Advertisement background image searching method and device, storage medium and electronic equipment

Abstract

The invention relates to an advertisement background image searching method and device, a storage medium and electronic equipment. The method comprises the following steps: performing visual semantic analysis on a target advertisement map to extract background key elements and theme colors of the target advertisement map; performing semantic similarity matching in an offline template library based on background key elements and theme colors of the target advertisement map to screen out initial candidate results; performing visual matching based on the color distance between the theme color of the initial candidate result and the theme color of the target advertisement picture to determine a final matching result; and when the final matching result cannot be determined, inputting the target advertisement graph into the multi-modal large model, so that the multi-modal large model generates a target background graph according to the background key elements and the image style of the target advertisement graph. The technical problem that an existing advertisement background adaptation method is difficult to consider semantic association and color consistency is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of multimedia intelligent processing, and in particular to an advertisement background image searching method and device, a storage medium and an electronic device. BACKGROUND

[0002] With the growing demand for multi-scene application of digital advertising, the advertising content needs to be adapted to different sizes and proportions of display forms, such as converting a vertical poster into a horizontal video cover or a webpage banner. Existing adaptation methods mainly include manual design, simple algorithm filling and template library matching. Although the manual method can maintain aesthetics, it is low in efficiency and high in cost; the algorithm filling method is prone to produce picture distortion or texture repetition, which destroys the visual focus; and the template library matching relies on limited preset templates and is difficult to cover diversified advertising scenarios. The existing technologies generally have low automation degree, poor visual consistency and high resource consumption, and cannot realize intelligent and personalized generation of advertisement image backgrounds. Specifically, the current methods lack understanding of the semantic content of the advertisement image, resulting in a mismatch between the generated background and the theme; there is a lack of effective color migration mechanism, causing inconsistent picture style; at the same time, the template library update relies on manual operation and is difficult to automatically expand or optimize according to user feedback, limiting the continuous adaptation ability of the system. Therefore, there is an urgent need for an intelligent background image searching method that combines offline template matching and online generation mechanism to realize efficient and automated multi-size advertising adaptation while ensuring consistent style. SUMMARY

[0003] The present application provides an advertisement background image searching method, device, storage medium and electronic device to solve the technical problem that the existing advertisement background adaptation method cannot balance semantic association and color consistency.

[0004] In a first aspect, the present application provides an advertisement background image searching method, comprising: performing visual semantic analysis on a target advertisement image to extract background key elements and a theme color of the target advertisement image; performing semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results; performing visual matching based on the color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine a final matching result; and when the final matching result cannot be determined, inputting the target advertisement image into a multi-modal large model to enable the multi-modal large model to generate a target background image according to the background key elements and the image style of the target advertisement image.

[0005] In a second aspect, the application provides a search device for an advertisement background image, comprising: an extraction module configured to perform visual semantic analysis on a target advertisement image to extract background key elements and a theme color of the target advertisement image; a screening module configured to perform semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results; a matching module configured to perform visual matching based on a color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine final matching results; a first generation module configured to input the target advertisement image to a multi-modal large model when the final matching results cannot be determined, so that the multi-modal large model generates a target background image according to the background key elements and the image style of the target advertisement image.

[0006] As an optional example, the extraction module comprises: a first identification unit configured to input the target advertisement image to a picture-to-text large model so that the picture-to-text large model identifies the background key elements and the image style in the advertisement image; and an analysis unit configured to perform color analysis on the target advertisement image by using a theme color extraction algorithm to obtain the main color distribution of the target advertisement image and determine the color with the highest proportion as the theme color of the target advertisement image.

[0007] As an optional example, the device further comprises: a creation module configured to, before performing semantic similarity matching in the offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results, establish a multi-dimensional keyword database comprising element dimension, scene dimension and special effect dimension, and randomly or regularly combine the multi-dimensional keyword database to generate a keyword set; a second generation module configured to input the keyword set to a text-to-picture large model so that the text-to-picture large model generates an image description text according to the keyword set; a third generation module configured to input the image description text to a picture-to-text large model so that the picture-to-text large model generates an initial background image according to the image description text; a processing module configured to perform multi-group color migration processing on the initial background image to form result images with different color tones, thereby obtaining the offline template library; and a storage module configured to store the theme color of each result image and the corresponding keyword information as index data of the offline template library.

