Image-text design page identifying, typesetting and arranging method

Abstract

The invention relates to the technical field of computer vision and artificial intelligence, and discloses an image-text design page recognition, typesetting and arrangement method, which comprises the following steps of: analyzing a historical work set of a designer, abstractly calculating into a design principle, and establishing a bidirectional mapping model of the design principle and specific implementation; a multi-agent typesetting decision-making mechanism is constructed, different typesetting tasks are decomposed into a plurality of special agents for processing, and all the agents work cooperatively; a two-way knowledge exchange mechanism is realized, and the typesetting reasoning process is presented in a visual mode; a novel design scheme is generated through recombination and variation of the mastered design principle, a probability model is established to evaluate a variation scheme, and a creative scheme is screened out; the feedback of a designer is quickly learned after a small amount of interaction, so that continuous learning and optimization are realized; the image-text design page identifying, typesetting and arranging method can understand the style of a designer, has creativity, supports two-way communication and can continuously learn.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and artificial intelligence, and more specifically, to a method for identifying, typesetting, and organizing graphic design pages. Background Art

[0002] With the explosive growth of digital publishing and multimedia content, the demand for graphic design typesetting has increased geometrically. Designers need to integrate personal styles with design principles to form unique typesetting effects. However, the resources of professional designers are limited and difficult to meet the market demand. In the prior art, automatic typesetting software mainly uses preset templates and simple rules. Although the typesetting system based on machine learning can learn simple styles, it lacks the understanding of designers' high-level thinking. Traditional human-computer interaction systems usually execute one-way instructions and have obvious deficiencies in aspects such as quantitative extraction of design styles, creativity of AI systems, two-way communication between humans and machines, and system adaptability.

[0003] Therefore, how to provide a method for identifying, typesetting, and organizing graphic design pages that can understand designers' styles, be creative, support two-way communication, and can continuously learn has become an urgent problem to be solved. Summary of the Invention

[0004] The present invention provides a method for identifying, typesetting, and organizing graphic design pages, which solves the technical problem of how to provide a method for identifying, typesetting, and organizing graphic design pages that can understand designers' styles, be creative, support two-way communication, and can continuously learn in the prior art.

[0005] The present invention provides a method for identifying, typesetting, and organizing graphic design pages, including the following steps: By analyzing the historical works of designers, extract high-level design thinking and decision-making patterns, abstract them into computable design principles, and establish a two-way mapping model between design principles and specific implementations; Based on the extracted design principles, construct a multi-agent typesetting decision-making mechanism, and decompose different typesetting tasks into multiple specialized agents through multi-agent reinforcement learning, so that each agent works collaboratively, integrating designers' personal styles while ensuring basic design specifications; Based on the multi-agent typesetting decision-making mechanism, implement a two-way knowledge communication mechanism, and present the typesetting reasoning process in an intuitive way through decision-making process visualization technology, enabling designers to understand the basis of system decisions and provide targeted guidance; Based on the two-way knowledge communication mechanism, generate novel design schemes through recombination and mutation of the mastered design principles, and establish a probability model to evaluate the feasibility and innovation of the mutated schemes, and screen out the schemes that conform to designers' styles and are creative; Based on a novel design scheme, the Meta-Learning fast adaptation algorithm quickly learns the designer's feedback after a small number of interactions, records the interaction history between the designer and the system and the creative evolution path, and realizes continuous learning and optimization.

[0006] Furthermore, the extraction of high-level design thinking and decision-making patterns includes the following steps: Preprocess the designer's historical portfolio, including image segmentation, text recognition, element classification, and hierarchical structure recognition; Use unsupervised learning algorithms to perform clustering and correlation analysis on the preprocessed design element feature data to identify the layout rules in the designer's works; Abstract the extracted layout rules into high-level design principles and construct a mathematical representation model.

[0007] Furthermore, the multi-agent layout decision-making mechanism includes agents: text arrangement agent, image layout agent, color coordination agent, and hierarchical organization agent. Each agent shares state information through a communication network and uses a consensus algorithm to coordinate decision-making conflicts.

[0008] Furthermore, the two-way knowledge exchange mechanism includes the following steps: Record the key information of the system decision-making process and organize the decision-making process information into a directed decision graph; Convert the recorded decision-making process into an intuitive visual representation for easy understanding by the designer; Collect the designer's feedback on the system decision-making and perform structured processing to provide data for subsequent learning; Integrate the designer's knowledge into the system decision-making model to form a knowledge accumulation mechanism for continuous learning.

[0009] Furthermore, principle mutation and innovation scheme generation include the following steps: Construct a series of principle mutation operators, including parameter perturbation operator, crossover recombination operator, rule transformation operator, and cross-domain transfer operator; Use the mutation operator to generate diverse design schemes and evaluate their feasibility and innovativeness; Based on the evaluated mutated principles, generate specific innovative design schemes; Generate feasible boundary cases to help the designer clarify the boundaries of personal design principles.

