A digital art design auxiliary method and system based on artificial intelligence

Abstract

The present invention discloses a digital art design assistance method and system based on artificial intelligence, and the method includes data collection and processing, preliminary sketch generation, color matching assistance, adaptive style fusion and digital art design assistance. The present invention relates to the field of digital art design technology, and specifically refers to a digital art design assistance method and system based on artificial intelligence. The present scheme adopts an art design assistance comprehensive method combining the three functions of sketch design, coloring design and style design, and through functional classification modular design, it can ensure the separate use of functions and integrated design, thereby improving the overall consistency and maintainability of the method; adopts a parallel improved generative adversarial network method to perform preliminary sketch generation; adopts an improved generative adversarial network method combined with color source enhancement to perform color matching assistance; adopts a multi-path generative adversarial network method combined with style extraction enhancement to perform style design enhancement.

Description

Technical Field

[0001] The present invention relates to the technical field of digital art design, and in particular to an artificial intelligence-based digital art design auxiliary method and system. Background Art

[0002] The AI-based digital art design assistance method is a tool that uses AI technology, especially deep learning, computer vision, and generative adversarial networks, to support the art creation and design process. It can help designers achieve functions such as automated and personalized design generation, style transfer, and image optimization during the creative process. By learning from a large amount of art data, AI can play an important role in providing creative inspiration, improving design efficiency, and achieving precise design tuning, thereby making art design more efficient, innovative, and diversified.

[0003] However, in the existing digital art design auxiliary methods, there is a technical problem that the existing digital art design ideas often adopt a single-stage design, directly generating output results according to specific needs, but this process cannot effectively provide substantial help for the whole process of art design; in the existing sketch design methods, there is a technical problem that sketch generation, as the underlying technology of digital art design, is prone to disordered content or poor reference, which leads to poor sketch usability and cannot provide strong support for digital art design; in the existing coloring design methods, there is a technical problem that the coloring of sketches is not only related to the artistic laws of color matching itself, but also involves the overall image. Style control, but the existing technology has a low degree of granularity for coloring tasks, which makes it difficult to meet the process from sketch to colored image, and also leads to the technical problem that the coloring style is messy and difficult to control, and cannot reflect the intelligence of digital art; in the existing style design methods, there is a technical problem that the learning and transfer of image style is a relatively complex task for automated methods, and the traditional standard adversarial generative network is difficult to meet the precise production of multiple styles, and thus cannot provide effective support for the stylized design of digital art design. At the same time, stylized design is closely related to the content of digital art images, and traditional methods are difficult to meet the technical problems of clarity, detail and style fusion requirements of artistic style generation. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digital art design assistance method and system based on artificial intelligence. In view of the technical problem that in the existing digital art design assistance methods, the existing digital art design ideas often adopt a single-stage design, and directly generate output results according to specific needs, but this process cannot effectively provide substantial help for the whole process of art design, this scheme creatively adopts an art design assistance comprehensive method that combines the three functions of sketch design, coloring design and style design, and through functional classification and modular design, it can ensure the separate use of functions and integrated design, thereby improving the overall consistency and maintainability of the method; in view of the technical problem that in the existing sketch design methods, sketch generation, as the underlying technology of digital art design, is prone to disordered content or poor reference, which leads to poor sketch usability and cannot provide strong support for digital art design, this scheme creatively adopts a parallel improved generative adversarial network method to perform preliminary sketch generation. By improving the loss function and optimization algorithm, combined with the intervention of three-dimensional manual evaluation when constructing the original data, the content generation of the sketch design is fixed within the scope of basic innovation, practicality and popular aesthetics, thereby improving the accuracy and usability of the sketch design. In the existing coloring design methods, the coloring of the sketch is not only related to the artistic laws of color matching itself, but also involves the control of the overall style of the image. The existing technology has a low degree of granularity for the coloring task, which makes it difficult to meet the process from sketch to colored image. At the same time, it also leads to the technical problem that the coloring style is messy and difficult to control, and cannot reflect the intelligence of digital art. This scheme creatively adopts an improved generative adversarial network method combined with color source enhancement to assist in color matching. By optimizing the coloring style loss and combining the color transmission of principal component analysis after image space transformation, the overall fit of the coloring design to human aesthetics and color perception is optimized, the design effect of color matching and style is improved, and a good intermediate result is also provided for subsequent stylized design.In the existing style design methods, learning and transferring image styles is a relatively complex task for automated methods. Traditional standard adversarial generative networks are difficult to meet the precise production of multiple styles, and thus cannot provide effective support for the stylized design of digital art design. At the same time, stylized design is closely related to the content of digital art images. Traditional methods are difficult to meet the technical problems of clarity, detail and style fusion requirements of artistic style generation. This solution creatively adopts a multi-path generative adversarial network method combined with style extraction enhancement to enhance style design. Through the multi-path generator network, it can optimize the correspondence between style category annotations and specific style images, and also optimize the complex design of multi-style fusion. By constructing an adaptive style fusion network, the adaptive fusion of style features is performed from the two perspectives of contour and coloring, realizing the multi-style fusion production in digital art design, and providing strong support for the overall performance of digital art design assistance, and providing a practical reference for the field of intelligent digital art design. ;

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based digital art design auxiliary method, which includes the following steps:

[0006] Step S1: data collection and processing;

[0007] Step S2: preliminary generation of sketch;

[0008] Step S3: color matching assistance;

[0009] Step S4: Adaptive style fusion;

[0010] Step S5: Digital art design assistance.