[0008] As an optional example, the screening module comprises: a conversion unit configured to convert the background key elements of the target advertisement image into a text vector representation and convert the keyword information of each background image in the offline template library into a text vector representation; a first calculation unit configured to calculate, according to the text vector representation of the target advertisement image and the text vector representation of each background image in the offline template library, a text similarity between the target advertisement image and each background image in the offline template library by using a semantic matching model; and a first determination unit configured to determine, as initial candidate images, the background images in the offline template library having a text similarity greater than a first threshold value, to obtain the initial candidate result.

[0009] As an optional example, the matching module comprises: a second calculation unit configured to calculate a color distance value between the theme color of the target advertisement image and the theme color of each initial candidate image; a second determination unit configured to determine, as secondary candidate images, the initial candidate images in the initial candidate result having a color distance value less than or equal to a second threshold value, to obtain a secondary candidate result; a third calculation unit configured to perform weighted scoring on the text similarity and the color distance value of each secondary candidate image, to obtain a comprehensive score; and a third determination unit configured to determine, as the final matching result, the secondary candidate image in the secondary candidate result having the highest comprehensive score and greater than a third threshold value.

[0010] As an optional example, the first generation module comprises: a second identification unit configured to, when the final matching result cannot be determined, input the target advertisement image into the multi-modal large model, so that the multi-modal large model identifies the background key elements and the image style of the target advertisement image by using a text-to-image large model; a first generation unit configured to, according to the background key elements and the image style of the target advertisement image, generate, by the multi-modal large model, image generation prompt words describing the background key elements and the image style of the target advertisement image by using a text generation large model; a second generation unit configured to, according to the image generation prompt words, generate, by the multi-modal large model, an initial background image by using a text-to-image large model; and an adjustment unit configured to, by the multi-modal large model, perform color adjustment on the initial background image by using a color transfer algorithm, with the theme color of the target advertisement image as a reference color, to obtain a target background image consistent with the image style and the theme color of the target advertisement image.

[0011] As an optional example, the device further comprises a saving module configured to, when the final matching result cannot be determined, input the target advertisement image into the multi-modal large model, so that the multi-modal large model generates a target background image according to the background key elements and the image style of the target advertisement image, and save the target background image and its corresponding keyword, theme color and generation information into the offline template library.

[0012] Thirdly, this application provides a storage medium storing a computer program, wherein the computer program is executed by a processor to perform the above-mentioned method for searching advertising background images.

[0013] Fourthly, this application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-described method for searching advertising background images through the computer program.

[0014] The technical solutions provided in this application have the following advantages compared with the prior art: This application employs a method that involves visual semantic analysis of the target advertisement image to extract key background elements and theme colors; semantic similarity matching is performed in an offline template library based on the key background elements and theme colors of the target advertisement image to filter initial candidate results; visual matching is performed between the theme colors of the initial candidate results and the color distance of the theme colors of the target advertisement image to determine the final matching result; when the final matching result cannot be determined, the target advertisement image is input into a multimodal large model, so that the multimodal large model generates a target background image based on the key background elements and image style of the target advertisement image. In this method, key background elements and theme color information are extracted by visual semantic analysis of the target advertisement image, and matching is performed in an offline template library based on semantic similarity, combined with visual matching based on the color distance of the theme colors to determine the optimal background image; when matching fails, the multimodal large model automatically generates the target background image based on the key background elements and image style. By using a dual matching mechanism that integrates semantic understanding and visual features, high-precision matching of advertising background images in terms of semantic consistency and visual style is achieved, thereby improving the intelligence and automation level of advertising design, significantly increasing the efficiency of material generation and reducing labor costs. This solves the technical problem that existing advertising background adaptation methods cannot simultaneously take into account semantic association and color consistency. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0018] Figure 1 This is a flowchart of an optional method for searching advertising background images according to an embodiment of this application; Figure 2 This is a flowchart illustrating the specific implementation of an optional method for searching advertising background images according to an embodiment of this application. Figure 3 This is a schematic diagram of the structure of an optional advertising background image search device according to an embodiment of this application; Figure 4 This is a schematic diagram of an optional electronic device according to an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0021] According to a first aspect of the embodiments of this application, a method for searching advertising background images is provided, optionally, such as... Figure 1 As shown, the above method includes: S102, Perform visual semantic analysis on the target ad image to extract key background elements and theme color of the target ad image; S104. Based on the background key elements and theme color of the target ad image, perform semantic similarity matching in the offline template library to filter out the initial candidate results; S106, Visual matching is performed based on the color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine the final matching result; S108, when the final matching result cannot be determined, the target ad image is input into the multimodal large model so that the multimodal large model generates a target background image based on the background key elements and image style of the target ad image.