[0010] Furthermore, the Meta-Learning fast adaptation algorithm adopts the Model-Agnostic Meta-Learning algorithm and achieves fast adaptation through gradient update: ; where represents in the task Based on the optimal parameters of the fast adaptation process, the parameters are updated by gradient descent so that the model can quickly adapt to the new task representation in the task Based on the optimal parameters of the fast adaptation process, the parameters are updated by gradient descent so that the model can quickly adapt to the new task is the inner learning rate, denotes the meta - learning model with as the parameters, denotes the loss function on the task denotes taking the gradient with respect to the parameters for calculating the rate of change of the loss function with respect to the parameters of the change rate.

[0011] Furthermore, the recording of the interaction history and the creative evolution path includes the following steps: Construct an interactive session data structure to record system operations and designer feedback; Construct a creative evolution graph, where nodes represent design solutions and edges represent the evolution relationships between solutions; Add designer evaluations and modification information to each evolution node to form an annotated evolution record; Implement a version control mechanism to record the change trajectories of design principles and preferences over time.

[0012] Furthermore, the bidirectional mapping model between design principles and specific implementations includes: The principle - to - implementation mapping function : where, denotes mapping the abstract design principle to the specific implementation parameters; is implemented through a multi - layer neural network: ; where, is the activation function, and are the first weight matrix and the second weight matrix respectively, and are the first bias vector and the second bias vector respectively, denotes the set of specific implementation parameters, including typesetting layout, font, color, spacing specific parameters; The implementation - to - principle mapping function : where, denotes mapping the specific implementation back to the abstract principle; is implemented through a reverse inference network: ; wherein, is an activation function, and are the third weight matrix and the fourth weight matrix respectively, and are the third bias vector and the fourth bias vector respectively, represents a set of abstract design principles, including high-level design concepts such as balance, contrast, and hierarchy; Consistency loss function: ; Ensure the consistency of the bidirectional mapping, wherein, represents a set of abstract design principles, represents the principle-to-implementation mapping function, represents the implementation-to-principle mapping function, represents a set of specific implementation parameters, represents the consistency loss function of the bidirectional mapping, represents the L2 norm, which is used to calculate the Euclidean distance between vectors.

[0013] Furthermore, the designer style knowledge base is constructed through the following steps: Extract the explicit and implicit preferences of the designer from the interaction history; Construct a hierarchical style model, including the set of preferences, the relationships between preferences, and the weights of each preference; Apply an incremental update algorithm to dynamically adjust the style model according to new interaction data; Implement a style consistency checking mechanism to ensure the internal consistency of the style model and resolve possible preference conflicts.

[0014] A computer-readable storage medium is used to store computer-readable instructions, which can run a method for identifying, typesetting, and organizing graphic design pages when the computer-readable instructions are read by a computer.

[0015] The beneficial effects of the present invention are as follows: By analyzing the historical portfolio of designers, the present invention extracts high-level design thinking and decision-making patterns, abstracts them into computable design principles, and establishes a two-way mapping model between design principles and specific implementations; Based on the extracted design principles, a multi-agent typesetting decision-making mechanism is constructed. Through multi-agent reinforcement learning, different typesetting tasks are decomposed into multiple specialized agents for processing, enabling the agents to work together, incorporating the personal style of designers while ensuring basic design specifications; Based on the multi-agent typesetting decision-making mechanism, a two-way knowledge exchange mechanism is realized. Through the visualization technology of the decision-making process, the typesetting reasoning process is presented in an intuitive manner, enabling designers to understand the decision-making basis of the system and provide targeted guidance; Based on the two-way knowledge exchange mechanism, by reorganizing and mutating the mastered design principles, novel design schemes are generated, and a probability model is established to evaluate the feasibility and innovation of the mutated schemes, screening out schemes that not only conform to the designer's style but also have creativity; Based on the novel design scheme, through the Meta-Learning fast adaptation algorithm, the system quickly learns the designer's feedback after a small number of interactions, records the interaction history between the designer and the system and the creative evolution path, realizes continuous learning and optimization, so as to be able to understand the designer's style, be creative, support two-way communication and be able to continuously learn. A method for identifying, typesetting, and organizing graphic design pages; Extracting the designer's style through deep learning algorithms to make the typesetting matching degree of the system stable; Enhancing creativity by using multi-agent systems and principle mutation algorithms; Realizing two-way communication between AI and designers to improve the acceptance rate of system recommendations; Based on the Meta-Learning algorithm to achieve fast adaptation, the present invention realizes the transformation from a passive tool to a creative partner, with a certain improvement in work efficiency and a certain improvement in designer satisfaction, providing an innovative solution for the automation and personalization of graphic design typesetting. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flowchart of the method for identifying, typesetting, and organizing graphic design pages provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0018] In at least one embodiment of the present invention, a method for identifying, typesetting, and organizing graphic design pages is disclosed, as Figure 1 shown, including: Step 1: By analyzing the designer's historical portfolio, extract high-level design thinking and decision-making patterns, abstract them into computable design principles, and establish a two-way mapping model between design principles and specific implementations.