[0011] Further, in step S1, the data collection process is used to collect the original data set required for digital art design assistance, and perform preliminary enhancement and optimization processing, specifically, by collecting images, obtaining the original data set for digital art design assistance, and by basic enhancement processing and annotation, obtaining the optimized digital art design data set;

[0012] The image collection specifically refers to performing image classification and collection operations;

[0013] The image classification collection specifically refers to the collection of concept sketch data, sketch coloring data, and stylized art image data;

[0014] The digital art design auxiliary original data set specifically includes sketch data, coloring data and stylized data;

[0015] The basic enhancement processing and marking steps include:

[0016] Step S11: image operation, specifically performing random rotation, scaling and cropping operations on the optimized digital art design data set to obtain an image operation enhanced data set;

[0017] Step S12: Attribute optimization, specifically adjusting the hue, brightness and contrast of the image operation enhancement data to obtain an attribute enhancement variant data set;

[0018] Step S13: noise blurring, specifically adding random noise and blurring to the attribute enhanced variant data set to obtain a noise blurring enhanced variant data set;

[0019] Step S14: style mixing, specifically, performing standard generative adversarial network model training based on the noise blur enhancement variant dataset, performing stylized image generation, and obtaining a style expansion variant dataset;

[0020] Step S15: annotation enhancement, specifically, performing element labeling operation through the image operation, the attribute optimization, the noise blurring and the style mixing to obtain an optimized digital art design data set;

[0021] The element labeling operation specifically refers to image content labeling, color position labeling, artistic style labeling and manual evaluation labeling; the manual evaluation labeling specifically refers to three-dimensional quantitative evaluation, including the artistic innovation dimension, practical value dimension and aesthetic popularity dimension;

[0022] Further, in step S2, the sketch is initially generated to generate a digital art sketch according to the demand, specifically, based on the optimized digital art design data set, a parallel improved generative adversarial network method is used to perform preliminary sketch generation to obtain digital art sketch data;

[0023] The parallel improved generative adversarial network specifically includes loss function improvement and optimization function improvement;

[0024] The method of using a parallel improved generative adversarial network to perform preliminary sketch generation and obtain digital art sketch data specifically includes the following steps:

[0025] Step S21: constructing a basic generative adversarial network, specifically using a pre-trained convolutional neural network as a feature extraction network, and constructing a basic generative adversarial network including a generator and a discriminator;

[0026] Step S22: construct an improved loss function for integrating gradient penalty and generating loss function to optimize model training effect;

[0027] Step S23: constructing an improved optimization function, specifically executing a standard genetic algorithm to optimize model parameters, and optimizing and improving the fitness function of the standard genetic algorithm in combination with image annotation, the calculation formula is:

[0028] ;

[0029] Where f(·) is the improved fitness function, w C is the weight of artistic innovation dimension, Score C (·) is the artistic innovation dimension evaluation value, which is obtained through manual evaluation in the image element labeling operation stage. U is the weight of the practical value dimension, Score U (·) is the evaluation value of the practical value dimension, w A is the weight of the aesthetic popularity dimension, Score A (·) is the evaluation value of the aesthetic popularity dimension;

[0030] Step S24: training the sketch preliminary generation model, specifically, training the sketch preliminary generation model by constructing the basic generative adversarial network, constructing the improved loss function, and constructing the improved optimization function, to obtain the sketch preliminary generation model Model SKE ;

[0031] Step S25: Preliminary generation of a sketch, specifically using the sketch to preliminarily generate a model Model SKE , based on the optimized digital art design data set, preliminary sketch generation is performed to obtain digital art sketch data.

[0032] Further, in step S3, the color matching assistance is used to assist in coloring the digital art sketch, specifically, based on the optimized digital art design data set and the digital art sketch data, an improved generative adversarial network method combined with color source enhancement is used to perform color matching assistance to obtain digital art coloring data, specifically including the following steps:

[0033] Step S31: improving the generative adversarial network initialization, specifically executing the operations of step S21 to step S23, performing improved generative adversarial network initialization, which is used as a model for color matching assistance, and extracting a color matching feature map from the optimized digital art design data set and the digital art sketch data using a pre-trained convolutional neural network through the improved generative adversarial network initialization, and calculating the Gram matrix corresponding to the feature map, which is used as an auxiliary matrix for color source enhancement;

[0034] In the color matching auxiliary model training, the digital art sketch data is specifically used as the original content image, and the sketch coloring data in the optimized digital art design data set is used as the coloring style target image;

[0035] Step S32: constructing a coloring style loss, specifically integrating a color content loss function and an improved coloring style loss function, and constructing a coloring style loss. The calculation formula is:

[0036] ;

[0037] Where, L total is the coloring style loss function, is the color content loss weight, L con is the color content loss function, and specifically adopts the standard content loss function as the color content loss function. is the improved coloring style loss weight, L csys is the improved coloring style loss function, L is the total number of feature map layers, l is the feature map layer index, N l is the total number of feature maps of the l-th layer, M l is the feature map size of the feature map of layer l, W l is the feature map weight of the feature map of layer l, i is the feature map index, j is the feature map position index, is the Gram matrix of the position of the jth feature map corresponding to the i-th feature map of the l-th layer feature map, is the feature map matrix of the coloring style target image corresponding to the jth feature map position of the i-th feature map of the l-th layer feature map;

[0038] Step S33: constructing a color source enhancement analysis operator, specifically, constructing the color source enhancement analysis operator by sequentially performing color space conversion, principal component analysis dimensionality reduction, and style color transmission to obtain color matching mapping data;

[0039] The color space conversion specifically refers to converting the color of the original image from the RGB space to the CIELAB color space to obtain color space converted image data;

[0040] The principal component analysis dimensionality reduction specifically performs principal component analysis dimensionality reduction operation on the color space conversion image data, and performs style color transmission by calculating the color matching between the generated image and the colored style target image;

[0041] Step S34: coloring style comprehensive mixing, specifically, using alpha blending to perform coloring style comprehensive mixing to obtain a coloring style matching image output, and by obtaining the coloring style matching image output, optimizing the color matching auxiliary model to generate an output result of digital art coloring;

[0042] Step S35: Color matching auxiliary model training, specifically, by initializing the improved generative adversarial network, constructing the coloring style loss, constructing the color source enhancement analysis operator and the coloring style comprehensive mixing, the color matching auxiliary model training is performed to obtain the color matching auxiliary model Model COL ;

[0043] Step S36: color matching assistance, specifically, using the color matching assistance model Model according to the optimized digital art design data set and the digital art sketch data COL , assist in color matching and obtain digital art coloring data.