[0022] Optionally, to address the technical challenge of existing advertising background adaptation methods in simultaneously considering semantic relevance and color consistency, this embodiment employs visual semantic analysis of the target advertising image to extract key background elements and theme color information. Matching is then performed in an offline template library based on semantic similarity, followed by visual matching using theme color distance to determine the optimal background image. When matching fails, a multimodal large model automatically generates the target background image based on key background elements and image style. This dual matching mechanism, combining semantic understanding and visual feature fusion, achieves high-precision matching of advertising background images in terms of semantic consistency and visual style, enhancing the intelligence and automation of advertising design, significantly improving material generation efficiency, and reducing labor costs. Specifically, as shown... Figure 2 The specific implementation flowchart shown below: First, visual semantic analysis is performed on the target advertisement image to extract key background elements and theme color information. This visual semantic analysis involves using a pre-trained image recognition model to identify and extract features from the overall scene, compositional elements, main object outlines, and color relationships of the advertisement image, thereby generating a background feature description that reflects the visual style and semantic intent of the advertisement image.

[0023] Secondly, based on the extracted background key elements and theme color information, semantic similarity matching is performed in a pre-built offline template library. The offline template library includes a large number of classified and labeled background images, each of which stores its semantic tags, style feature vectors, and theme color information. By calculating the semantic feature similarity between the target ad image and each background image in the offline template library, initial candidate results with high semantic relevance to the target ad image are selected.

[0024] Then, visual matching is performed based on the color distance between the theme colors of the initial candidate results and the theme colors of the target advertisement image. By calculating the Euclidean distance or color difference value between the two in the color space, the initial candidate results are further filtered and sorted to determine the final matching result that best matches the visual style consistency.

[0025] When the final result cannot be determined based on semantic matching and color distance matching, the target ad image is input into a multimodal large model. The multimodal large model combines visual and semantic features for cross-modal understanding, and generates a target background image that conforms to the semantics and visual style of the ad content based on the background key elements, composition, and style features of the target ad image, thereby achieving adaptive background generation.

[0026] Optionally, this embodiment significantly improves the search accuracy and matching effect of advertising background images by combining visual semantic parsing, semantic similarity matching, and theme color visual matching. Compared with traditional retrieval methods based on keywords or single image features, this method can more comprehensively understand the semantics and aesthetic intent of advertising images, thereby achieving higher consistency in visual style. When no ideal matching results are found in the offline template library, the system can automatically generate background images that conform to semantic and visual features through a multimodal large model, realizing the fusion of search and generation, improving the intelligence and automation level of advertising material production, significantly reducing design costs, and improving advertising production efficiency.

[0027] As an optional example, visual semantic analysis is performed on the target ad image to extract key background elements and theme colors, including: Input the target ad image into the image-to-text model so that the image-to-text model can identify the key background elements and image style in the ad image; A theme color extraction algorithm is used to perform color analysis on the target ad image to obtain the main color distribution of the target ad image and determine the color with the highest proportion as the theme color of the target ad image.

[0028] Optionally, in this embodiment, key background information and visual features are automatically extracted from the target advertisement image. Specifically, the target advertisement image is input into a graph-to-text model, enabling the model to perform semantic recognition on the image content, extract representative key background elements (such as natural scenes, buildings, people, etc.) and overall image style features from the advertisement image, thereby obtaining the core description of the advertisement's visual semantics. Subsequently, a theme color extraction algorithm is used to perform pixel-level color clustering analysis on the target advertisement image, statistically analyze the distribution of the main colors in the image, and determine the color with the highest proportion as the theme color of the target advertisement image to characterize its overall tone tendency. By combining semantic parsing and color extraction, the semantic level and visual style features of the advertisement image can be accurately captured without manual intervention, providing high-quality feature input for subsequent background matching and generation. This achieves intelligent understanding of the semantic information of the advertisement image and automatic extraction of the theme color, significantly improving the accuracy and visual consistency of background generation and reducing the cost of manual analysis.

