Artificial intelligence aided design color matching decision-making system and method

Abstract

The invention discloses an artificial intelligence aided design color matching decision-making system and method, and relates to the technical field of computer aided design. Comprising the steps of obtaining a design semantic text, user visual preference and scene illumination parameters; converting the text data into emotion vectors through machine learning; mapping the emotion vectors into a color base vector set by using a pre-trained emotion color mapping model; carrying out personalized parameter adjustment on the color basis vector based on the visual preference data of the user; environment self-adaptive correction is carried out in combination with scene illumination parameters; performing multi-dimensional intelligent evaluation on the corrected color set to generate a comprehensive confidence score and a color matching scheme; when the score is lower than a preset threshold value, iterative optimization is carried out, and finally an optimized color matching scheme is output. The problem that a traditional color matching tool is difficult to consider design intentions, user preferences and physical scenes is effectively solved, and the quality, personalization and practicability of a color design scheme are improved.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, specifically to an artificial intelligence-assisted design color matching decision system and method. Background Technology

[0002] Artificial intelligence-assisted design is an important development direction for the digital upgrade of the creative industries. It aims to improve design efficiency and quality by generating visual solutions that meet user needs through data-driven approaches and intelligent algorithms. In recent years, with the advancement of natural language processing and computer vision technologies, color scheme generation has gradually moved towards semantic and personalized approaches.

[0003] Existing technologies mostly recommend color schemes by analyzing design theme keywords or user historical preference data, achieving automated generation and style matching of color combinations. These methods can provide basic color references in general design scenarios and filter color libraries based on simple style tags (such as "minimalist" and "retro"), possessing a certain degree of automation. However, for professional-level design applications, color schemes need to maintain consistency between visual expression and design intent in complex and dynamic real-world environments. Existing technologies fail to integrate and coordinate user abstract emotional semantics, personalized visual physiological preferences, and the physical lighting conditions of the deployment environment into a unified decision-making process. This results in color schemes that often meet requirements at the conceptual stage, but in practical applications, due to differences in individual user visual perception or changes in ambient light, problems such as severe color distortion, deviation in emotional communication, or decreased visual comfort occur.

[0004] This disconnect means that the design process still relies on designers' human experience for repeated adjustments, making it impossible to achieve end-to-end intelligent decision-making from "concept generation" to "environmental implementation," thus limiting the deep application of artificial intelligence in key design stages. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-assisted design color matching decision-making system and method.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, this invention discloses an artificial intelligence-assisted design color matching decision-making method, comprising the following steps:

[0008] Acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data;

[0009] The designed semantic text data is converted into a pre-defined emotion vector using a machine learning algorithm, and the emotion vector is mapped to a set of color basis vectors using a pre-trained emotion color mapping model.

[0010] Based on the user's visual preference data, the color basis vector set is adjusted with personalized parameters to generate a personalized color set;

[0011] Based on the scene lighting parameter data, the personalized color set is subjected to environmental adaptive correction to generate a corrected color set;

[0012] The corrected color set is subjected to multi-dimensional intelligent evaluation, a comprehensive confidence score is calculated, and an initial color scheme is generated based on the corrected color set.

[0013] When the overall confidence score is lower than a preset threshold, a parallel optimization thread is started to iteratively optimize the parameters of at least two initial color schemes, and the final color scheme is selected based on the game result of the corresponding overall confidence scores.

[0014] Output the final color scheme after iterative optimization.

[0015] Secondly, this invention discloses an artificial intelligence-assisted design color matching decision-making system, which uses the aforementioned artificial intelligence-assisted design color matching decision-making method, including:

[0016] The data acquisition module is used to acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data;

[0017] The feature extraction module is used to convert the designed semantic text data into a pre-defined emotion vector using a machine learning algorithm, and to map the emotion vector into a set of color basis vectors using a pre-trained emotion color mapping model.

[0018] The adjustment module is used to adjust the personalized parameters of the color basis vector set based on the user's visual preference data to generate a personalized color set;

[0019] An environment adaptation module is used to perform environment adaptive correction on the personalized color set based on the scene lighting parameter data, and generate a corrected color set.

[0020] The evaluation module is used to perform multi-dimensional intelligent evaluation on the corrected color set, calculate the comprehensive confidence score, and generate an initial color scheme based on the corrected color set.