[0044] Further, in step S4, the adaptive style fusion is used to perform digital art stylized design in combination with the color image, specifically, based on the optimized digital art design data set and the digital art coloring data, a multi-path generative adversarial network method combined with style extraction enhancement is used to perform adaptive style fusion to obtain stylized digital art image data;

[0045] The multi-path generative adversarial network combined with style extraction enhancement specifically includes a multi-path generator network, a style extraction enhancement network, an adaptive style fusion network and a global discriminator network;

[0046] The step of adopting the improved generative adversarial network method combined with style extraction enhancement to perform adaptive style fusion to obtain stylized digital art image data includes:

[0047] Step S41: constructing a multi-path generator network, specifically constructing a multi-path generator network including a global space path, a category space path and a category context path, a feature fusion layer and a conditional normalization layer;

[0048] The global space path is used to accept a complete image as input;

[0049] The category space path is used to accept the annotated data of the style type as input;

[0050] The category context path is used to process context information of the style type;

[0051] The feature fusion layer is used to fuse the outputs of the global space path, the category space path and the category context path;

[0052] The conditional normalization layer is used to take the annotated data of the style type as conditional input, generate normalization parameters, and optimize the content feature map generation;

[0053] Step S42: constructing a style extraction enhancement network, specifically applying a pre-trained convolutional neural network to extract style features, and obtaining style feature vector data through channel attention weighting;

[0054] Step S43: constructing an adaptive style fusion network, specifically, performing contour style feature transformation and color style feature transformation in sequence according to the style feature vector data to obtain style change feature vector data, and performing feature fusion on the style change feature vector data and the style feature vector data through a feature fusion module to obtain adaptive style fusion feature data, and generating a style image according to the adaptive style fusion feature data;

[0055] Step S44: constructing an adaptive loss function, specifically using an integrated loss function of multi-scale adversarial loss, perceptual loss and feature matching loss as an adaptive loss function for optimizing adaptive style fusion model training;

[0056] Step S45: training the adaptive style fusion model, specifically, training the adaptive style fusion model by constructing the multi-path generator network, constructing the style extraction enhancement network, constructing the adaptive style fusion network and constructing the adaptive loss function, to obtain the adaptive style fusion model Model STY ;

[0057] Step S46: Adaptive style fusion, specifically using the adaptive style fusion model Model STY , performing adaptive style fusion based on the optimized digital art design data set and the digital art coloring data to obtain stylized digital art image data.

[0058] Further, in step S5, the digital art design assistance is used to combine sketch generation, color matching and style fusion to perform comprehensive digital art design assistance, specifically, by initially generating the sketch, generating digital art sketch data, and performing sketch design assistance based on the digital art sketch data, by the color matching assistance, obtaining digital art coloring data, and performing color matching design assistance based on the digital art coloring data, by the adaptive style fusion, obtaining stylized digital art image data, and performing stylized design assistance based on the stylized digital art image data, and performing comprehensive digital art design assistance through the image design assistance, the color matching design assistance and the stylized design assistance to obtain comprehensive reference data for digital art design assistance;

[0059] The digital art design auxiliary reference data specifically includes output image data, output sketch parameters, output coloring parameters and output art style parameters.

[0060] The present invention provides an artificial intelligence-based digital art design auxiliary system, which includes a data collection and processing module, a sketch preliminary generation module, a color matching auxiliary module, an adaptive style fusion module and a digital art design auxiliary module;

[0061] The data collection and processing module is used for data collection and processing, and obtains an optimized digital art design data set through data collection and processing, and sends the optimized digital art design data set to the sketch preliminary generation module;

[0062] The sketch preliminary generation module is used for preliminary sketch generation, obtains digital art sketch data through preliminary sketch generation, and sends the digital art sketch data to the color matching auxiliary module and the digital art design auxiliary module;

[0063] The color matching auxiliary module is used for color matching assistance, obtains digital art coloring data through color matching assistance, and sends the digital art coloring data to the adaptive style fusion module and the digital art design auxiliary module;

[0064] The adaptive style fusion module is used for adaptive style fusion, obtains stylized digital art image data through adaptive style fusion, and sends the stylized digital art image data to the digital art design auxiliary module;

[0065] The digital art design assistance module is used for digital art design assistance, and obtains digital art design assistance reference data through digital art design assistance.

[0066] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0067] (1) In view of the technical problem that the existing digital art design auxiliary methods often adopt a single-stage design approach, directly generating output results based on specific needs, but this process cannot effectively provide substantial assistance for the entire process of art design, this solution creatively adopts an art design auxiliary comprehensive method that combines the three functions of sketch design, coloring design and style design, and through functional classification and modular design, it can ensure the separate use of functions and integrated design, thereby improving the overall coherence and maintainability of the method;

[0068] (2) In view of the technical problem that sketch generation, as the underlying technology of digital art design, is prone to disordered content or poor reference in existing sketch design methods, which leads to poor usability of sketches and cannot provide strong support for digital art design, this solution creatively adopts a parallel improved generative adversarial network method to perform preliminary sketch generation. By improving the loss function and optimization algorithm, combined with the intervention of three-dimensional manual evaluation when constructing the original data, the content generation of sketch design is fixed within the range of basic innovation, practicality and popular aesthetics, thereby improving the accuracy and usability of sketch design.

[0069] (3) In the existing coloring design methods, the coloring of sketches is not only related to the artistic rules of color matching itself, but also involves the control of the overall style of the image. The existing technology has a low degree of granularity for coloring tasks, which makes it difficult to meet the process from sketches to colored images. It also leads to the technical problem that the coloring style is messy and difficult to control, and cannot reflect the intelligence of digital art. This solution creatively adopts an improved generative adversarial network method combined with color source enhancement to assist in color matching. By optimizing the coloring style loss and combining the color transmission of principal component analysis after image space transformation, the overall fit of the coloring design to human aesthetics and color perception is optimized, the design effect of color matching and style is improved, and a good intermediate result is provided for subsequent stylized design.