[0029] As an optional example, before performing semantic similarity matching in an offline template library based on the background key elements and theme color of the target ad image to filter out initial candidate results, the method also includes: Establish a multidimensional keyword database that includes element, scene, and special effects dimensions, and then randomly or regularly combine the multidimensional keyword database to generate a keyword set; The keyword set is input into the Wenshengwen model so that the Wenshengwen model can generate image description text based on the keyword set. Input the image description text into the text-generated image model so that the text-generated image model can generate an initial background image based on the image description text. Multiple color migration processes are performed on the initial background image to form result images with different hues, resulting in an offline template library; The theme color and corresponding keyword information of each result image are stored as index data for the offline template library.

[0030] Optionally, in this embodiment, an offline template library for generating and retrieving advertising background images is constructed to improve the semantic matching accuracy and visual consistency of advertising images in multi-scene adaptation. Specifically, such as... Figure 2 The specific implementation flowchart shown below: First, a multi-dimensional keyword database is established, encompassing element, scene, and effect dimensions. The element dimension describes the main constituent elements of an image (e.g., snow-capped mountains, grasslands, city buildings), the scene dimension depicts the environment type (e.g., nature, urban, fantasy), and the effect dimension represents the visual style (e.g., oil painting, cartoon, realistic, game style). Keyword sets are generated through random or rule-based combinations to enrich the semantic space of the template generation.

[0031] Secondly, the target advertisement image is input into a large-scale image-to-text model (such as GPT-4O) to extract the main background elements and image style information. Then, the extracted elements and style are used as input conditions and input into the large-scale image-to-text model to generate high-quality image description text for the image-to-text model. The generation rules include: all input elements must appear without introducing additional content; the image style must be a minimalist expression of the input style; and the semantic description must be as detailed as possible. Subsequently, this image description text is input into the large-scale image-to-text model to generate an initial background image, ensuring that the generated background is consistent with the original advertisement image in terms of semantics and style.

[0032] Next, an intelligent color transfer operation is performed to improve the visual matching of the background. Specifically, a theme color extraction algorithm is used to obtain the main color distribution of the advertisement image, and the color with the highest proportion is selected as the target reference color. Based on LAB color space transformation, the color mean and standard deviation of the initial background image are normalized and adjusted so that the generated background is closer to the target reference color in terms of hue, resulting in a visually consistent image.

[0033] Finally, multiple sets of color-transfer processed images and their corresponding keyword information are stored in an offline template library to form structured index data.

[0034] Optionally, this embodiment achieves a fully automated process for generating advertising background images, from semantic generation to color optimization, by automatically building an offline template library. This enables the generation of high-quality background materials with a unified style, relevant theme, and visual harmony. It significantly improves the automation level of cross-size and cross-scene adaptation of advertising images, reduces manual design costs, and ensures consistency between the overall style of the advertisement and the main visual.

[0035] As an optional example, based on the background key elements and theme color of the target ad image, semantic similarity matching is performed in an offline template library to filter out initial candidate results, including: Convert the background key elements of the target ad image into text vector representations, and convert the keyword information of each background image in the offline template library into text vector representations; Based on the text vector representation of the target ad image and the text vector representation of each background image in the offline template library, the text similarity between the target ad image and each background image in the offline template library is calculated using a semantic matching model. Background images with text similarity greater than the first threshold in the offline template library are identified as initial candidate images, thus obtaining initial candidate results.

[0036] Optionally, in this embodiment, the semantic similarity-based advertising background image matching is used to filter candidate backgrounds from the offline template library that have the highest semantic relevance to the target advertising image. Specifically: First, the key background elements of the target ad image are converted into text vector representations to quantify the semantic features of the ad image. At the same time, the keyword information of each background image in the offline template library is uniformly converted into text vector representations to construct a comparable semantic space.

[0037] Subsequently, semantic matching models (such as BERT or CLIP, cross-modal semantic matching models) are used to calculate the semantic similarity score between the target ad image vector and each background image vector in the template library. Based on the calculation results, background images with similarity scores higher than a first threshold are identified as initial candidate images, thus forming an initial candidate result set. This process effectively filters out background materials that are highly relevant to the target ad image in terms of content elements, scene semantics, and style description.

[0038] Optionally, this embodiment achieves semantic-level accurate matching between the advertisement image and the background image in the template library, significantly improving the efficiency of automated screening and semantic consistency, and providing a high-quality candidate basis for the subsequent visual matching and generation stages.

