An image generation data processing system based on AIGC

Abstract

The present invention relates to the field of data processing, and in particular to an AIGC-based image generation data processing system, comprising: a data acquisition unit; a state analysis unit for determining a conflict coefficient according to a determination condition, and determining a text state according to the number of similar keywords and the conflict coefficient; a removal analysis unit for determining a keyword removal method for each redundant combination according to a set condition; a processing determination unit for determining an input state of a standard input text according to a text fit and a feature dimension, and determining an image processing method according to the input state; a first processing unit for determining an image update method as a full feature update according to a determination condition, or determining a key feature update method according to a text type and a type of text state; and a second processing unit for directly generating an image according to a keyword. The present invention can improve image generation efficiency, and can improve the degree of conformity between a generated image and user expectations.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an AIGC-based image generation data processing system. Background Art

[0002] AIGC is limited by the model processing capabilities when generating images. When there are many image tasks, it often faces the problem of low generation efficiency, which makes it difficult to meet user needs. In addition, vocabulary conflicts in the input text may cause AIGC to deviate from the understanding of instructions, thereby generating images that do not meet expectations. Therefore, how to improve the efficiency of image generation and improve the degree of conformity between generated images and user expectations is a technical problem that needs to be solved urgently by those skilled in the art.

[0003] Chinese patent publication number CN117333414A discloses an AIGC data analysis method for rapid image generation, including: by determining the image generation standard, finding a pre-execution sequence set that matches a variety of image generation requirements, identifying the data to be converted and determining the target execution sequence, establishing a rapid generation channel and training to generate an AIGC-based image generation module, and transmitting the data to be converted to the image generation module for image conversion and feature extraction, and finally determining the image generation result. If the image generation result does not meet the image verification standard, an image compensation process based on the spatial dimension is performed to determine the compensated generated image. It can be seen that the above technical solution has the following problems: the image generation method is single, and it is impossible to adaptively generate images according to the actual state of the text, resulting in poor generation efficiency when there are many image tasks, and the text input situation is not considered, resulting in poor conformity between the generated image and the user's expectations, and unable to meet user needs. Summary of the invention

[0004] To this end, the present invention provides an image generation data processing system based on AIGC to overcome the problems in the prior art: the image generation method is single, and the image cannot be adaptively generated according to the actual state of the text, resulting in poor generation efficiency when there are many image tasks, and the text input situation is not considered, resulting in the generated image being poorly consistent with the user's expectations and unable to meet user needs.

[0005] To achieve the above object, the present invention provides an image generation data processing system based on AIGC, comprising:

[0006] A data collection unit, used for collecting text information;

[0007] A state analysis unit connected to the data collection unit, for determining a conflict coefficient according to a determination condition, and determining a text state according to the number of similar keywords and the conflict coefficient;

[0008] A removal analysis unit, which is connected to the data acquisition unit and the state analysis unit respectively, and is used to determine the redundant combination according to the reference frequency and the distance reference value in the first text state, and determine the keyword removal method of each redundant combination according to the set conditions to remove the redundant keywords according to the keyword relevance or the combination matching coefficient;

[0009] A processing and judging unit connected to the rejection and analysis unit is used to determine the input state of the standard input text according to the text fit and the feature dimension, and determine the image processing method according to the input state, such as determining the image update method according to the judgment condition or directly generating the image according to the keyword;

[0010] A first processing unit, connected to the processing determination unit, is used to determine the image update mode as all feature update according to the determination condition, or to determine the key feature update mode as update according to key product features or update according to product background and product position according to the text type and a type of text status;

[0011] The second processing unit is connected to the processing and determining unit and is used to directly generate an image according to the keyword.

[0012] Furthermore, the state analysis unit determines the text state according to the number of similar keywords and the conflict coefficient, and the text state includes:

[0013] The first text state in which the number of similar keywords is greater than the standard number and the conflict coefficient is greater than or equal to the preset conflict coefficient;

[0014] The second text state in which the number of similar keywords is equal to the standard number or the conflict coefficient is less than the preset conflict coefficient.