[0019] In this step, by analyzing the designer's historical portfolio, extract high-level design thinking and decision-making patterns, and abstract them into computable design principles.

[0020] Step 1.1: Preprocessing of design work data.

[0021] Input the designer's historical portfolio, including the original data of graphic and text layout pages, and preprocess this data, including image segmentation, text recognition, element classification, and hierarchical structure recognition. Analyze the attributes of each design element, such as position, size, color, contrast, font, etc., through computer vision algorithms and deep learning networks, and output a standardized design element feature vector.

[0022] Step 1.2: Typesetting rule extraction algorithm.

[0023] Use unsupervised learning algorithms to perform clustering and correlation analysis on the preprocessed design element feature data to identify the typesetting rules in the designer's works. The specific implementation includes: Apply hierarchical clustering algorithms to identify the spatial relationship rules between elements and generate an element relationship diagram: ; Among them, represents the set of design elements, represents the spatial relationship between elements; Use the attention mechanism to analyze the distribution of visual importance of elements and calculate the visual weight of each element: ; Among them, represents the attention between element and element , represents the correlation function between element features, represents the total number of elements, and respectively represent the th and th element feature vectors, represents the visual weight of element ; Extract the commonly used element combination patterns of the designer through frequent pattern mining algorithms and output the typesetting rule set: ; Among them, respectively represent different typesetting rules, and m represents the total number of specific typesetting patterns.

[0024] Step 1.3: Principle Abstraction and Representation Model

[0025] Abstract the extracted layout rules into high-level design principles and construct a mathematical representation model. The specific implementation includes: Use the hierarchical Bayesian model to map the layout rules to abstract design principles, expressed as: ; where, represents the th design principle, represents the posterior probability of the th design principle given the layout rule , represents the likelihood probability of observing the layout rule under the th design principle, represents the prior probability of the th design principle.

[0026] Construct a design principle representation vector: ; where, respectively represent the attribute values of the principle on the th dimension, represents the total number of attribute values.

[0027] The relationships between design principles are represented by a graph structure: ; where, is the set of design principles, represents the association relationship between principles.

[0028] Step 1.4: Two-way Mapping Model between Principles and Implementations

[0029] Construct a two-way mapping model between design principles and specific implementations to support two-way reasoning from principles to implementations and from implementations to principles. The specific implementation is as follows: Principle-to-Implementation Mapping Function : where, represents mapping the abstract design principle to specific implementation parameters; is implemented through a multi-layer neural network: ; where, is the activation function, and are the first weight matrix and the second weight matrix respectively, and are the first bias vector and the second bias vector respectively, represents a set of specific implementation parameters, including typesetting layout, font, color, and spacing specific parameters; Implementation-to-Principle Mapping Function : Among them, represents the reverse mapping of specific implementation to abstract principles; is implemented through a reverse inference network: ; Among them, is the activation function, and are the third weight matrix and the fourth weight matrix respectively, and are the third bias vector and the fourth bias vector respectively, represents a set of abstract design principles, including high-level design concepts such as balance, contrast, and hierarchy; Consistency Loss Function: ; Ensure the consistency of the bidirectional mapping. Among them, represents a set of abstract design principles, represents the Principle-to-Implementation Mapping Function, represents the Implementation-to-Principle Mapping Function, represents a set of specific implementation parameters, represents the consistency loss function of the bidirectional mapping, represents the L2 norm, which is used to calculate the Euclidean distance between vectors.

[0030] Train the mapping model by minimizing the consistency loss function, and output a design principle implementation mapping model that can support bidirectional reasoning.

[0031] Step 2: Based on the extracted design principles, construct a multi-agent typesetting decision-making mechanism, and decompose different typesetting tasks into multiple specialized agents through multi-agent reinforcement learning, enabling each agent to work collaboratively and integrating the personal style of the designer while ensuring basic design specifications.

[0032] Based on the extracted design principles, construct a personalized typesetting decision-making system, which decomposes different typesetting tasks into multiple specialized agents for processing to achieve collaborative work.

[0033] Step 2.1: Agent Function Decomposition and Structure Definition.

[0034] Decompose the typesetting task into multiple functional components, and define the functions and structures of each agent. The specific implementation includes: According to the functional requirements of the design task, the typesetting task is decomposed into subtasks such as text arrangement, image layout, color coordination, and hierarchical organization; Define corresponding specialized agents for each subtask , and each agent contains a state space , an action space , a transition function , and a reward function ; Define the communication protocol and interaction interface between agents, and establish a multi-agent network topology structure: ; Among them, represents the set of agents, represents the communication connection between agents.

[0035] Step 2.2: Construction of a principle-based reward function.