[0021] The optimization iteration module is used to start a parallel optimization thread when the overall confidence score is lower than a preset threshold, to iteratively optimize the parameters of at least two initial color schemes, and to select the final color scheme based on the game result of the corresponding overall confidence scores.

[0022] The output module is used to output the final color scheme after iterative optimization.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] 1. This invention establishes a collaborative data processing flow integrating design semantics, user preferences, and physical scene parameters. This achieves multi-dimensional requirement fusion in the color scheme generation process, transforming abstract design intentions (such as "composure" and "vitality") into quantifiable emotional vectors, which further drive color generation, ensuring the accuracy and consistency of the color scheme's semantic communication. Simultaneously, by introducing user visual preference data, the initially generated colors can be personalized and fine-tuned, making the output more aligned with the visual sensitivities of different users, effectively improving the subjective satisfaction and individual adaptability of the color scheme.

[0025] 2. This invention solves the problem of inaccurate performance of color schemes in practical application scenarios. By integrating scene lighting parameters and intelligently converting and correcting the color set in a device-independent color space, the final output color scheme can simulate the visual effect under target lighting in advance, thereby enhancing the physical scene robustness of the color scheme and the reliability of its practical application.

[0026] 3. This invention introduces an evaluation-optimization closed-loop mechanism based on comprehensive confidence scoring, endowing the system with self-verification and continuous improvement capabilities. This mechanism not only provides a multi-dimensional and quantifiable comprehensive evaluation of the overall quality of the color scheme, but more importantly, when the evaluation results do not meet expectations, the system can automatically trigger an iterative optimization process to adjust the scheme parameters in a targeted manner. This overcomes the quality fluctuation problem that may exist in traditional one-time scheme generation. Through cyclical feedback and correction, it gradually approaches the optimal solution, thereby stably outputting a high-quality, highly reliable final color scheme, ensuring the scientific rigor and superiority of the decision-making results. Attached Figure Description

[0027] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0028] Figure 1 This is a flowchart of the steps of the present invention;

[0029] Figure 2 This is a flowchart illustrating the working principle of the present invention.

[0030] Figure 3 This is a flowchart of the personalized parameter adjustment process of the present invention;

[0031] Figure 4 This is a system module connection diagram of the present invention. Detailed Implementation

[0032] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0033] Application Overview:

[0034] In existing technologies, color scheme generation in the design field largely relies on preset color rule libraries or style extraction from existing images, making it difficult to simultaneously satisfy the accurate communication of design intent, individual user visual preferences, and realistic representation of physical scenes. Traditional methods, when dealing with abstract design semantics (such as "high-end" or "stable"), often simply associate them with fixed colors, lacking the ability to analyze semantics at multiple levels and in a quantifiable manner. When it is necessary to consider the unique visual sensitivities of users or specific ambient lighting conditions, existing methods typically exhibit single-dimensional optimization, leading to color schemes that may deviate from intended meaning, have low individual acceptance, or poor scene adaptability in practical applications, making it difficult to support highly reliable design decisions.

[0035] To address the aforementioned issues, the research revealed a deep coupling relationship between design semantics, user visual preferences, and scene lighting parameters. By establishing a dynamic mapping model from semantics to color, multi-dimensional requirements can be synergistically optimized. The research found that machine learning algorithms can transform abstract textual semantics into numerical emotion vectors, providing a calculable driving signal for color generation. Simultaneously, user visual preferences can be represented as sensitivity weights to color attributes, used for personalized modulation of the initial color scheme. Scene lighting parameters can be corrected for physical realism in color schemes through a color space transformation model. Furthermore, by introducing a multi-dimensional evaluation and iterative optimization mechanism, a closed-loop verification and improvement of decision quality can be achieved.

[0036] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] Example 1:

[0038] like Figure 1 As shown, the AI-assisted color matching decision-making method includes the following steps:

[0039] Acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data;

[0040] We use machine learning algorithms to convert semantic text data into emotion vectors of a predetermined number of bits, and then use a pre-trained emotion color mapping model to map the emotion vectors into a set of color basis vectors.

[0041] Personalized color sets are generated by adjusting the parameters of the color basis vector set based on user visual preference data.