[0039] As an optional example, visual matching is performed based on the color distance between the theme color of the initial candidate results and the theme color of the target ad image to determine the final matching result, including: Calculate the color distance between the theme color of the target ad image and the theme color of each initial candidate image; The initial candidate images with color distance values ​​less than or equal to the second threshold in the initial candidate results are identified as secondary candidate images, thus obtaining secondary candidate results; The text similarity and color distance values ​​of each secondary candidate image are weighted and scored to obtain a comprehensive score; The secondary candidate image with the highest overall score that is greater than the third threshold is selected as the final matching result.

[0040] Optionally, to further determine the final background image that best matches the target advertisement image in terms of color and visual style among the candidate background images after initial semantic screening, this embodiment employs an advertisement background visual matching method based on theme color distance. Specifically, firstly, the color distance value between the theme color of the target advertisement image and the theme color of each initial candidate image is calculated. By representing the main colors of both in the same color space (such as CIEDE2000 or LAB space), their color difference distance is calculated, thereby obtaining the degree of difference between the two images in overall visual tone.

[0041] Secondly, based on the calculation results, candidate images with color distance values ​​less than or equal to the second threshold in the initial candidate results are identified as secondary candidate images, thus obtaining the secondary candidate results. This step effectively eliminates candidate images with excessively large color deviations, ensuring that the images entering the next stage maintain coordination with the target advertising image in terms of dominant color tendency.

[0042] Next, a weighted score is applied to the textual semantic similarity and color distance value of each secondary candidate image. Semantic similarity reflects the relevance of image content, while color distance reflects visual harmony. By weighted fusion of these two metrics, a comprehensive score is obtained to achieve a balance between semantic consistency and visual unity.

[0043] Finally, the candidate image with the highest comprehensive score among the secondary candidate results and greater than the third threshold is determined as the final matching result, thereby outputting the background image that best matches the theme semantics and visual style of the target advertisement image.

[0044] Optionally, this embodiment introduces a dual semantic and color matching mechanism to achieve a step-by-step filtering process from semantic association to visual harmony. This significantly improves the accuracy and aesthetics of matching advertising background images, ensuring that the generated results maintain high visual quality while maintaining thematic consistency and color harmony. It also reduces manual intervention and automates and increases the efficiency of intelligent adaptation of advertising backgrounds.

[0045] As an optional example, when the final matching result cannot be determined, the target ad image is input into a multimodal large model, which generates a target background image based on the background key elements and image style of the target ad image, including: When the final matching result cannot be determined, the target ad image is input into the multimodal big model so that the multimodal big model can use the graph-generated text big model to identify the background key elements and image style of the target ad image; Based on the background key elements and image style of the target ad image, the multimodal big model uses the text-to-text big model to generate image-generated prompts describing the background key elements and image style of the target ad image; Based on the image, prompt words are generated. The multimodal large model utilizes the text-generated image large model to generate the initial background image. The multimodal large model uses a color transfer algorithm to adjust the color of the initial background image, using the theme color of the target advertisement image as a reference color, to obtain a target background image that matches the image style and theme color of the target advertisement image.

[0046] Optionally, in this embodiment, an intelligent advertising background generation method based on a multimodal large model is provided. This method automatically generates a background image that is semantically consistent and stylistically unified with the target advertising image when a suitable background cannot be obtained through an offline template library. Specifically, firstly, when the final matching result cannot be determined through the offline template library, the target advertising image is input into the multimodal large model. The multimodal large model first calls the image-to-text large model to perform semantic recognition on the target advertising image, extracting key background elements (such as snow-capped mountains, beaches, cities, grasslands, etc.) and image styles (such as realistic, cartoon, oil painting, sketch, game style, etc.) to form a visual semantic description of the advertising image.

[0047] Secondly, based on the identified key background elements and image style information, the multimodal big data model calls the text-to-text big data model to generate image generation prompts. These prompts are used to accurately describe the background semantic features and visual style of the target advertisement image. The generation rules include: ensuring that the input elements appear completely, not introducing irrelevant content, maintaining a consistent style, and presenting a minimalist visual expression overall, while enriching the details of the image through detailed language descriptions.

[0048] Then, the multimodal large model inputs the generated image prompts into the text-to-image large model to generate the initial background image. This step achieves the mapping from text semantics to visual image, ensuring that the generated background maintains structural and stylistically consistency with the advertisement image.