[0070] (4) In the existing style design methods, learning and transferring image styles is a relatively complex task for automated methods. Traditional standard adversarial generative networks are difficult to meet the precise production of multiple styles, and thus cannot provide effective support for the stylized design of digital art design. At the same time, stylized design is closely related to the content of digital art images. Traditional methods are difficult to meet the technical problems of clarity, detail and style fusion requirements of artistic style generation. This scheme creatively adopts a multi-path generative adversarial network method combined with style extraction enhancement to enhance style design. Through the multi-path generator network, it can optimize the correspondence between style category annotations and specific style images, as well as the complex design of multi-style fusion. By constructing an adaptive style fusion network, style features are adaptively fused from the two perspectives of contour and coloring, realizing the multi-style fusion production in digital art design, and providing strong support for the overall performance of digital art design assistance, and providing a practical reference for the field of intelligent digital art design. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 A flowchart of a digital art design auxiliary method based on artificial intelligence provided by the present invention;

[0072] Figure 2A schematic diagram of a digital art design assistance system based on artificial intelligence provided by the present invention;

[0073] Figure 3 A schematic diagram of the process flow diagram initially generated for the step S2 sketch;

[0074] Figure 4 This is a schematic diagram of the process of color matching assistance in step S3;

[0075] Figure 5 Schematic diagram of the process of adaptive style fusion in step S4.

[0076] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0077] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0078] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0079] Example 1, see Figure 1 The present invention provides an artificial intelligence-based digital art design auxiliary method, which comprises the following steps:

[0080] Step S1: data collection and processing;

[0081] Step S2: preliminary generation of sketch;

[0082] Step S3: color matching assistance;

[0083] Step S4: Adaptive style fusion;

[0084] Step S5: Digital art design assistance.

[0085] By performing the above operations, in view of the technical problem that in the existing digital art design auxiliary methods, the existing digital art design ideas often adopt single-stage design, directly generating output results according to specific needs, but this process cannot effectively provide substantial help for the whole process of art design, this scheme creatively adopts an art design auxiliary comprehensive method combining the three functions of sketch design, coloring design and style design, and through functional classification modular design, it can ensure the separate use of functions and integrated design, thereby improving the overall consistency and maintainability of the method.

[0086] Example 2, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. In step S1, the data collection process is used to collect the original data set required for digital art design assistance, and perform preliminary enhancement and optimization processing, specifically, by collecting images, obtaining the original data set for digital art design assistance, and by basic enhancement processing and annotation, obtaining the optimized digital art design data set;

[0087] The image collection specifically refers to performing image classification and collection operations;

[0088] The image classification collection specifically refers to the collection of concept sketch data, sketch coloring data, and stylized art image data;

[0089] The digital art design auxiliary original data set specifically includes sketch data, coloring data and stylized data;

[0090] The basic enhancement processing and marking steps include:

[0091] Step S11: image operation, specifically performing random rotation, scaling and cropping operations on the optimized digital art design data set to obtain an image operation enhanced data set;

[0092] Step S12: Attribute optimization, specifically adjusting the hue, brightness and contrast of the image operation enhancement data to obtain an attribute enhancement variant data set;

[0093] Step S13: noise blurring, specifically adding random noise and blurring to the attribute enhanced variant data set to obtain a noise blurring enhanced variant data set;

[0094] Step S14: style mixing, specifically, performing standard generative adversarial network model training based on the noise blur enhancement variant dataset, performing stylized image generation, and obtaining a style expansion variant dataset;

[0095] Step S15: annotation enhancement, specifically, performing element labeling operation through the image operation, the attribute optimization, the noise blurring and the style mixing to obtain an optimized digital art design data set;

[0096] The element labeling operation specifically refers to image content labeling, color position labeling, artistic style labeling, and manual evaluation labeling;

[0097] Table 1 is a schematic table of the content of the optimized digital art design dataset. As shown in the table, the optimized digital art design dataset specifically includes concept sketch data, sketch coloring data, stylized art image data, content annotation data, background element annotation data, composition annotation data, style mixing annotation data and manual evaluation annotation data;

[0098] Table 1 Schematic table of the content of the optimized digital art design dataset

[0099]

[0100] The key elements specifically include character annotation, object annotation, background element annotation, decorative item annotation and environment annotation;

[0101] The color position specifically includes fill color annotation, edge color annotation, gradient color annotation, shadow color annotation and highlight color annotation;

[0102] The style types specifically include impressionist style annotation, abstract style annotation, surrealist style annotation, stereoscopic style annotation and minimalist style annotation;

[0103] The detail elements specifically include stroke annotation, light and shadow annotation, texture annotation, reflection annotation and granularity annotation;

[0104] The background elements specifically include sky annotation, natural landscape annotation, urban landscape annotation, indoor landscape annotation and man-made object annotation;

[0105] The composition elements specifically include symmetry annotation, focus annotation, perspective annotation, scale annotation and blank space annotation;

[0106] The fusion ratio refers to the fusion ratio between different style types, which is automatically calculated by the standard generative adversarial network model and has a value range of [0,100%];

[0107] The three-dimensional quantitative evaluation specifically includes the dimensions of artistic innovation, practical value and aesthetic popularity, which are obtained by converting manual scores into percentage values, and the value range is [0,100%].

[0108] Example 3, see Figure 1 , Figure 2and Figure 3 , this embodiment is based on the above embodiment, in step S2, the sketch is initially generated, and is used to generate a digital art sketch according to demand, specifically, based on the optimized digital art design data set, a parallel improved generative adversarial network method is used to perform preliminary sketch generation to obtain digital art sketch data;

[0109] The parallel improved generative adversarial network specifically includes loss function improvement and optimization function improvement;

[0110] The method of using a parallel improved generative adversarial network to perform preliminary sketch generation and obtain digital art sketch data specifically includes the following steps:

[0111] Step S21: constructing a basic generative adversarial network, specifically using a pre-trained convolutional neural network as a feature extraction network, and constructing a basic generative adversarial network including a generator and a discriminator;

[0112] The pre-trained convolutional neural network specifically adopts a pre-trained VGG19 network;

[0113] The objective function calculation formula of the basic generative adversarial network is:

[0114] ;

[0115] Where G is the generator identifier, D is the discriminator identifier, V(·) is the objective function, min is the minimization function, max is the maximization function, x is the real data, which is used to represent the specific sketch image in the optimized digital art design dataset, z is the input noise of the generator, which is used to represent the generated fake sketch data, is the expected log probability that the discriminator D judges the real data x to be true, is the expected log-probability that the discriminator judges the fake sketch data generated by the generator G as fake;