[0042] Based on scene lighting parameter data, the personalized color set is subjected to environmental adaptive correction to generate a corrected color set;

[0043] The corrected color set is intelligently evaluated from multiple dimensions, a comprehensive confidence score is calculated, and an initial color scheme is generated based on the corrected color set.

[0044] When the overall confidence score is lower than the preset threshold, a parallel optimization thread is started to iteratively optimize the parameters of at least two initial color schemes, and the final color scheme is selected based on the game result of the corresponding overall confidence scores.

[0045] Output the final color scheme after iterative optimization.

[0046] Specifically, such as Figure 2 As shown, the working principle of this application is as follows: The system first receives the design semantic text description through the user interface, collects or calls a pre-established user visual preference profile, and obtains the lighting parameters of the target application scene. The design semantic text is parsed and quantified by the embedded machine learning algorithm and converted into a numerical emotion vector with a predetermined number of bits. This emotion vector represents the emotion and style tendency behind the text in a compact form. This emotion vector is then used as input to a pre-trained emotion color mapping model. The emotion color mapping model learns the correspondence between emotion and color in a large number of design samples and can map the abstract emotion vector into a set of color basis vectors composed of several basic colors, laying the foundation for the color scheme.

[0047] To further enhance the personalization and scene adaptability of the solution, the system adjusts the parameters of the color basis vector set based on user visual preference data. Specifically, it fine-tunes and filters the basic colors according to the user's sensitivity and preference for brightness, saturation, contrast, etc., thereby generating a personalized color set that better matches the user's subjective experience. The system integrates scene lighting parameters and, based on color science and optical principles, simulates and corrects the personalized color set in a device-independent color space, enabling it to predict and adapt to the real visual effects in the target environment, generating a corrected color set.

[0048] To ensure the quality and diversity of the output schemes, the system performs multi-dimensional intelligent evaluation of the corrected color set, comprehensively considering factors such as its matching degree with the original semantics, the internal harmony of colors, and environmental stability, to calculate a comprehensive confidence score. Based on this score, at least two initial color schemes with different emphases are generated. If the comprehensive confidence score fails to reach a preset threshold, the system initiates parallel optimization threads to synchronously iteratively optimize the parameters (such as the ratio of primary and secondary colors, brightness and darkness relationships) of the multiple initial color schemes. During this process, each optimization thread works independently and is continuously evaluated. The system ultimately selects the optimal final color scheme based on the game result of the comprehensive confidence scores of each scheme after optimization (such as the highest score winning or weighted fusion), thereby ensuring that the output results achieve a higher standard of balance in semantic accuracy, visual aesthetics, and environmental adaptability.

[0049] In this embodiment, the pre-trained emotion color mapping model employs a three-layer fully connected neural network architecture. The input layer has 128 neurons, perfectly corresponding to the generated 128-dimensional emotion vector; the hidden layer contains 256 neurons, using the ReLU activation function for non-linear transformation; the output layer has 150 neurons, corresponding to the 3D CIELAB color space coordinates of 50 color basis vectors. The CIELAB color space was defined by the International Commission on Illumination (CIE) in 1976 to provide a perceptually uniform color model. The model's forward propagation process is described by the formula... Implementation, where E is the emotion vector input, , This is the weight matrix. , B is the bias term, and B is the set of color basis vectors for the final output.

[0050] During the model training phase, the system uses 100,000 expert-annotated "design semantics-ideal color matching" samples for supervised learning. All color data is uniformly converted to the CIELAB color space and normalized. The training process consists of 1000 training epochs with an initial learning rate of 0.001, which decays to 0.5 times the original rate every 200 epochs. The batch size is set to 256 samples. The Adam optimizer is used. The loss function uses a weighted combination of mean squared error and color distribution similarity. To ensure the model's generalization ability, a five-fold cross-validation strategy is used during training. When the color similarity score on the validation set (calculated using the reciprocal of the CIEDE2000 color difference formula) no longer improves after 50 consecutive rounds, an early stopping mechanism is automatically triggered to terminate training.

[0051] After a full training cycle, the emotion color mapping model demonstrated excellent performance on the test set, achieving a color prediction similarity of 0.89±0.05 and a semantic consistency score of 0.92±0.03. In actual deployment, the model parameters are fixed on the system server, and the time for a single prediction is less than 5 milliseconds (using GPU acceleration), providing reliable technical support for the system's real-time color matching decisions.