[0015] Further, the state analysis unit determines the conflict coefficient according to the determination condition;

[0016] The judgment condition for the state analysis unit response is that the number of similar keywords is greater than the standard number, and the conflict coefficient is determined based on the word meaning conflict degree and the interval reference value;

[0017] The judgment condition for the state analysis unit response is that the number of similar keywords is equal to the standard number, and the conflict coefficient is determined according to the degree of conflict in word meaning.

[0018] Further, the elimination analysis unit responds to the set conditions to determine the keyword elimination method;

[0019] The setting condition for eliminating the response of the analysis unit is that the keyword relevance is greater than or equal to the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the keyword relevance;

[0020] The setting condition for eliminating the response of the analysis unit is that the keyword relevance is less than the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the combination matching coefficient.

[0021] Furthermore, the processing determination unit determines the input state of the standard input text according to the text fit and the feature dimension, and the input state includes:

[0022] A first input state in which the text fit is greater than or equal to a preset text fit and the feature dimension is less than a preset feature dimension;

[0023] A second input state in which the text fit is less than a preset text fit or the feature dimension is greater than or equal to a preset feature dimension;

[0024] The standard input text includes text processed by keyword elimination in a first text state and text in a second text state.

[0025] Further, the processing determination unit responds to the input state to determine the image processing mode;

[0026] The processing and determining unit responds to the first input state, and the image processing mode is to determine the image update mode according to the determination condition;

[0027] The processing and determining unit responds to the second input state, and the image processing method is to directly generate an image according to the keyword.

[0028] Furthermore, the judgment condition responded by the first processing unit is that the feature similarity mean is greater than or equal to the preset feature similarity mean and the image complexity is less than the preset image complexity, and the image update mode is all feature update;

[0029] When the image update mode is to update all features, the image is updated based on any historical generated image in the pre-screened set, and the image features corresponding to the same keywords in the historical standard text corresponding to the historical generated image and the standard input text are similarly replaced;

[0030] The judgment condition of the response of the first processing unit is that the feature similarity mean is less than the preset feature similarity mean or the image complexity is greater than or equal to the preset image complexity, and the image update method is to determine the key feature update method according to the text type and a class of text status;

[0031] The pre-screening set is a set of historical generated images corresponding to historical standard texts whose matching reference values ​​with the standard input text are greater than a preset matching reference value;

[0032] The type of text status includes a first state in which the product area ratio is greater than or equal to the preset product area ratio or the background complexity is less than the preset background complexity, and a second state in which the product area ratio is less than the preset product area ratio and the background complexity is greater than or equal to the preset background complexity.

[0033] Furthermore, the first processing unit updates the key features of the second type of text and the first type of text in the first state according to the key commodity features, wherein:

[0034] The historical generated image with the largest comprehensive evaluation value in the pre-screened set is used as the benchmark image for image update, and the image features corresponding to the keywords in the standard input text whose image influence value is greater than the preset image influence value are recorded as key product features, and the key product features are replaced.

[0035] Furthermore, the first processing unit updates the key features of a type of text in the second state according to the commodity background and commodity location, wherein:

[0036] The historical generated image with the smallest comprehensive evaluation value in the pre-screened set is used as the reference image to update the image, the image features corresponding to each background keyword are updated, and the product moving distance is determined according to the background difference;

[0037] The commodity moving distance is negatively correlated with the background difference.

[0038] Furthermore, the second processing unit directly generates an image according to the keyword, wherein:

[0039] Features are generated according to the keywords corresponding to the standard input text. When generating features for a single keyword, the image features corresponding to any keyword are randomly selected until the image features corresponding to each keyword are selected, and then an image is generated according to the image features corresponding to each keyword.

[0040] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical scheme of the present invention, the conflict coefficient is determined according to the judgment conditions, so that the determination of the conflict coefficient is more in line with the actual text input state, and the text state is determined according to the number of similar keywords and the conflict coefficient, and the degree of semantic conflict of the input text is effectively reflected by the number of similar keywords and the conflict coefficient, and then in the first text state, the keyword removal method is determined according to the set conditions, and the relevance of redundant keywords in the input text is effectively reflected by the set conditions, and then different keyword removal methods are adaptively selected according to the set conditions, so that the selection of the text removal method is more in line with the actual application scenario, avoiding the problem of vocabulary conflict in the input text not being considered in the prior art, resulting in the poor degree of conformity between the generated image and the user's expectations, thereby improving the accuracy and efficiency of the generated image.