[0036] Convert the extracted design principles into the reward function of the agent to guide the decision-making behavior of the agent. The specific implementation is as follows: For each agent , construct a reward function based on the design principle: ; Among them, and represent the current state and the next state respectively, represents the action, represents the reward component related to the th design principle, is the weight of the reward function, represents the total number of design principles.

[0037] Principle compliance evaluation function: ; Among them, represents the degree of compliance measurement function between the design principle and the current state , is the state feature extraction function, represents the feature vector extracted from the state .

[0038] Adjust the weight of the reward function through meta-learning so that the reward function can accurately reflect the personal style preferences of the designer.

[0039] Step 2.3: Multi-agent reinforcement learning training.

[0040] Train a multi-agent system using reinforcement learning algorithms so that each agent learns an optimal decision-making strategy. The specific implementation is as follows: Adopt a multi-agent deep reinforcement learning framework, and each agent learns its Q function based on the Deep Q-Network (DQN): ; Among them, is the discount factor, is the next action, is the immediate reward, is the expected value, is the state-action value function, represents selecting the action that can maximize the Q value among all possible next actions .

[0041] Apply the experience replay technique to store interaction experiences in the replay buffer : ; Among them, represents the reward, and randomly sample batch data for learning.

[0042] ; Among them, and are the parameters of the current network and the target network respectively, represents the agent 's loss function, which is used to measure the mean square error between the predicted value and the target value of the Q network.

[0043] Use a central coordinator to coordinate the actions among multiple agents to maximize the overall reward: ; Among them, is the weight coefficient of the reward of the -th agent, is the total number of agents.

[0044] Step 2.4: Implementation of the collaborative decision-making mechanism.

[0045] Build a collaborative decision-making mechanism among multiple agents to ensure that each agent can work in coordination and produce a consistent typesetting result. The specific implementation is as follows: Build an agent communication network to allow agents to share key state information: ; Among them, represents the message sent from agent to agent . is a message encoding function, represents the current state of the agent , represents the action of the agent .

[0046] Develop a consensus algorithm to coordinate the decision-making conflicts of each agent: ; Among them, represents the optimal action of the agent , represents the action corresponding to the maximum value of the function , represents all the messages received by the agent .

[0047] Implement a hierarchical decision-making structure, where high-level agents are responsible for global planning, and low-level agents execute specific tasks. Ensure decision consistency through two-way information flow from top to bottom and from bottom to top.

[0048] Through the above steps, output a multi-agent system that can make typesetting decisions according to the designer's personal style.

[0049] Step 3: Based on the multi-agent typesetting decision-making mechanism, implement a two-way knowledge exchange mechanism, and present the typesetting reasoning process in an intuitive way through decision process visualization technology, so that designers can understand the system's decision-making basis and give targeted guidance.

[0050] Implement a two-way knowledge exchange mechanism between the AI system and the designer, and present the typesetting reasoning process in an intuitive way through decision process visualization technology.

[0051] Step 3.1: Record and structure the system decision-making process.

[0052] Record the key information of the system decision-making process and perform structured processing. The specific implementation is as follows: During the process of the multi-agent system making typesetting decisions, record information such as the state transition, action selection, reward value, and decision-making basis of each agent; Organize the decision-making process information into a directed decision-making graph; ; Among them, the node represents the system state, and the edge represents the state transition and the corresponding decision-making action.

[0053] Add metadata to each decision-making node, including decision-making basis, optional solution scoring, and application of design principles to form a decision-making knowledge base 。

[0054] Step 3.2: Visual expression of the decision-making process.

[0055] Convert the recorded decision-making process into an intuitive visual expression for easy understanding by designers. The specific implementation is as follows: Build a decision tree visualization model, decompose the complex decision-making process into an intuitive tree structure, where the size of the nodes represents the importance of the decision, and the thickness of the edges represents the state transition probability; Develop an algorithm for highlighting key decision points, and identify key decision points through the information gain formula: ; where represents the information entropy, represents the decision attribute, represents the decision attribute of all possible value sets, represents the decision attribute when the value is the corresponding sub-state set.

[0056] Implement an interactive hierarchical expansion function, allowing designers to expand or collapse different branches of the decision tree as needed to explore decision paths of interest.

[0057] Step 3.3: Collection and structuring of designers' feedback.

[0058] Collect designers' feedback on the system's decisions and perform structured processing to provide data for subsequent learning. The specific implementation is as follows: Develop an interactive feedback interface that allows designers to rate, modify, and annotate the system's decisions; Map designers' feedback to the corresponding decision nodes and design principles to establish a ternary association of feedback - decision - principle: ; where represents the feedback, represents the decision, represents the design principle.

[0059] Perform semantic analysis on the feedback data, extract designers' implicit preferences and decision-making bases, and convert them into a knowledge representation that can be understood by the system.

[0060] Step 3.4: Construction of a two-way knowledge exchange model.