[0049] Finally, the multimodal large model, based on a color transfer algorithm, uses the theme color of the target ad image as a reference to perform color mapping and adjustment on the initial background image. Specifically, the reference color is converted into a LAB space and matched with the color distribution of the initial background image using the mean and standard deviation, thereby generating a target background image that is consistent with the theme color and overall style of the target ad image.

[0050] Optionally, this embodiment implements an adaptive background generation mechanism in scenarios where template matching fails. It can automatically understand the semantics and style characteristics of the ad image, generating background content that highly matches it, significantly improving the automated production capability and visual consistency of ad creatives, reducing manual design intervention, and enhancing generation efficiency and application flexibility.

[0051] As an alternative example, when the final matching result cannot be determined, the method further includes feeding the target ad image into a multimodal large model so that the multimodal large model generates a target background image based on the background key elements and image style of the target ad image. Save the target background image, its corresponding keywords, theme colors, and generated metadata to the offline template library.

[0052] Optionally, this embodiment provides a closed-loop optimization method for advertising background images, used to automatically generate and save high-quality background images when a suitable background cannot be obtained through template matching. Specifically, when the target advertising image cannot be matched to the final result, it is input into a multimodal large model, which generates a target background image based on the background key elements and image style of the target advertising image. Subsequently, the generated target background image, along with its corresponding keyword information, theme color, and generation metadata, is saved to an offline template library for subsequent search and matching. Through this strategy, the offline template library can continuously expand with new background materials, achieving dynamic updates and self-optimization of the template library. This not only ensures a high degree of consistency in semantics and visual style of advertising backgrounds but also continuously enriches the template library through an automatic database entry mechanism, improving the efficiency and accuracy of advertising background image matching and generation, reducing manual intervention costs, and realizing intelligent and automated management and production of advertising materials.

[0053] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0054] According to another aspect of the embodiments of this application, a search device for advertising background images is also provided, such as... Figure 3 As shown, it includes: The extraction module 302 is used to perform visual semantic analysis on the target advertisement image to extract the background key elements and theme color of the target advertisement image; The filtering module 304 is used to perform semantic similarity matching in the offline template library based on the background key elements and theme color of the target advertisement image in order to filter out the initial candidate results. Matching module 306 is used to perform visual matching based on the color distance between the theme color of the initial candidate result and the theme color of the target advertisement image to determine the final matching result; The first generation module 308 is used to input the target ad image into the multimodal large model when the final matching result cannot be determined, so that the multimodal large model can generate a target background image based on the background key elements and image style of the target ad image.

[0055] It should be noted that the extraction module 302 in this embodiment can be used to execute step S102 in this application embodiment, the filtering module 304 in this embodiment can be used to execute step S104 in this application embodiment, the matching module 306 in this embodiment can be used to execute step S106 in this application embodiment, and the first generation module 308 in this embodiment can be used to execute step S108 in this application embodiment.

[0056] As an optional example, the extraction module includes: The first recognition unit is used to input the target advertisement image into the image-to-text model so that the image-to-text model can recognize the key background elements and image style in the advertisement image. The analysis unit is used to perform color analysis on the target ad image using a theme color extraction algorithm to obtain the main color distribution of the target ad image and determine the color with the highest proportion as the theme color of the target ad image.

[0057] As an optional example, the above-described apparatus further includes: Before the initial candidate results are filtered by semantic similarity matching in an offline template library based on the background key elements and theme colors of the target ad image, a multi-dimensional keyword database containing element, scene, and special effects dimensions is created. The multi-dimensional keyword database is then randomly or regularly combined to generate a keyword set. The second generation module is used to input the keyword set into the Wenshengwenda model so that the Wenshengwenda model can generate image description text based on the keyword set. The third generation module is used to input the image description text into the image-generated image model so that the image-generated image model can generate an initial background image based on the image description text. The processing module is used to perform multiple color migration processes on the initial background image to form result images with different hues, thus obtaining an offline template library; The storage module is used to store the theme color and corresponding keyword information of each result image as index data for the offline template library.