[0041] Furthermore, in the present invention, the input state of the standard input text is determined according to the text fit and the feature dimension, and the text fit and the feature dimension are used to effectively reflect the degree of correlation between the text and the historical generated image and the complexity of the image to be generated, and then different image update methods are adaptively selected according to the input state, so that the selection of the image update method is more in line with the actual application scenario, avoiding the problem of a single image generation method in the prior art and the inability to adaptively generate the image according to the actual state of the text, which helps to save time and resources and thus improves the image generation efficiency.

[0042] Furthermore, in the present invention, the image update method is determined according to the judgment conditions, and the judgment conditions effectively reflect the similarity and complexity of each image in the pre-screening set, and then the image update method is adaptively determined according to the judgment conditions, so that the selection of the image update method is more in line with the actual application scenario. At the same time, image generation based on historical generated images can improve image generation efficiency and maintain the stability of the generated images.

[0043] Furthermore, the present invention effectively distinguishes texts by text type and a type of text status, and then adaptively selects different key feature updating methods according to the text type and a type of text status, so that the selection of key feature updating methods is more in line with actual application scenarios, avoiding the problem of a single image generation method in the prior art and the inability to adaptively generate images according to the actual status of the text, and can improve the utilization rate of resources of generated images and reduce the cost of repeatedly generating similar images. At the same time, it also ensures the difference between the generated images and the generated images, thereby improving the image generation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A unit connection diagram of the image generation data processing system based on AIGC of the present invention;

[0045] Figure 2 This is a flow chart of the present invention for determining a keyword elimination method according to set conditions;

[0046] Figure 3 A flow chart of determining an image processing method according to an input state in the present invention;

[0047] Figure 4 The present invention is a flow chart of determining an image update method according to judgment conditions. DETAILED DESCRIPTION

[0048] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0049] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0050] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0051] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0052] See also Figures 1 to 4 As shown, the present invention provides an image generation data processing system based on AIGC, comprising:

[0053] A data collection unit, used for collecting text information;

[0054] A state analysis unit connected to the data collection unit, for determining a conflict coefficient according to a determination condition, and determining a text state according to the number of similar keywords and the conflict coefficient;

[0055] A removal analysis unit, which is connected to the data acquisition unit and the state analysis unit respectively, and is used to determine the redundant combination according to the reference frequency and the distance reference value in the first text state, and determine the keyword removal method of each redundant combination according to the set conditions to remove the redundant keywords according to the keyword relevance or the combination matching coefficient;

[0056] A processing and judging unit connected to the rejection and analysis unit is used to determine the input state of the standard input text according to the text fit and the feature dimension, and determine the image processing method according to the input state, such as determining the image update method according to the judgment condition or directly generating the image according to the keyword;

[0057] A first processing unit, connected to the processing determination unit, is used to determine the image update mode as all feature update according to the determination condition, or to determine the key feature update mode as update according to key product features or update according to product background and product position according to the text type and a type of text status;

[0058] The second processing unit is connected to the processing and determining unit and is used to directly generate an image according to the keyword.

[0059] The application scenario of the present invention is the image generation of AIGC of commodity marketing pictures. The text information includes but is not limited to input text, historical generated images, historical standard text, training images and text descriptions. The input text is a number of characters input by the user to generate the target image. The historical generated image is an image generated according to the historical standard text in the historical image generation process. A single historical standard text is a number of characters corresponding to a single historical generated image in the historical image generation process. The training image is an image used for AIGC model training. A single text description is a number of characters corresponding to a single training image. The present invention sets up a feature library. Several categories are set in the feature library. A single category corresponds to several keywords. A single keyword corresponds to several image features. The categories include but are not limited to color, size, shape, structure, product name and background. Keywords include but are not limited to red, fireworks, round, and pot lid. Image features are vectors corresponding to keywords for generating pictures. This is content that is easy for technicians in this field to understand and will not be elaborated on in detail.

[0060] In the present invention, the type of the generated product marketing image needs to be selected in advance. When the generated product marketing image is a white background image, the type of the corresponding input text is a first-class text. When the generated product marketing image is a scene image, the type of the corresponding input text is a second-class text.