[0061] Based on decision visualization and feedback collection, achieve two-way knowledge exchange between the system and designers. The specific implementation is as follows: Build a knowledge transfer channel, where the system transfers decision-making bases and suggestions to designers, and designers transfer feedback and knowledge to the system; Develop a knowledge fusion algorithm to integrate designers' knowledge into the system decision-making model: ; Among them, represents the original model, represents the fused model, represents the knowledge fusion operation.

[0062] Implement an interactive historical memory model to record the interaction history between the system and the designer, forming a continuous learning knowledge accumulation mechanism.

[0063] Through the above steps, output a decision-making process visualization system that supports two-way knowledge exchange, enabling designers to understand the system decision-making basis and effectively guide the system.

[0064] Step 4: Based on the two-way knowledge exchange mechanism, generate novel design solutions by reorganizing and mutating the mastered design principles, and establish a probability model to evaluate the feasibility and innovativeness of the mutated solutions, and screen out solutions that conform to the designer's style and are creative.

[0065] Generate novel design solutions by reorganizing and mutating the mastered design principles, and evaluate their feasibility and innovativeness.

[0066] Step 4.1: Construction of principle mutation operators.

[0067] Construct a series of principle mutation operators for generating variants of the original design principles. The specific implementation is as follows: Parameter perturbation operator: Randomly perturb the parameter values of the design principle: ; Among them, represents the perturbed parameter value, represents the original parameter value, obeys a normal distribution with a mean of 0.

[0068] Cross-recombination operator: Combine the characteristics of two different design principles: ; Among them, is the weight coefficient, is the newly generated design principle, and are two different original design principles respectively.

[0069] Rule transformation operator: Transform the application rules of the design principle, such as symmetry conversion, hierarchical reorganization; Cross-domain migration operator: Introduce principle characteristics from other design fields and adapt them to the current field: ; Among them, is a cross - domain adaptation function, is the source domain principle, is the target domain principle, is the new design principle generated after migration.

[0070] Step 4.2: Generation and evaluation of mutation schemes.

[0071] Generate diverse design schemes using mutation operators and evaluate their feasibility and innovativeness. The specific implementation is as follows: Generate a set of candidate design principles through mutation operators: ; Among them, represents the th mutated candidate design principle, represents the total number of candidate principles.

[0072] Screen the candidate principles using a feasibility evaluation function: ; Among them, represents the feasibility score of the candidate principle , represents the conditional probability of the candidate principle under the historical design specification , used to determine whether the mutated principle conforms to the basic design specification;.

[0073] Apply an innovativeness evaluation function to calculate the novelty of the candidate principles: ; Among them, represents the novelty score of the candidate principle , represents finding the maximum similarity value with the candidate principle in the existing principle set , is the similarity function.

[0074] Comprehensively evaluate the value of the candidate principles: ; Among them, and are weight coefficients respectively, represents the comprehensive value score of the candidate principle , used to measure the overall quality of the candidate principle, and select the candidate principle with the highest value to enter the next stage.

[0075] Step 4.3: Generation and screening of innovative schemes.

[0076] Generate a specific innovative design solution based on the evaluated variation principles. The specific implementation is as follows: Apply the filtered variation principles to the bidirectional mapping model to generate specific implementation parameters: , where represents the generated implementation parameters.

[0077] Use a multi-agent system to generate a set of candidate design solutions based on the new parameters: ; where represents the th candidate design solution, represents the total number of candidate design solutions.

[0078] Apply a compliance evaluation function to evaluate the consistency of each solution with the designer's style: ; where represents the degree of compliance of candidate solution with the designer's style, represents the candidate solution to be evaluated, represents the designer's historical portfolio, represents the similarity calculation function.

[0079] Based on the weighted scores of innovation and consistency, screen out the final recommended solution: ; where represents the comprehensive score of solution , represents the innovation score, represents the style consistency score, and are the weight coefficients respectively.

[0080] Step 4.4: Generation of boundary cases and exploration of principle boundaries.

[0081] Generate nearly feasible boundary cases to help the designer clarify the boundaries of personal design principles. The specific implementation is as follows: Perform boundary perturbation on the design principles to generate principle variants on the edge of feasibility: ; where represents the generated boundary principle variant after perturbation, represents the original design principle, is the perturbation step size, is the gradient of the feasibility function.

[0082] Generate the corresponding boundary design scheme based on the boundary principle: ; where represents the generated boundary design scheme, which is obtained by mapping the boundary principle to the specific implementation parameter space, Identify the key features of the principle boundary through comparative analysis: ; where represents the set of key features of the principle boundary, represents a feature, represents a critical value, represents a tolerance range.

[0083] Construct a visual representation model of the principle boundary to intuitively display the applicable range and boundary conditions of the design principle.

[0084] Through the above steps, output a design scheme that not only conforms to the designer's style but also has innovation, and provide the exploration results of the principle boundary.

[0085] Step 5: Based on the novel design scheme, use the Meta-Learning fast adaptation algorithm to quickly learn the designer's feedback after a small number of interactions, and record the interaction history between the designer and the system and the creative evolution path to achieve continuous learning and optimization.