[0052] For example, in a practical application on an e-commerce design platform, when the design requirement of "spring flower-themed promotional poster" is input, the system accurately generates a color scheme with pink and green as the main colors through the model. After adaptive correction to ambient light, it is directly applied to the mobile poster design, improving design efficiency and quality consistency.

[0053] Through the above technical solution, this application realizes a complete process from multi-dimensional requirement input to intelligent color generation and personalized adjustment, and then to environmental adaptation and quality closed-loop verification. This application can automatically generate color matching schemes that accurately reflect design intent, fully cater to individual user differences, and maintain visual effects in real physical scenarios, thereby improving the intelligence level, reliability, and user experience of design decisions.

[0054] This application further proposes that the specific steps for converting semantic text data into a pre-defined sentiment vector using a machine learning algorithm include:

[0055] Input the user-input design semantic text data (such as "high-end men's watch packaging, conveying composure and trust") into a pre-trained text sentiment analysis model.

[0056] Text sentiment analysis models perform deep processing on text sequences through embedding layers, converting the text sequence (containing each word or phrase) into a contextual semantic vector, thereby accurately capturing the deep meaning and sentiment of the text.

[0057] Based on the generated contextual semantic vector, a normalized sentiment probability distribution with a predetermined number of digits (e.g., K dimensions) is generated through the fully connected output layer of the text sentiment analysis model, serving as the sentiment vector. Where K is the predetermined number of bits. This K-dimensional vector is the final emotion vector. Each dimension represents the probability of the intensity of a specific emotion or style attribute (such as "high-end," "composed," "trustworthy," etc.). Emotion Vector Used to represent the design sentiments and stylistic tendencies contained in the input text.

[0058] This application combines natural language understanding with numerical representation to achieve precise quantification of design intent, laying a solid and reliable data foundation for the entire color decision-making process. It ensures that the color generation process is highly aligned with design goals from the outset, improving the accuracy and consistency of the final color scheme's semantic expression, thereby enhancing the overall system's intelligence and decision-making reliability.

[0059] This application further proposes that the pre-trained emotion color mapping model is initialized through supervised learning using an expert-annotated sample database. Each sample has a clearly defined correspondence between a specific design semantic (such as a text description) and a set of ideal color combinations considered to best represent that semantic. Through supervised learning on such high-quality samples, the model can initially establish a mapping from emotion vectors... The robust mapping relationship to the set of color basis vectors provides a reliable initial knowledge foundation for the system.

[0060] During system operation, the system implicitly collects users' actual interaction behavior with the final output color scheme, such as directly accepting or rejecting it, or manually fine-tuning certain colors before saving. These actions are converted into implicit feedback tags by the system, along with the emotion vector used when generating the scheme. Together they form a new training sample.

[0061] The system periodically (e.g., weekly or monthly) incrementally learns the emotion color mapping model using a cumulative set of new training samples to dynamically adjust its parameters. During this process, the model's parameters are dynamically adjusted so that its mapping rules are not only based on initial expert knowledge but also incorporate the group preferences and aesthetic trends exhibited by a large number of real users in actual design scenarios. This allows the model to gradually adapt to changes in user aesthetics and even develop more accurate mapping capabilities for specific users or design styles.

[0062] Through the above technical solutions, this application effectively solves the problems of knowledge solidification and insufficient adaptability that may exist in static models, making the color decision system a dynamic organism that can continuously evolve over time and become more intelligent with use, thereby improving the personalization of its long-term services, user satisfaction, and practicality and reliability in different design contexts.

[0063] like Figure 3 As shown, this application further proposes that, based on user visual preference data, the color basis vector set is adjusted with personalized parameters to generate a personalized color set. The steps for adjusting the personalized parameters specifically include:

[0064] User visual preference data is represented as a sensitivity weight vector, which contains four core dimension parameters: brightness sensitivity. Hue sensitivity Saturation sensitivity and contrast preference parameters These parameters are obtained through analysis of user history or explicit preference settings, and are represented by normalized numerical values ​​to indicate the user's sensitivity and preference intensity to various color attributes.