[0058] As an optional example, the filtering module includes: The conversion unit is used to convert the background key elements of the target ad image into text vector representations, and to convert the keyword information of each background image in the offline template library into text vector representations; The first calculation unit is used to calculate the text similarity between the target advertisement image and each background image in the offline template library using a semantic matching model, based on the text vector representation of the target advertisement image and the text vector representation of each background image in the offline template library. The first determining unit is used to determine the background images in the offline template library whose text similarity is greater than a first threshold as initial candidate images, and to obtain the initial candidate results.

[0059] As an optional example, the matching module includes: The second calculation unit is used to calculate the color distance value between the theme color of the target advertisement image and the theme color of each initial candidate image; The second determining unit is used to determine the initial candidate images whose color distance values ​​are less than or equal to the second threshold in the initial candidate results as secondary candidate images, and obtain secondary candidate results; The third calculation unit is used to weight and score the text similarity and color distance values ​​of each secondary candidate image to obtain a comprehensive score; The third determining unit is used to determine the second-level candidate image with the highest comprehensive score among the second-level candidate results and which is greater than the third threshold as the final matching result.

[0060] As an optional example, the first generation module includes: The second recognition unit is used to input the target advertisement image into the multimodal big model when the final matching result cannot be determined, so that the multimodal big model can use the image-to-text big model to identify the background key elements and image style of the target advertisement image. The first generation unit is used to generate image-generated prompts describing the background key elements and image style of the target advertisement image using a multimodal big model and a text-to-text big model. The second generation unit is used to generate prompt words based on the image. The multimodal large model uses the text-generated image large model to generate the initial background image. The adjustment unit is used in multimodal large models to adjust the color of the initial background image using the theme color of the target advertisement image as a reference color through a color transfer algorithm, so as to obtain a target background image that is consistent with the image style and theme color of the target advertisement image.

[0061] As an optional example, the above-described apparatus further includes: The save module is used to input the target ad image into the multimodal big model when the final matching result cannot be determined. After the multimodal big model generates the target background image based on the background key elements and image style of the target ad image, it saves the target background image and its corresponding keywords, theme colors and generated meta information to the offline template library.

[0062] For other examples of this embodiment, please refer to the examples above, which will not be repeated here.

[0063] Figure 4 This is a schematic diagram of an optional electronic device according to an embodiment of this application, such as... Figure 4 As shown, it includes a processor 402, a communication interface 404, a memory 406, and a communication bus 408. The processor 402, communication interface 404, and memory 406 communicate with each other via the communication bus 408. Memory 406 is used to store computer programs; When processor 402 executes a computer program stored in memory 406, it performs the following steps: Perform visual semantic analysis on the target ad image to extract key background elements and theme colors; Based on the background key elements and theme color of the target ad image, semantic similarity matching is performed in the offline template library to filter out initial candidate results; Visual matching is performed based on the color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine the final matching result; When the final matching result cannot be determined, the target ad image is input into the multimodal large model so that the multimodal large model can generate a target background image based on the background key elements and image style of the target ad image.

[0064] Optionally, in this embodiment, the communication bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The symbol is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0065] The memory may include RAM, or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0066] As an example, the memory 406 described above may include, but is not limited to, the extraction module 302, filtering module 304, matching module 306, and first generation module 308 from the search device for the aforementioned advertising background image. Furthermore, it may include, but is not limited to, other module units from the search device for the aforementioned advertising background image, which will not be elaborated upon in this example.

[0067] The processor mentioned above can be a general-purpose processor, including but not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0068] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0069] Those skilled in the art will understand that Figure 4 The structure shown is for illustrative purposes only. The device that implements the above-described method for searching advertising background images can be a terminal device, such as a smartphone (e.g., an Android phone, an iOS phone), a tablet computer, a PDA, a mobile internet device (MID), a PAD, or other terminal devices. Figure 4 This does not limit the structure of the aforementioned electronic devices. For example, the electronic device may also include components that are more... Figure 4 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 4 The different configurations shown.

[0070] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, ROM, RAM, disk or optical disk, etc.

[0071] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, which, when executed by a processor, performs the steps in the above-described method for searching advertising background images.

[0072] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0073] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0074] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0075] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0076] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0078] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0079] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An advertisement background image search method, characterized by, The method comprises the following steps: performing visual semantic analysis on a target advertisement image to extract background key elements and a theme color of the target advertisement image; performing semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results; performing visual matching based on the color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine final matching results; when the final matching results cannot be determined, inputting the target advertisement image into a multi-modal large model to enable the multi-modal large model to generate a target background image according to the background key elements and the image style of the target advertisement image.