[0086] Based on the Meta-Learning fast adaptation algorithm and the interaction history record, achieve the continuous learning and optimization of the system.

[0087] Step 5.1: Construction of the Meta-Learning fast adaptation model.

[0088] Construct a fast adaptation model based on meta-learning to enable the system to quickly learn the designer's feedback through a small number of interactions. The specific implementation is as follows: Define the task distribution , where each task represents a specific design requirement; Construct the meta-learning model , and train the parameters through a two-layer optimization process: ; where represents the fast adaptation on task based on parameter , represents the meta-learning model with as the parameter, represents on task The loss function on represents the task distribution, and represents the parameter values when the objective function reaches its minimum.

[0089] Implement the Model - Agnostic Meta - Learning algorithm and achieve fast adaptation through gradient updates: ; Among them, is the inner - layer learning rate, represents taking the gradient of the parameter for calculating the change rate of the loss function with respect to the parameter .

[0090] According to the characteristics of the design field, construct the task representation and the loss function to ensure that the model can capture the subtle differences in design preferences.

[0091] Step 5.2: Interaction history and creative evolution record.

[0092] Record the interaction history between the designer and the system and the creative evolution path to provide a data basis for continuous learning. The specific implementation is as follows: Construct the interaction session data structure: ; Among them, represents a complete interaction session record between the system and the designer, represents the system operation, represents the designer's feedback.

[0093] Construct the creative evolution graph: ; Among them, the node represents the design scheme, and the edge represents the evolution relationship between the schemes, represents the directed graph structure composed of the design schemes and their evolution relationships during the creative evolution process.

[0094] Add the designer's evaluation and modification information to each evolution node to form an annotated evolution record; Implement a version control mechanism to record the changing trajectory of design principles and preferences over time.

[0095] Step 5.3: Construction of the designer style knowledge base.

[0096] Based on the interaction history and the creative evolution record, construct the designer style knowledge base. The specific implementation is as follows: Extract the explicit and implicit preferences of the designer from the interaction history: ; Among them, represents a preference extraction function, which is used to extract the designer's preference information from the interaction session record ; represents the set of designer preferences extracted, including explicitly expressed preferences and implicit preferences inferred through behavior analysis.

[0097] Build a hierarchical style model: ; Among them, represents the designer's style model, represents the weight of each preference.

[0098] Develop an incremental update algorithm to dynamically adjust the style model according to new interaction data: ; Among them, represents the updated style model, represents the new interaction session data, represents the model update function.

[0099] Implement a style consistency check mechanism to ensure the internal consistency of the style model and solve possible preference conflicts.

[0100] Step 5.4: Continuous learning and model optimization.

[0101] Based on the designer style knowledge base and the Meta-Learning model, realize the continuous learning and optimization of the system. The specific implementation is as follows: Develop an experience replay mechanism to sample data from historical interactions regularly for model update; Implement an adaptive learning rate adjustment strategy: ; Among them, represents the learning rate at the t-th step, represents the initial learning rate, represents the number of learning steps, is the smoothing constant.

[0102] Build a knowledge distillation model to fuse the model knowledge of multiple versions into a unified optimized model: ; Among them, represents the distilled model, represents the first, second,..., x-th different versions of the source model, represents the total number of different versions of the source model, represents the knowledge distillation function.

[0103] Build a model evolution mechanism to adjust the system learning strategy according to the changes in the designer's style, and adapt to the growth and changes of the designer's style.

[0104] Through the above steps, a collaborative creative system that can continuously learn and optimize is output, which can adapt to the changes and growth of the designer's style.

[0105] This embodiment realizes the following technical effects through an innovative method for identifying, typesetting, and organizing graphic design pages: Improve work efficiency: After 50 interactions, the matching degree between the typesetting result and the designer's expectation reaches 85%, the manual adjustment time is reduced by 67%, the designer satisfaction is increased by 58%, and the work efficiency is increased by 3.2 times. This efficiency improvement stems from the system's deep understanding and precise application of the designer's personal style.

[0106] Enhance innovation ability: Through the principle mutation and cross - domain migration mechanism, the system can provide designers with novel and feasible design solutions, break through creative bottlenecks, and expand design possibilities. Experiments show that among the innovative solutions generated by the system, 30% are rated as inspiring by designers, and 15% are directly adopted and applied.

[0107] Promote human - machine collaboration: The system transforms from a passive tool into a creative partner, forms a positive interaction with the designer, co - evolves, and continuously improves the capabilities of both parties. Long - term usage data shows that the collaboration efficiency between the designer and the system shows a logarithmic growth trend with the increase in usage time.

[0108] Personalized adaptation: The system can adapt to the styles and preferences of different designers, provide customized typesetting services, and its adaptability continuously improves with the increase in interactions. In multi - designer tests, the system can identify and adapt to the basic style characteristics of new designers after an average of 10 interactions.