[0061] The confirmation method of the redundant combination is to perform a redundant detection on each keyword in the input text, and when performing a redundant detection on a single keyword, record the keyword as a target keyword, record other keywords excluding the target keyword as reference keywords, record the target keyword and a reference keyword whose reference frequency is less than a preset reference frequency and whose distance reference value is less than a preset distance reference value as a redundant combination, and continue to perform a redundant detection on keywords that have not been subjected to the redundant detection, record the keywords in each redundant combination as redundant keywords, and record the keywords that are not included in the redundant combination as standard keywords;

[0062] The reference frequency is confirmed in the following manner: for a keyword, the keyword is recorded as the target keyword, and other keywords in the input text are recorded as reference keywords. The reference number corresponding to each reference keyword is detected. The reference number is the number of times a single reference keyword and the target keyword appear simultaneously in the text description corresponding to each training image. The minimum value of the reference number corresponding to each reference keyword is recorded as the reference frequency. The distance reference value is the number of characters between a reference keyword and the target keyword.

[0063] The values ​​of the preset reference frequency and the preset distance reference value can be determined by the user according to the actual application scenario. The higher the user's demand for the degree of conformity between the generated image and the user's expectations, the smaller the values ​​of the preset reference frequency and the preset distance reference value. A value of a preset reference frequency and a preset distance reference value is provided, and the average value of the reference frequency corresponding to the historical records that can meet the user's needs is recorded as the preset reference frequency, and the average value of the distance reference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset distance reference value.

[0064] In the present invention, several historical records are correspondingly set up, and any historical record records the reference frequency, distance reference value, conflict coefficient and keyword relevance, etc. in the historical process of at least one image generation, and each historical record corresponds to a qualified mark, which records the efficiency of image generation and whether the degree of conformity between the generated image and the user's expectations meets the user's needs. The qualified mark can be recorded manually.

[0065] Specifically, the state analysis unit determines the text state according to the number of similar keywords and the conflict coefficient, and the text state includes:

[0066] The first text state in which the number of similar keywords is greater than the standard number and the conflict coefficient is greater than or equal to the preset conflict coefficient;

[0067] The second text state in which the number of similar keywords is equal to the standard number or the conflict coefficient is less than the preset conflict coefficient.

[0068] The number of keywords of the same type is confirmed by detecting the category corresponding to each keyword in the user input text, recording the number of keywords corresponding to the same category as the category number, recording the maximum value of the number of each category as the number of keywords of the same type, and the standard number is 1;

[0069] The value of the preset conflict coefficient can be determined by the user according to the actual application scenario. The higher the user's demand for the degree of conformity between the generated image and the user's expectations, the smaller the value of the preset conflict coefficient. A value of the preset conflict coefficient is provided, and the average value of the conflict coefficient corresponding to the historical records that can meet the user's needs is recorded as the preset conflict coefficient.

[0070] Specifically, the state analysis unit determines the conflict coefficient according to the determination condition;

[0071] The judgment condition for the state analysis unit response is that the number of similar keywords is greater than the standard number, and the conflict coefficient is determined based on the word meaning conflict degree and the interval reference value;

[0072] The judgment condition for the state analysis unit response is that the number of similar keywords is equal to the standard number, and the conflict coefficient is determined according to the degree of conflict in word meaning.

[0073] The determination condition includes a first determination condition and a second determination condition, the first determination condition is that the number of similar keywords is greater than the standard number, and the second determination condition is that the number of similar keywords is equal to the standard number;

[0074] Under the first judgment condition, conflict coefficient = word meaning conflict degree - interval reference value; under the second judgment condition, conflict coefficient is positively correlated with word meaning conflict degree; word meaning conflict degree is negatively correlated with frequency reference value, frequency reference value is the minimum value of the minimum frequencies corresponding to each keyword in the input text, the minimum frequency is the minimum value of each reference frequency corresponding to a keyword, the interval reference value is confirmed by recording the category with a number of categories greater than 1 as the category to be detected, recording each keyword in a single category to be detected as a reference keyword, recording the minimum value of the interval distance corresponding to each reference keyword as the interval reference value, and the interval distance is confirmed by detecting the number of characters between the reference keyword and each adjacent reference keyword, and recording the minimum number of characters as the interval distance.