[0116] Step S22: Construct an improved loss function to integrate gradient penalty and generate loss function to optimize model training effect. The calculation formula is:

[0117] ;

[0118] Where, L WGP is the improved loss function, x is the real data, z is the input noise of the generator, is the expected loss of the discriminator D for judging the real data x as true, is the loss expectation of the discriminator judging the fake sketch data generated by the generator G as fake, is the gradient penalty hyperparameter, is the interpolated sampled data sampled from the real data x and the generated fake sketch data, the interpolated sampled data The calculation formula is: ,in, is a random interpolation parameter with a value range of (0,1), G(·) is the generator function, is the discriminator D for interpolated sampled data The gradient of , specifically through the Lipschitz gradient penalty, is used to constrain the discriminator gradient. The discriminator D that complies with the Lipschitz gradient penalty judges the interpolation sampling data Loss expectation for real;

[0119] Step S23: constructing an improved optimization function, specifically executing a standard genetic algorithm to optimize model parameters, and optimizing and improving the fitness function of the standard genetic algorithm in combination with image annotation, the calculation formula is:

[0120] ;

[0121] Where f(·) is the improved fitness function, w C is the weight of artistic innovation dimension, Score C (·) is the artistic innovation dimension evaluation value, which is obtained through manual evaluation in the image element labeling operation stage, and w U is the weight of the practical value dimension, Score U (·) is the evaluation value of the practical value dimension, w A is the weight of the aesthetic popularity dimension, Score A (·) is the evaluation value of the aesthetic popularity dimension;

[0122] The artistic innovation dimension weight w C The specific value is 0.55, and the practical value dimension weight w U The specific value of is 0.25, and the weight of the aesthetic popularity dimension is w A The specific value of is 0.2;

[0123] Step S24: training the sketch preliminary generation model, specifically, training the sketch preliminary generation model by constructing the basic generative adversarial network, constructing the improved loss function, and constructing the improved optimization function, to obtain the sketch preliminary generation model Model SKE ;

[0124] Step S25: Preliminary generation of a sketch, specifically using the sketch to preliminarily generate a model Model SKE , based on the optimized digital art design data set, preliminary sketch generation is performed to obtain digital art sketch data.

[0125] By performing the above operations, in view of the technical problem that in the existing sketch design methods, sketch generation, as the underlying technology of digital art design, is prone to disordered content or poor reference, which leads to poor usability of sketches and cannot provide strong support for digital art design, this scheme creatively adopts a parallel improved generative adversarial network method to perform preliminary sketch generation. By improving the loss function and optimization algorithm in two directions, combined with the intervention of three-dimensional manual evaluation when constructing the original data, the content generation of the sketch design is fixed within the range of basic innovation, practicality and popular aesthetics, thereby improving the accuracy and usability of the sketch design.

[0126] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the color matching assistance is used to assist in coloring the digital art sketch. Specifically, based on the optimized digital art design data set and the digital art sketch data, an improved generative adversarial network method combined with color source enhancement is used to perform color matching assistance to obtain digital art coloring data. Specifically, the following steps are included:

[0127] Step S31: improving the generative adversarial network initialization, specifically executing the operations of step S21 to step S23, performing improved generative adversarial network initialization, which is used as a model for color matching assistance, and extracting a color matching feature map from the optimized digital art design data set and the digital art sketch data using a pre-trained convolutional neural network through the improved generative adversarial network initialization, and calculating the Gram matrix corresponding to the feature map, which is used as an auxiliary matrix for color source enhancement;

[0128] In the color matching auxiliary model training, the digital art sketch data is specifically used as the original content image, and the sketch coloring data in the optimized digital art design data set is used as the coloring style target image;

[0129] Step S32: constructing a coloring style loss, specifically integrating a color content loss function and an improved coloring style loss function, and constructing a coloring style loss. The calculation formula is:

[0130] ;

[0131] Where, L total is the coloring style loss function, is the color content loss weight, L con is the color content loss function, and specifically adopts the standard content loss function as the color content loss function. is the improved coloring style loss weight, L csysis the improved coloring style loss function, L is the total number of feature map layers, l is the feature map layer index, N l is the total number of feature maps of the l-th layer, M l is the feature map size of the feature map of layer l, W l is the feature map weight of the feature map of layer l, i is the feature map index, j is the feature map position index, is the Gram matrix of the position of the jth feature map corresponding to the i-th feature map of the l-th layer feature map, is the feature map matrix of the coloring style target image corresponding to the jth feature map position of the i-th feature map of the l-th layer feature map;

[0132] The color content loss weight The specific value of is 0.2, and the improved coloring style loss weight The specific value of is 0.8;

[0133] Step S33: constructing a color source enhancement analysis operator, specifically, constructing the color source enhancement analysis operator by sequentially performing color space conversion, principal component analysis dimensionality reduction, and style color transmission to obtain color matching mapping data;

[0134] The color space conversion specifically refers to converting the color of the original image from the RGB space to the CIELAB color space to obtain color space converted image data;

[0135] The principal component analysis dimensionality reduction specifically performs principal component analysis dimensionality reduction operation on the color space conversion image data, and performs style color transmission by calculating the color matching between the generated image and the coloring style target image. The calculation formula is:

[0136] ;

[0137] Where F(·) is the cumulative distribution function, L T is the brightness dimension vector of the coloring style target image, A T is the red and green dimension vector of the coloring style target image, B T is the blue-yellow dimension vector of the coloring style target image, q T is the principal component vector of the coloring style target image, L S is the brightness dimension vector of the generated image, A S is the red and green dimension vector of the generated image, B S is the blue-yellow dimension vector of the generated image, q S is the principal component vector of the generated image;

[0138] Step S34: coloring style comprehensive mixing, specifically, using alpha blending to perform coloring style comprehensive mixing to obtain a coloring style matching image output, and by obtaining the coloring style matching image output, optimizing the color matching auxiliary model to generate an output result of digital art coloring;

[0139] Step S35: Color matching auxiliary model training, specifically, by initializing the improved generative adversarial network, constructing the coloring style loss, constructing the color source enhancement analysis operator and the coloring style comprehensive mixing, the color matching auxiliary model training is performed to obtain the color matching auxiliary model Model COL ;

[0140] Step S36: color matching assistance, specifically, using the color matching assistance model Model according to the optimized digital art design data set and the digital art sketch data COL , assist in color matching and obtain digital art coloring data.