2. The method of claim 1, wherein, The method of performing visual semantic analysis on a target advertisement image to extract background key elements and a theme color of the target advertisement image comprises the following steps: inputting the target advertisement image into a picture-to-text large model to enable the picture-to-text large model to identify background key elements and an image style in the advertisement image; performing color analysis on the target advertisement image by using a theme color extraction algorithm to obtain a main color distribution of the target advertisement image and determine the color with the highest proportion as the theme color of the target advertisement image.

3. The method of claim 1, wherein, Before performing semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results, the method further comprises the following steps: establishing a multi-dimensional keyword database containing element dimension, scene dimension and special effect dimension, and randomly or regularly combining the multi-dimensional keyword database to generate a keyword set; inputting the keyword set into a text-to-picture large model to enable the text-to-picture large model to generate an image description text according to the keyword set; inputting the image description text into a picture-to-text large model to enable the picture-to-text large model to generate an initial background image according to the image description text; performing multi-group color migration processing on the initial background image to form result images with different color tones, thereby obtaining the offline template library; storing the theme color of each result image and the corresponding keyword information as index data of the offline template library.

4. The method of claim 1, wherein, The method of performing semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen out initial candidate results comprises the following steps: converting the background key elements of the target advertisement image into a text vector representation, and converting the keyword information of each background image in the offline template library into a text vector representation; calculating the text similarity between the target advertisement image and each background image in the offline template library by using a semantic matching model according to the text vector representation of the target advertisement image and the text vector representation of each background image in the offline template library; determining the background images with a text similarity greater than a first threshold value in the offline template library as initial candidate images, thereby obtaining the initial candidate results.

5. The method of claim 4, wherein, The method of performing visual matching based on the color distance between the theme color of the initial candidate results and the theme color of the target advertisement image to determine final matching results comprises the following steps: calculating the color distance value between the theme color of the target advertisement image and the theme color of each initial candidate image; determine an initial primary candidate image with a color distance value less than or equal to a second threshold value in the initial candidate result as a secondary candidate image, to obtain a secondary candidate result; weighting the text similarity and the color distance value of each secondary candidate image to obtain a comprehensive score; determine a secondary candidate image with the highest comprehensive score and greater than a third threshold value in the secondary candidate result as the final matching result.

6. The method of claim 1, wherein, when the final matching result cannot be determined, input the target advertisement image into a multi-modal large model, so that the multi-modal large model generates a target background image according to the background key elements and image style of the target advertisement image, including: when the final matching result cannot be determined, input the target advertisement image into the multi-modal large model, so that the multi-modal large model identifies the background key elements and image style of the target advertisement image by using a picture-to-text large model; according to the background key elements and image style of the target advertisement image, the multi-modal large model generates image generation prompt words describing the background key elements and image style of the target advertisement image by using a text-to-image large model; according to the image generation prompt words, the multi-modal large model generates an initial background image by using a text-to-image large model; the multi-modal large model performs color adjustment on the initial background image by taking the theme color of the target advertisement image as a reference color through a color migration algorithm, to obtain a target background image consistent with the image style and theme color of the target advertisement image.

7. The method of claim 1, wherein, after the target advertisement image is input into the multi-modal large model when the final matching result cannot be determined, so that the multi-modal large model generates a target background image according to the background key elements and image style of the target advertisement image, the method further includes: save the target background image, its corresponding keywords, theme color and generation information to the offline template library.

8. An advertisement background image search device characterized by comprising: including: an extraction module for performing visual semantic analysis on a target advertisement image to extract background key elements and a theme color of the target advertisement image; a screening module for performing semantic similarity matching in an offline template library based on the background key elements and the theme color of the target advertisement image to screen an initial candidate result; a matching module for performing visual matching based on the color distance between the theme color of the initial candidate result and the theme color of the target advertisement image to determine a final matching result; a first generation module for inputting the target advertisement image into a multi-modal large model when the final matching result cannot be determined, so that the multi-modal large model generates a target background image according to the background key elements and image style of the target advertisement image.

9. A computer-readable storage medium storing a computer program, the computer-readable storage medium being characterized by, The computer program is run by the processor to execute the method described in any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 7 by the computer program. The memory stores a computer program, and the processor is configured to execute the method described in any one of claims 1 to 7 by the computer program.

Images