[0075] Specifically, the elimination analysis unit responds to the set conditions to determine the keyword elimination method;

[0076] The setting condition for eliminating the response of the analysis unit is that the keyword relevance is greater than or equal to the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the keyword relevance;

[0077] The setting condition for eliminating the response of the analysis unit is that the keyword relevance is less than the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the combination matching coefficient.

[0078] The setting condition includes a first setting condition and a second setting condition, the first setting condition is that the keyword relevance is greater than or equal to the preset keyword relevance, and the second setting condition is that the keyword relevance is less than the preset keyword relevance;

[0079] Eliminate redundant keywords according to keyword relevance, wherein for a single redundant combination, retain the redundant keyword with the greatest keyword relevance, and delete the other redundant keywords;

[0080] Eliminate redundant keywords according to the combination matching coefficient, wherein, for a single redundant combination, retain a redundant keyword with the largest combination matching coefficient, and delete the other redundant keywords;

[0081] Each keyword in the input text is converted into a text vector. The present invention adopts TF-IDF technology to convert the text vector; each keyword corresponding to a single redundant combination is recorded as a keyword to be analyzed. For a single keyword to be analyzed, the cosine value of the angle between the text vector corresponding to the keyword to be analyzed and the reference vector is calculated. The keyword relevance is the minimum value of the cosine value of the angle corresponding to each keyword to be analyzed in the single redundant combination. The reference vector is the sum of the text vectors corresponding to each standard keyword. It can be understood that the value range of the keyword relevance is [0,1].

[0082] The value of the preset keyword relevance can be determined by the user according to the actual application scenario. The higher the user's demand for the degree of conformity between the generated image and the user's expectations, the greater the value of the preset keyword relevance. A value of the preset keyword relevance is provided, and the average value of the keyword relevance corresponding to the historical records that can meet the user's needs is recorded as the preset keyword relevance.

[0083] The method for confirming the combined matching coefficient is that, for a single keyword to be analyzed, the keyword to be analyzed is recorded as the target keyword to be analyzed, and the standard reference times corresponding to each standard keyword are detected. The standard reference times are the number of times a single standard keyword and the target keyword to be analyzed appear simultaneously in the text description corresponding to each training image. The average value of the standard reference times corresponding to each standard keyword is recorded as the combined matching coefficient.

[0084] Specifically, the processing determination unit determines the input state of the standard input text according to the text fit and the feature dimension, and the input state includes:

[0085] A first input state in which the text fit is greater than or equal to a preset text fit and the feature dimension is less than a preset feature dimension;

[0086] A second input state in which the text fit is less than a preset text fit or the feature dimension is greater than or equal to a preset feature dimension;

[0087] The standard input text includes text processed by keyword elimination in a first text state and text in a second text state.

[0088] Among them, the text fit is the average of the fit reference values ​​of the text descriptions corresponding to each training image. For the text description corresponding to a single training image, the text description is recorded as the target text description, the number of keywords in the target text description is recorded as A, and the number of keywords in the standard input text is recorded as B. The fit reference value of the target text description = the number of identical keywords in the target text description and the standard input text / (the larger value of A and B); the feature dimension is the number of categories corresponding to each keyword in the standard input text.

[0089] The values ​​of the preset text fit and the preset feature dimension can be determined by the user according to the actual application scenario. The higher the user's demand for improving the image generation efficiency, the smaller the value of the preset text fit and the larger the value of the preset feature dimension. The values ​​of the preset text fit and the preset feature dimension are provided, and the image processing method is to detect the historical records of the image update method according to the judgment conditions, and the average value of the text fit corresponding to the historical records that can meet the user's needs is recorded as the preset text fit, and the average value of the feature dimension corresponding to the historical records that can meet the user's needs is recorded as the preset feature dimension.

[0090] Specifically, the processing determination unit responds to the input state to determine the image processing mode;

[0091] The processing and determining unit responds to the first input state, and the image processing mode is to determine the image update mode according to the determination condition;

[0092] The processing and determining unit responds to the second input state, and the image processing method is to directly generate an image according to the keyword.