[0141] By performing the above operations, in the existing coloring design methods, the coloring of sketches is not only related to the artistic laws of color matching itself, but also involves the control of the overall style of the image. The existing technology has a low degree of granularity for coloring tasks, which makes it difficult to meet the process from sketches to colored images. It also leads to the technical problem that the coloring style is messy and difficult to control, and the intelligence of digital art cannot be reflected. This solution creatively adopts an improved generative adversarial network method combined with color source enhancement to assist in color matching. By optimizing the coloring style loss and combining the color transmission of principal component analysis after image space transformation, the overall fit of the coloring design to human aesthetics and color perception is optimized, the design effect of color matching and style is improved, and a good intermediate result is provided for subsequent stylized design.

[0142] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the adaptive style fusion is used to perform digital art stylized design in combination with color images. Specifically, based on the optimized digital art design data set and the digital art coloring data, a multi-path generative adversarial network method combined with style extraction enhancement is used to perform adaptive style fusion to obtain stylized digital art image data;

[0143] The multi-path generative adversarial network combined with style extraction enhancement specifically includes a multi-path generator network, a style extraction enhancement network, an adaptive style fusion network and a global discriminator network;

[0144] The step of adopting the improved generative adversarial network method combined with style extraction enhancement to perform adaptive style fusion to obtain stylized digital art image data includes:

[0145] Step S41: constructing a multi-path generator network, specifically constructing a multi-path generator network including a global space path, a category space path and a category context path, a feature fusion layer and a conditional normalization layer;

[0146] The global space path is used to accept the complete image as input, and specifically includes a three-layer convolutional network, each layer of the network includes a convolution layer, a normalization layer and an activation function layer;

[0147] The category space path is used to accept the style type annotation data as input, and specifically includes a three-layer convolutional network, each layer of the network includes a convolution layer, a normalization layer and an activation function layer;

[0148] The category context path is used to process context information of the style type, specifically including a lightweight ResNet subnet, a global average pooling layer, and an attention extraction layer;

[0149] The feature fusion layer is used to fuse the outputs of the global space path, the category space path and the category context path;

[0150] The conditional normalization layer is used to take the annotated data of the style type as conditional input, generate normalization parameters, and optimize the content feature map generation;

[0151] Step S42: constructing a style extraction enhancement network, specifically applying a pre-trained convolutional neural network to extract style features, and obtaining style feature vector data through channel attention weighting;

[0152] Step S43: constructing an adaptive style fusion network, specifically, performing contour style feature transformation and color style feature transformation in sequence according to the style feature vector data to obtain style change feature vector data, and performing feature fusion on the style change feature vector data and the style feature vector data through a feature fusion module to obtain adaptive style fusion feature data, and generating a style image according to the adaptive style fusion feature data;

[0153] The calculation formulas for the outline style feature transformation and the color style feature transformation are:

[0154] ;

[0155] In the formula, is the contour style feature transformation feature data, E c is the adaptive style orthogonal matrix, D cis the adaptive style diagonal matrix, f c is the style feature vector data, where the adaptive style orthogonal matrix E c and the adaptive style diagonal matrix D c satisfy , is the style change feature vector data;

[0156] Step S44: constructing an adaptive loss function, specifically using an integrated loss function of multi-scale adversarial loss, perceptual loss and feature matching loss as an adaptive loss function for optimizing adaptive style fusion model training;

[0157] Step S45: training the adaptive style fusion model, specifically, training the adaptive style fusion model by constructing the multi-path generator network, constructing the style extraction enhancement network, constructing the adaptive style fusion network and constructing the adaptive loss function, to obtain the adaptive style fusion model Model STY ;

[0158] Step S46: Adaptive style fusion, specifically using the adaptive style fusion model Model STY , based on the optimized digital art design data set and the digital art coloring data, adaptive style fusion is performed to obtain stylized digital art image data.

[0159] By performing the above operations, in view of the fact that in the existing style design methods, the learning and transfer of image styles is a relatively complex task for automated methods, the traditional standard adversarial generative network is difficult to meet the precise production of multiple styles, and thus cannot provide effective support for the stylized design of digital art design. At the same time, stylized design is closely related to the content of digital art images, and traditional methods are difficult to meet the technical problems of clarity, detail and style fusion requirements of artistic style generation. This solution creatively adopts a multi-path generative adversarial network method combined with style extraction enhancement to enhance style design. Through the multi-path generator network, it can optimize the correspondence between style category annotations and specific style images, and also optimize the complex design of multi-style fusion. By constructing an adaptive style fusion network, the adaptive fusion of style features is performed from the two perspectives of contour and coloring, realizing the multi-style fusion production in digital art design, and also providing strong support for the overall performance of digital art design assistance, and providing a practical reference for the field of intelligent digital art design.

[0160] Example 6, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S5, the digital art design assistance is used to combine sketch generation, color matching and style fusion to perform comprehensive digital art design assistance. Specifically, digital art sketch data is generated through the preliminary sketch generation, and sketch design assistance is performed based on the digital art sketch data. Digital art coloring data is obtained through the color matching assistance, and color matching design assistance is performed based on the digital art coloring data. Stylized digital art image data is obtained through the adaptive style fusion, and stylized design assistance is performed based on the stylized digital art image data. Comprehensive digital art design assistance is performed through the image design assistance, the color matching design assistance and the stylized design assistance to obtain comprehensive reference data for digital art design assistance.

[0161] The digital art design auxiliary reference data specifically includes output image data, output sketch parameters, output coloring parameters and output art style parameters.