[0109] Transparent decision - making process: Through decision - making visualization technology, designers can understand the basis for the system's decisions, enhance trust, and can specifically guide the system's improvement. User research shows that the increase in decision - making transparency has increased the acceptance rate of system recommendations by designers by 45%.

[0110] In summary, the method for identifying, typesetting, and organizing graphic design pages provided in this embodiment not only solves the key problems in the prior art, but also achieves significant technical breakthroughs in aspects such as efficiency improvement, innovation promotion, and human - machine collaboration.

[0111] This section shows the application and its effects of the present invention in the actual scenario of graphic design typesetting and organization through specific examples: A design company needs to develop various brand manuals for its clients. Since different designers are responsible for different pages, the overall style of the manuals is inconsistent, and a large amount of manual adjustment is required for each typesetting to ensure unity. This system is applied to this scenario to learn and coordinate the typesetting styles of multiple designers.

[0112] Data collection and preprocessing of designers' works: 500 historical design works of 5 senior designers were collected, including brand manuals, brochures, etc. The system preprocessed these works and extracted element attributes. Some results are shown in Table 1.

[0113] Table 1: Example of design element feature extraction results

[0114] Extraction of typesetting rules and principle modeling: The system analyzed the spatial relationship and visual importance between elements, and extracted the core typesetting rules of Designer A, as shown in Table 2.

[0115] Table 2: Results of typesetting rule extraction for Designer A

[0116] Based on these rules, the system constructed an abstract model of design principles for Designer A. Some core principles are shown in Table 3.

[0117] Table 3: Parameters of the abstract model of principles for Designer A

[0118] Training process of the multi-agent typesetting decision-making mechanism: The system decomposed the typesetting task into 4 specialized agents: text arrangement agent, image layout agent, color coordination agent, and hierarchical organization agent. Table 4 shows the performance changes of the text arrangement agent during the training process.

[0119] Table 4: Performance changes of the text arrangement agent during training

[0120] Visualization of the decision-making process and knowledge exchange: The system uses a decision tree visualization model to show the typesetting decision-making process to designers. Figure 1 (The actual graph is omitted here) shows the reasoning process of the system's decision on the title position of a certain page, clearly indicating that the system makes a decision based on the principle that the title of the first chapter of the brand manual should be centered above.

[0121] Designers provided feedback on the system's decision through the interface. The system collected these feedbacks and processed them structurally. Some feedback data are shown in Table 5.

[0122] Table 5: Example of designers' feedback data

[0123] Innovation solution generation and final application: The system generated multiple innovative design solutions by mutating the principles of Designer A. A comparison of some mutated principles is shown in Table 6 below.

[0124] Table 6: Results of principle mutation

[0125] The designer selected P007 as the final application solution, and the system generated the overall layout of the brand manual based on this. A comparison of the example effects of one page is shown in Table 7 below.

[0126] Table 7: Comparison of layout effects

[0127] By applying this system in 10 actual projects, data related to work efficiency was collected, as shown in Table 8 below.

[0128] Table 8: Data on work efficiency improvement

[0129] The data shows that after 50 interactions, the matching degree between the typesetting result and the designer's expectation reached 85%, and the average work efficiency increased by 3.3 times.

[0130] Personalized adaptation ability: The adaptability of the system to the styles of 5 different designers was tested, and the number of interactions required for the system to learn and adapt to different designers' styles was recorded, as shown in Table 9 below.

[0131] Table 9: Test data on personalized adaptation ability

[0132] The results show that the system can adapt to the style characteristics of new designers after an average of 10 interactions, and the final matching degree reaches stability, with the average designer satisfaction reaching 4.6 points (out of 5).

[0133] In summary, this embodiment has shown significant technical effects in practical applications, especially in terms of work efficiency improvement and personalized adaptation ability, effectively solving the key technical problems in the field of graphic design page recognition, typesetting, and collation.

[0134] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A method for identifying, typesetting and organizing a graphic and text design page, characterized in that, It includes the following steps: By analyzing the designer's historical portfolio, extract high-level design thinking and decision-making patterns, abstract them into computable design principles, and establish a two-way mapping model between design principles and specific implementations; Based on the extracted design principles, construct a multi-agent typesetting decision-making mechanism. Decompose different typesetting tasks into multiple specialized agents through multi-agent reinforcement learning, enable each agent to work collaboratively, and integrate the designer's personal style while ensuring basic design specifications; Based on the multi-agent typesetting decision-making mechanism, implement a two-way knowledge exchange mechanism. Present the typesetting reasoning process in an intuitive way through decision process visualization technology, so that the designer can understand the system's decision-making basis and give targeted guidance; Based on the two-way knowledge exchange mechanism, generate novel design solutions through the recombination and mutation of the mastered design principles, and establish a probability model to evaluate the feasibility and innovation of the mutant solutions, and screen out solutions that conform to the designer's style and are creative; Based on the novel design solution, use the Meta-Learning fast adaptation algorithm to quickly learn the designer's feedback after a small number of interactions, and record the interaction history between the designer and the system and the creative evolution path to achieve continuous learning and optimization.