[0093] Among them, the judgment conditions include a first judgment condition and a second judgment condition. The first judgment condition is that the feature similarity mean is greater than or equal to the preset feature similarity mean and the image complexity is less than the preset image complexity. The second judgment condition is that the feature similarity mean is less than the preset feature similarity mean or the image complexity is greater than or equal to the preset image complexity.

[0094] Specifically, the judgment condition responded by the first processing unit is that the feature similarity mean is greater than or equal to the preset feature similarity mean and the image complexity is less than the preset image complexity, and the image update mode is all feature update;

[0095] When the image update mode is to update all features, the image is updated based on any historical generated image in the pre-screened set, and the image features corresponding to the same keywords in the historical standard text corresponding to the historical generated image and the standard input text are similarly replaced;

[0096] The judgment condition of the response of the first processing unit is that the feature similarity mean is less than the preset feature similarity mean or the image complexity is greater than or equal to the preset image complexity, and the image update method is to determine the key feature update method according to the text type and a class of text status;

[0097] The pre-screening set is a set of historical generated images corresponding to historical standard texts whose matching reference values ​​with the standard input text are greater than a preset matching reference value;

[0098] The type of text status includes a first state in which the product area ratio is greater than or equal to the preset product area ratio or the background complexity is less than the preset background complexity, and a second state in which the product area ratio is less than the preset product area ratio and the background complexity is greater than or equal to the preset background complexity.

[0099] Among them, the feature similarity mean is the average value of the number of identical categories corresponding to each historical generated image in the pre-screening set, and the number of identical categories is the number of identical categories between the historical standard text corresponding to a single historical generated image and the standard input text;

[0100] Image complexity is the average value of the number of mutation pixels corresponding to each historical generated image in the pre-screening set. For a single pixel, if the minimum pixel value difference between the pixel and the pixel adjacent to the pixel is greater than the preset difference, the pixel is a mutation pixel. The value of the preset difference can be set by the user according to the actual situation. The greater the user's demand for the accuracy of confirming image complexity, the smaller the preset difference value is. A preset difference value is provided, and the preset difference is the minimum value of the minimum pixel value difference between each mutation pixel in the historical record and the pixel adjacent to the corresponding mutation pixel.

[0101] The value of the preset fit reference value can be determined by the user according to the actual application scenario. The greater the user's demand for image generation accuracy, the larger the value of the preset fit reference value. A value of the preset fit reference value is provided, and the average value of the fit reference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset fit reference value.

[0102] The values ​​of the preset feature similarity mean and the preset image complexity can be determined by the user according to the actual application scenario. The higher the user's demand for improving the image generation efficiency, the smaller the value of the preset feature similarity mean and the larger the value of the preset image complexity. A value of the preset feature similarity mean and the preset image complexity is provided. The image update method is detected by determining the historical records of the key feature update method according to the text type and a class of text states. The average value of the feature similarity mean corresponding to the historical records that can meet the user's needs is recorded as the preset feature similarity mean, and the average value of the image complexity corresponding to the historical records that can meet the user's needs is recorded as the preset image complexity.

[0103] Product area ratio = area of ​​product area in the reference image / area of ​​the reference image. The present invention identifies the product area and background area in the reference image through an edge detection algorithm. The background complexity is the number of mutation pixels in the background area of ​​the reference image.

[0104] The values ​​of the preset product area ratio and the preset background complexity can be determined by the user according to the actual application scenario. It can be understood that the smaller the value of the preset product area ratio and the larger the value of the preset background complexity, the greater the user's demand for updating the key feature according to the key product features. A value of a preset product area ratio and a preset background complexity is provided, and a historical record of detecting the key feature update method of updating according to the key product features, and the average value of the product area ratio corresponding to the historical records that can meet the user's needs is recorded as the preset product area ratio, and the average value of the background complexity corresponding to the historical records that can meet the user's needs is recorded as the preset background complexity.

[0105] Specifically, the first processing unit updates the key features of the second type of text and the first type of text in the first state according to the key commodity features, wherein:

[0106] The historical generated image with the largest comprehensive evaluation value in the pre-screened set is used as the benchmark image for image update, and the image features corresponding to the keywords in the standard input text whose image influence value is greater than the preset image influence value are recorded as key product features, and the key product features are replaced.