[0162] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an artificial intelligence-based digital art design auxiliary system, including a data collection and processing module, a sketch preliminary generation module, a color matching auxiliary module, an adaptive style fusion module and a digital art design auxiliary module;

[0163] The data collection and processing module is used for data collection and processing, and obtains an optimized digital art design data set through data collection and processing, and sends the optimized digital art design data set to the sketch preliminary generation module;

[0164] The sketch preliminary generation module is used for preliminary sketch generation, obtains digital art sketch data through preliminary sketch generation, and sends the digital art sketch data to the color matching auxiliary module and the digital art design auxiliary module;

[0165] The color matching auxiliary module is used for color matching assistance, obtains digital art coloring data through color matching assistance, and sends the digital art coloring data to the adaptive style fusion module and the digital art design auxiliary module;

[0166] The adaptive style fusion module is used for adaptive style fusion, obtains stylized digital art image data through adaptive style fusion, and sends the stylized digital art image data to the digital art design auxiliary module;

[0167] The digital art design assistance module is used for digital art design assistance, and obtains digital art design assistance reference data through digital art design assistance.

[0168] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0169] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0170] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A digital art design auxiliary method based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: data collection and processing, obtaining a digital art design auxiliary original data set through image collection, and obtaining an optimized digital art design data set through basic enhancement processing and annotation; Step S2: preliminary sketch generation, used to generate digital art sketches according to requirements, specifically based on the optimized digital art design data set, using a parallel improved generative adversarial network method to perform preliminary sketch generation to obtain digital art sketch data, specifically including the following steps: step S21: constructing a basic generative adversarial network; step S22: constructing an improved loss function; step S23: constructing an improved optimization function; step S24: preliminary sketch generation model training; step S25: preliminary sketch generation; Step S3: color matching assistance, used to assist in coloring the digital art sketch, specifically, based on the optimized digital art design data set and the digital art sketch data, an improved generative adversarial network method combined with color source enhancement is used to perform color matching assistance to obtain digital art coloring data, specifically including the following steps: step S31: improved generative adversarial network initialization; step S32: constructing a coloring style loss; step S33: constructing a color source enhancement analysis operator; step S34: comprehensive mixing of coloring styles; step S35: color matching assistance model training; step S36: color matching assistance; Step S4: Adaptive style fusion, used to combine color images for digital art stylized design, specifically based on the optimized digital art design data set and the digital art coloring data, using a multi-path generative adversarial network method combined with style extraction enhancement to perform adaptive style fusion to obtain stylized digital art image data, specifically including the following steps: Step S41: constructing a multi-path generator network; Step S42: constructing a style extraction enhancement network; Step S43: constructing an adaptive style fusion network; Step S44: constructing an adaptive loss function; Step S45: adaptive style fusion model training; Step S46: adaptive style fusion; Step S5: Digital art design assistance, performing comprehensive digital art design assistance to obtain digital art design assistance comprehensive reference data.

2. The artificial intelligence-based digital art design auxiliary method according to claim 1, characterized in that: In step S1, the data collection process is used to collect the original data set required for digital art design assistance, and perform preliminary enhancement and optimization processing, specifically, by collecting images, obtaining the original data set for digital art design assistance, and by basic enhancement processing and annotation, obtaining the optimized digital art design data set; The image collection specifically refers to performing image classification and collection operations; The image classification collection specifically refers to the collection of concept sketch data, sketch coloring data, and stylized art image data; The digital art design auxiliary original data set specifically includes sketch data, coloring data and stylized data; The basic enhancement processing and marking steps include: Step S11: image operation, specifically performing random rotation, scaling and cropping operations on the optimized digital art design data set to obtain an image operation enhanced data set; Step S12: Attribute optimization, specifically adjusting the hue, brightness and contrast of the image operation enhancement data to obtain an attribute enhancement variant data set; Step S13: noise blurring, specifically adding random noise and blurring to the attribute enhanced variant data set to obtain a noise blurring enhanced variant data set; Step S14: style mixing, specifically, performing standard generative adversarial network model training based on the noise blur enhancement variant dataset, performing stylized image generation, and obtaining a style expansion variant dataset; Step S15: annotation enhancement, specifically, performing element labeling operation through the image operation, the attribute optimization, the noise blurring and the style mixing to obtain an optimized digital art design data set; The element labeling operation specifically refers to image content labeling, color position labeling, artistic style labeling, and manual evaluation labeling; The manual evaluation and annotation specifically refers to three-dimensional quantitative evaluation, including the dimensions of artistic innovation, practical value and aesthetic popularity.

3. The digital art design auxiliary method based on artificial intelligence according to claim 2, characterized in that: In step S2, the parallel improved generative adversarial network method is used to perform preliminary sketch generation to obtain digital art sketch data, which specifically includes the following steps: Step S21: constructing a basic generative adversarial network, specifically using a pre-trained convolutional neural network as a feature extraction network, and constructing a basic generative adversarial network including a generator and a discriminator; Step S22: construct an improved loss function for integrating gradient penalty and generating loss function to optimize model training effect; Step S23: constructing an improved optimization function, specifically executing a standard genetic algorithm to optimize model parameters, and optimizing and improving the fitness function of the standard genetic algorithm in combination with image annotation, the calculation formula is: ； Where f(·) is the improved fitness function, w C is the weight of artistic innovation dimension, Score C (·) is the artistic innovation dimension evaluation value, which is obtained through manual evaluation in the image element labeling operation stage, and w U is the weight of the practical value dimension, Score U (·) is the evaluation value of the practical value dimension, w A is the weight of the aesthetic popularity dimension, Score A (·) is the evaluation value of the aesthetic popularity dimension; Step S24: training the sketch preliminary generation model, specifically, training the sketch preliminary generation model by constructing the basic generative adversarial network, constructing the improved loss function, and constructing the improved optimization function, to obtain the sketch preliminary generation model Model SKE ; Step S25: Preliminary generation of a sketch, specifically using the sketch to preliminarily generate a model Model SKE , based on the optimized digital art design data set, preliminary sketch generation is performed to obtain digital art sketch data.