2. The method for identifying, typesetting and organizing a graphic and text design page according to claim 1, wherein The extraction of high-level design thinking and decision-making patterns includes the following steps: Preprocess the designer's historical portfolio, including image segmentation, text recognition, element classification, and hierarchical structure recognition; Use unsupervised learning algorithms to perform clustering and correlation analysis on the preprocessed design element feature data to identify the typesetting rules in the designer's works; Abstract the extracted typesetting rules into high-level design principles and construct a mathematical representation model.

3. A method for identifying, typesetting, and organizing a graphic and text design page according to claim 1, characterized in that, The multi-agent typesetting decision-making mechanism includes agents: text arrangement agent, image layout agent, color coordination agent, and hierarchical organization agent. Each agent shares state information through a communication network and uses a consensus algorithm to coordinate decision-making conflicts.

4. A method for identifying, typesetting, and organizing a graphic and text design page according to claim 1, characterized in that, The two-way knowledge exchange mechanism includes the following steps: Record the key information of the system decision-making process and organize the decision-making process information into a directed decision graph; Convert the recorded decision-making process into an intuitive visual expression for easy understanding by the designer; Collect the designer's feedback on the system decision-making and perform structured processing to provide data for subsequent learning; Integrate the designer's knowledge into the system decision-making model to form a knowledge accumulation mechanism for continuous learning.

5. A method for identifying, typesetting, and organizing a graphic design page according to claim 1, characterized in that Principle mutation and innovative solution generation include the following steps: Construct a series of principle mutation operators, including parameter perturbation operators, crossover recombination operators, rule transformation operators, and cross-domain migration operators; Use the mutation operator to generate diverse design solutions and evaluate their feasibility and innovation; Based on the evaluated mutant principles, generate specific innovative design solutions; Generate feasible boundary cases to help the designer clarify the boundaries of personal design principles.

6. A method for identifying, typesetting, and organizing a graphic and text design page according to claim 1, characterized in that The Meta-Learning fast adaptation algorithm adopts the Model-Agnostic Meta-Learning algorithm and achieves fast adaptation through gradient update: ； Among them, represents the fast adaptation process on the task based on the optimal parameters. The model can quickly adapt to the new task by updating the parameters through gradient descent. The fast adaptation process on the task based on the optimal parameters. The model can quickly adapt to the new task by updating the parameters through gradient descent. is the inner learning rate, represents the meta-learning model with as the parameter. represents the loss function on the task represents taking the gradient of the parameter with respect to the parameter to calculate the rate of change of the loss function 7. A method for identifying, typesetting, and organizing a graphic design page according to claim 1, characterized in that, The recording of the interaction history and creative evolution path includes the following steps: Construct an interaction session data structure to record system operations and designer feedback; Construct an innovation evolution graph, where nodes represent design solutions and edges represent the evolution relationships between solutions; Add designer evaluations and modification information to each evolution node to form an annotated evolution record; Implement a version control mechanism to record the change trajectories of design principles and preferences over time.

8. A method for identifying, typesetting and organizing a graphic and text design page according to claim 1, characterized in that The bidirectional mapping model between design principles and specific implementations includes: Principle-to-Implementation Mapping Function : Among them, represents mapping abstract design principles to specific implementation parameters; Implemented through a multi-layer neural network: ； Among them, is the activation function, and are the first weight matrix and the second weight matrix respectively, and are the first bias vector and the second bias vector respectively, represents a specific set of implementation parameters, including specific parameters for layout, font, color, and spacing; Implementation of the principle mapping function : Among them, means mapping the specific implementation back to the abstract principle; Implemented through a reverse inference network: ； Among them, is the activation function, and are the third weight matrix and the fourth weight matrix respectively, and are the third bias vector and the fourth bias vector respectively, represents a set of abstract design principles, including high-level design concepts such as balance, contrast, and hierarchy; Consistency loss function: ； Ensure the consistency of the bi-directional mapping, where, represents a set of abstract design principles, represents the principle-to-implementation mapping function, represents the implementation-to-principle mapping function, represents a set of specific implementation parameters, represents the consistency loss function of the bi-directional mapping, represents the L2 norm, which is used to calculate the Euclidean distance between vectors.

9. A method for identifying, typesetting, and organizing a graphic and text design page according to claim 1, characterized in that The designer style knowledge base is constructed through the following steps: Extract the explicit and implicit preferences of designers from the interaction history; Construct a hierarchical style model, including a preference set, the relationships between preferences, and the weights of each preference; Apply an incremental update algorithm to dynamically adjust the style model according to new interaction data; Implement a style consistency check mechanism to ensure internal consistency of the style model and resolve possible preference conflicts.

10. A computer-readable storage medium, characterized in that, It is used to store computer-readable instructions, which can run a method for identifying, typesetting, and organizing a graphic design page as described in any one of claims 1-9 when read by a computer.

Images