design element characteristic data, through the formula: Calculate the structural similarity index between multiple design elements to obtain the style matching score; Among them, S represents the structural similarity index, m o is the average brightness of element o, m l is the average brightness of element l, σ o is the standard deviation of the brightness of element o, σ l is the standard deviation of the brightness of element l, σ ol is the brightness covariance between elements o and l, k1 and k2 are small constants added to ensure calculation stability; Based on the style matching score, the style consistency of the target advertising image is analyzed, the color, texture and shape of multiple design elements are adjusted, the style coordination of the visual effect is optimized, and the consistency calibration result is obtained.

8. The advertising design auxiliary system based on image style transfer is characterized by: According to any one of claims 1 to 7, the advertising design assistance method based on image style transfer comprises: The sample feature analysis module analyzes the edge strength features of the image based on the input style image sample, combines the color information and texture information, and performs quantitative analysis on the texture pattern and color distribution of the image. It generates a style feature dataset by matching importance weights for multiple features in the image. The design element adjustment module identifies multiple design elements in the target advertisement image based on the style feature data set, including graphics, text, and layout structure, and adjusts the color and texture of the target advertisement image in combination with the style features of the sample image to generate a style-transferred image; The visual feature marking module identifies and marks visual features of the processed image based on the style-transferred image, including edges and corners, and adjusts the processing parameters of the style transfer, including the brightness, texture, and color processing parameters of multiple visual feature points, by monitoring the changes in key visual features in continuous frame images, to generate continuous frame processing results; The design element analysis module analyzes the style matching between multiple design elements in the advertisement image based on the continuous frame processing results through structural similarity evaluation, performs style consistency detection on the image, and generates a style coordination score; The image feature adjustment module adjusts the color, texture and shape of the advertisement image based on the style coordination score, optimizes the coordination of the image style, and generates a consistency calibration result.

Images