4. The artificial intelligence-based digital art design auxiliary method according to claim 3, characterized in that: In step S3, the improved generative adversarial network method combined with color source enhancement is used to assist in color matching to obtain digital art coloring data, which specifically includes the following steps: Step S31: improving the generative adversarial network initialization, specifically executing the operations of step S21 to step S23, performing improved generative adversarial network initialization, which is used as a model for color matching assistance, and extracting a color matching feature map from the optimized digital art design data set and the digital art sketch data using a pre-trained convolutional neural network through the improved generative adversarial network initialization, and calculating the Gram matrix corresponding to the feature map, which is used as an auxiliary matrix for color source enhancement; In the color matching auxiliary model training, the digital art sketch data is specifically used as the original content image, and the sketch coloring data in the optimized digital art design data set is used as the coloring style target image; Step S32: constructing a coloring style loss, specifically integrating a color content loss function and an improved coloring style loss function, and constructing a coloring style loss. The calculation formula is: ； Where, L total is the coloring style loss function, is the color content loss weight, L con is the color content loss function, and specifically adopts the standard content loss function as the color content loss function. is the improved coloring style loss weight, L csys is the improved coloring style loss function, L is the total number of feature map layers, l is the feature map layer index, N l is the total number of feature maps of the l-th layer, M l is the feature map size of the feature map of layer l, W l is the feature map weight of the feature map of layer l, i is the feature map index, j is the feature map position index, is the Gram matrix of the position of the jth feature map corresponding to the i-th feature map of the l-th layer feature map, is the feature map matrix of the coloring style target image corresponding to the jth feature map position of the i-th feature map of the l-th layer feature map; Step S33: constructing a color source enhancement analysis operator, specifically, constructing the color source enhancement analysis operator by sequentially performing color space conversion, principal component analysis dimensionality reduction, and style color transmission to obtain color matching mapping data; The color space conversion specifically refers to converting the color of the original image from the RGB space to the CIELAB color space to obtain color space converted image data; The principal component analysis dimensionality reduction specifically performs principal component analysis dimensionality reduction operation on the color space conversion image data, and performs style color transmission by calculating the color matching between the generated image and the colored style target image; Step S34: coloring style comprehensive mixing, specifically, using alpha blending to perform coloring style comprehensive mixing to obtain a coloring style matching image output, and by obtaining the coloring style matching image output, optimizing the color matching auxiliary model to generate an output result of digital art coloring; Step S35: Color matching auxiliary model training, specifically, by initializing the improved generative adversarial network, constructing the coloring style loss, constructing the color source enhancement analysis operator and the coloring style comprehensive mixing, the color matching auxiliary model training is performed to obtain the color matching auxiliary model Model COL ; Step S36: color matching assistance, specifically, using the color matching assistance model Model according to the optimized digital art design data set and the digital art sketch data COL , assist in color matching and obtain digital art coloring data.

5. The artificial intelligence-based digital art design auxiliary method according to claim 4, characterized in that: In step S4, the step of using the improved generative adversarial network method combined with style extraction enhancement to perform adaptive style fusion to obtain stylized digital art image data includes: Step S41: constructing a multi-path generator network, specifically constructing a multi-path generator network including a global space path, a category space path and a category context path, a feature fusion layer and a conditional normalization layer; The global space path is used to accept the complete image as input; the category space path is used to accept the annotation data of the style type as input; the category context path is used to process the context information of the style type; the feature fusion layer is used to fuse the outputs of the global space path, the category space path and the category context path; the conditional normalization layer is used to take the annotation data of the style type as conditional input, generate normalization parameters, and optimize the content feature map generation; Step S42: constructing a style extraction enhancement network, specifically applying a pre-trained convolutional neural network to extract style features, and obtaining style feature vector data through channel attention weighting; Step S43: constructing an adaptive style fusion network, specifically, performing contour style feature transformation and color style feature transformation in sequence according to the style feature vector data to obtain style change feature vector data, and performing feature fusion on the style change feature vector data and the style feature vector data through a feature fusion module to obtain adaptive style fusion feature data, and generating a style image according to the adaptive style fusion feature data; Step S44: constructing an adaptive loss function, specifically using an integrated loss function of multi-scale adversarial loss, perceptual loss and feature matching loss as an adaptive loss function for optimizing adaptive style fusion model training; Step S45: training the adaptive style fusion model, specifically, training the adaptive style fusion model by constructing the multi-path generator network, constructing the style extraction enhancement network, constructing the adaptive style fusion network and constructing the adaptive loss function, to obtain the adaptive style fusion model Model STY ; Step S46: Adaptive style fusion, specifically using the adaptive style fusion model Model STY , performing adaptive style fusion based on the optimized digital art design data set and the digital art coloring data to obtain stylized digital art image data.

6. The artificial intelligence-based digital art design auxiliary method according to claim 5, characterized in that: In step S5, the digital art design assistance is used to combine sketch generation, color matching and style fusion to perform comprehensive digital art design assistance, specifically, by initially generating the sketch, generating digital art sketch data, and performing sketch design assistance based on the digital art sketch data, by the color matching assistance, obtaining digital art coloring data, and performing color matching design assistance based on the digital art coloring data, by the adaptive style fusion, obtaining stylized digital art image data, and performing stylized design assistance based on the stylized digital art image data, and performing comprehensive digital art design assistance through the image design assistance, the color matching design assistance and the stylized design assistance, to obtain comprehensive reference data for digital art design assistance; The digital art design auxiliary reference data specifically includes output image data, output sketch parameters, output coloring parameters and output art style parameters.

7. An artificial intelligence-based digital art design assistance system, used to implement an artificial intelligence-based digital art design assistance method as claimed in any one of claims 1 to 6, characterized in that: It includes data collection and processing module, preliminary sketch generation module, color matching auxiliary module, adaptive style fusion module and digital art design auxiliary module.

8. The artificial intelligence-based digital art design assistance system according to claim 7, characterized in that: The data collection and processing module is used for data collection and processing, and obtains an optimized digital art design data set through data collection and processing, and sends the optimized digital art design data set to the sketch preliminary generation module; The sketch preliminary generation module is used for preliminary sketch generation, obtains digital art sketch data through preliminary sketch generation, and sends the digital art sketch data to the color matching auxiliary module and the digital art design auxiliary module; The color matching auxiliary module is used for color matching assistance, obtains digital art coloring data through color matching assistance, and sends the digital art coloring data to the adaptive style fusion module and the digital art design auxiliary module; The adaptive style fusion module is used for adaptive style fusion, obtains stylized digital art image data through adaptive style fusion, and sends the stylized digital art image data to the digital art design auxiliary module; The digital art design assistance module is used for digital art design assistance, and obtains digital art design assistance reference data through digital art design assistance.

Images