[0107] Among them, comprehensive evaluation value = product area ratio - background complexity. For a single keyword in the standard input text, the keyword is recorded as the target analysis keyword, and other keywords excluding the target analysis keyword are recorded as reference analysis keywords. The image impact value is the cosine value of the angle between the text vector corresponding to the target analysis keyword and the reference analysis vector. The reference analysis vector is the sum of the text vectors corresponding to the reference analysis keywords.

[0108] The value of the preset image influence value can be determined by the user according to the actual application scenario. It can be understood that the greater the user's demand for the novelty of the generated image, the smaller the value of the preset image influence value is. A value of the preset image influence value is provided, and the average value of the image influence values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset image influence value;

[0109] When replacing key product features, the keyword corresponding to the key product feature is recorded as the first feature keyword, and an image feature corresponding to any first feature keyword different from the key product feature is randomly selected to replace the key product feature. After the replacement of each key product feature is completed, an image is generated based on the replaced image feature.

[0110] Specifically, the first processing unit updates the key features of a type of text in the second state according to the commodity background and commodity location, wherein:

[0111] The historical generated image with the smallest comprehensive evaluation value in the pre-screened set is used as the reference image to update the image, the image features corresponding to each background keyword are updated, and the product moving distance is determined according to the background difference;

[0112] The commodity moving distance is negatively correlated with the background difference.

[0113] Among them, the recognition of background keywords can be determined through a deep learning network, and the image features corresponding to each background keyword are updated. When the image features corresponding to a single background keyword are updated, the keyword is recorded as a second feature keyword, and the image features corresponding to the second feature keyword in the benchmark image are recorded as reference background features. An image feature corresponding to any second feature keyword different from the reference background feature is randomly selected to replace the reference background feature, and an image is generated according to the replaced image features; the background area generated after the image features corresponding to each background keyword are updated is recorded as an updated background, and the background area corresponding to the benchmark image is recorded as a benchmark background, and the background difference = the number of mutation pixels in the updated background - the number of mutation pixels in the benchmark background;

[0114] The image generated according to the standard input text is recorded as the generated image. It can be understood that the reference image and the generated image are both a rectangle, with the lower left corner of the rectangle as the origin, and a rectangular coordinate system is established along the length and width passing through the origin as the x-axis and y-axis respectively. The moving distance of the product is the shortest distance from the reference coordinates to the generated coordinates. The reference coordinates are the coordinates of the center of the circumscribed circle of the product area in the reference image, and the generated coordinates are the coordinates of the center of the circumscribed circle of the product area in the generated image.

[0115] Specifically, the second processing unit directly generates an image according to the keyword, wherein:

[0116] Features are generated according to the keywords corresponding to the standard input text. When generating features for a single keyword, the image features corresponding to any keyword are randomly selected until the image features corresponding to each keyword are selected, and then an image is generated according to the image features corresponding to each keyword.

[0117] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An image generation data processing system based on AIGC, characterized in that: include: A data collection unit, used for collecting text information; A state analysis unit connected to the data collection unit, for determining a conflict coefficient according to a determination condition, and determining a text state according to the number of similar keywords and the conflict coefficient; A removal analysis unit, which is connected to the data acquisition unit and the state analysis unit respectively, and is used to determine the redundant combination according to the reference frequency and the distance reference value in the first text state, and determine the keyword removal method of each redundant combination according to the set conditions to remove the redundant keywords according to the keyword relevance or the combination matching coefficient; A processing and judging unit connected to the rejection and analysis unit is used to determine the input state of the standard input text according to the text fit and the feature dimension, and determine the image processing method according to the input state, such as determining the image update method according to the judgment condition or directly generating the image according to the keyword; A first processing unit, connected to the processing determination unit, is used to determine the image update mode as all feature update according to the determination condition, or to determine the key feature update mode as update according to key product features or update according to product background and product position according to the text type and a type of text status; A second processing unit, connected to the processing and determining unit, for directly generating an image according to the keyword; The standard input text includes text processed by keyword elimination in a first text state and text in a second text state; One type of text is the type of input text corresponding to the generated product marketing image as a white background image.

2. The AIGC-based image generation data processing system according to claim 1, characterized in that: The state analysis unit determines the text state according to the number of similar keywords and the conflict coefficient, and the text state includes: The first text state in which the number of similar keywords is greater than the standard number and the conflict coefficient is greater than or equal to the preset conflict coefficient; The second text state in which the number of similar keywords is equal to the standard number or the conflict coefficient is less than the preset conflict coefficient.

3. The image generation data processing system based on AIGC according to claim 2, characterized in that: The state analysis unit determines the conflict coefficient according to the determination condition; The judgment condition for the state analysis unit response is that the number of similar keywords is greater than the standard number, and the conflict coefficient is determined based on the word meaning conflict degree and the interval reference value; The judgment condition for the state analysis unit response is that the number of similar keywords is equal to the standard number, and the conflict coefficient is determined according to the degree of conflict in word meaning.

4. The AIGC-based image generation data processing system according to claim 3, characterized in that: The elimination analysis unit responds to the set conditions to determine the keyword elimination method; The setting condition for eliminating the response of the analysis unit is that the keyword relevance is greater than or equal to the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the keyword relevance; The setting condition for eliminating the response of the analysis unit is that the keyword relevance is less than the preset keyword relevance, and the keyword elimination method is to eliminate redundant keywords according to the combination matching coefficient.

5. The image generation data processing system based on AIGC according to claim 4 is characterized in that: The processing and determining unit determines the input state of the standard input text according to the text fit and the feature dimension, and the input state includes: A first input state in which the text fit is greater than or equal to a preset text fit and the feature dimension is less than a preset feature dimension; A second input state in which the text fit is less than a preset text fit or the feature dimension is greater than or equal to a preset feature dimension.

6. The image generation data processing system based on AIGC according to claim 5, characterized in that: The processing determination unit responds to the input state to determine the image processing mode; The processing and determining unit responds to the first input state, and the image processing mode is to determine the image update mode according to the determination condition; The processing and determining unit responds to the second input state, and the image processing method is to directly generate an image according to the keyword.

7. The image generation data processing system based on AIGC according to claim 6, characterized in that: The judgment condition of the first processing unit response is that the feature similarity mean is greater than or equal to the preset feature similarity mean and the image complexity is less than the preset image complexity, and the image update mode is all feature update; When the image update mode is to update all features, the image is updated based on any historical generated image in the pre-screened set, and the image features corresponding to the same keywords in the historical standard text corresponding to the historical generated image and the standard input text are similarly replaced; The judgment condition of the response of the first processing unit is that the feature similarity mean is less than the preset feature similarity mean or the image complexity is greater than or equal to the preset image complexity, and the image update method is to determine the key feature update method according to the text type and a class of text status; The pre-screening set is a set of historical generated images corresponding to historical standard texts whose matching reference values ​​with the standard input text are greater than a preset matching reference value; The type of text status includes a first state in which the product area ratio is greater than or equal to the preset product area ratio or the background complexity is less than the preset background complexity, and a second state in which the product area ratio is less than the preset product area ratio and the background complexity is greater than or equal to the preset background complexity.

8. The AIGC-based image generation data processing system according to claim 7, characterized in that: The first processing unit updates the key features of the second type of text and the first type of text in the first state according to the key product features, wherein: The historical generated image with the largest comprehensive evaluation value in the pre-screened set is used as the reference image to update the image, and the image features corresponding to the keywords with image influence values ​​greater than the preset image influence values ​​in the standard input text are recorded as key product features, and the key product features are replaced; The second type of text is the type of input text corresponding to when the generated product marketing graph is a scene graph.

9. The AIGC-based image generation data processing system according to claim 8, characterized in that: The first processing unit updates the key features of a type of text in the second state in accordance with the commodity background and commodity location, wherein: The historical generated image with the smallest comprehensive evaluation value in the pre-screened set is used as the reference image to update the image, the image features corresponding to each background keyword are updated, and the product moving distance is determined according to the background difference; The commodity moving distance is negatively correlated with the background difference.

10. The AIGC-based image generation data processing system according to claim 6, characterized in that: The second processing unit directly generates an image according to the keyword, wherein: Features are generated according to the keywords corresponding to the standard input text. When generating features for a single keyword, an image feature corresponding to any keyword is randomly selected until all image features corresponding to the keywords are selected, and then an image is generated according to the image features corresponding to the keywords.

Images