[0065] Based on the sensitivity weight vector, the system calculates the offset of the first color vector using a preset transformation function. The first color vector refers to the color vector in the unadjusted set of color basis vectors output from the emotion color mapping model, used to represent the initial mapped color state. Specifically, for each first color vector in the color basis vector set... (In CIELAB color space) The offset of the first color vector It can be obtained through weighted calculation, and the formula is: ΔL, ΔH, and ΔS represent the basic adjustment amounts in the dimensions of brightness, hue, and saturation, respectively. These basic adjustment amounts are preset based on color psychology models and user experience data.

[0066] The system transforms and redistributes the weights of the first color vector in the color basis vector set based on the offset, generating a new color set. Specifically, the system applies the calculated offset to the corresponding first color vector to achieve color transformation, as shown in the formula: ,in This is the transformed second color vector. The second color vector refers to the color vector in the personalized color set generated after adjusting the personalized parameters of the first color vector based on the user's visual preference data. Simultaneously, the system adjusts the color vector according to the user's contrast preference parameters. The saliency weight coefficients of each color in the color basis vector set. The colors are then redistributed to generate a personalized color set.

[0067] For situations where contrast needs to be emphasized (user contrast preference parameter) (If the value is high), the system will amplify the difference in brightness or saturation between the primary and secondary colors, and strengthen this effect by adjusting the weighting coefficient. The formula is: Where e is the adjustment factor, These are the adjusted weighting coefficients.

[0068] Through the aforementioned transformation and redistribution process, the system ultimately generates a set of color basis vectors that fully reflect the user's personalized visual characteristics. This application establishes a precise mathematical mapping relationship, transforming subjective visual preferences into calculable color adjustment parameters, thus achieving a precise shift from standardized color schemes to truly personalized color schemes. This improves the alignment between the generated scheme and the user's subjective aesthetic needs, enhances the user's sense of identification and satisfaction with the design outcome, and ensures the standardization and reproducibility of the color adjustment process, providing a reliable technical foundation for large-scale personalized design services.

[0069] This application further proposes that, based on scene lighting parameter data, environmental adaptive correction is performed on a personalized color set to generate the corrected color set. The environmental adaptive correction steps specifically include:

[0070] Each second color vector in the personalized color set is converted to a device-independent color space, eliminating the interference of color gamut differences between different output devices on color performance.

[0071] Based on the color temperature and illuminance parameters in the scene lighting parameters, a color adaptation transformation function is used to correct the second color vector, generating a corrected third color vector. The third color vector refers to the color vector in the corrected color set generated after environmental adaptive correction of the second color vector based on the scene lighting parameter data. The core of the correction process lies in constructing a color adaptation transformation function based on the scene lighting parameters. Specifically, the system extracts the key physical quantity from the scene lighting parameters: color temperature. (Unit: K) and illuminance Lux (unit: lx), in this embodiment, color temperature ≥2500K. Color temperature is used to characterize the color properties of a light source, while illuminance reflects light intensity. The correction function can be expressed as:

[0072]

[0073] in The first color vector of the original LAB, This is the corrected third color vector. Transformation matrix. It is a 3×3 linear transformation matrix whose element values ​​are related to color temperature. Functional relationship, used to correct for color perception changes caused by different light source color temperatures; offset vector Primarily related to illuminance (Lux), it is used to compensate for differences in perceived brightness caused by variations in light intensity.

[0074] Through the above technical solution, this application can accurately predict and adjust the visual performance of each color under the target lighting environment, effectively overcome the color distortion problem that occurs under different lighting conditions, improve the conversion fidelity of the color scheme from the digital design environment to the physical application scenario, ensure visual consistency and reliability in the actual implementation process, and provide key technical support for cross-media color quality management.

[0075] This application further proposes to conduct multi-dimensional intelligent evaluation of the corrected color set. The multi-dimensional evaluation in calculating the comprehensive confidence score is achieved by establishing three independent yet complementary scoring dimensions, thus realizing a comprehensive quantitative evaluation of the color scheme quality. Specifically, this includes:

[0076] Based on the corrected color set, the corresponding calculated color sentiment vector is derived in reverse through the emotion color mapping model. It is used to characterize the emotional and stylistic features conveyed by the color scheme itself, and through emotion vectors With color emotion vector Cosine similarity calculation is used to obtain semantic consistency score. Semantic consistency score The formula quantifies the ability of a color scheme to convey intended emotion. ,in Denotes the vector dot product, || || represents the vector magnitude.

[0077] Based on the CIELAB color space distances of the primary, secondary, and accent colors in the corrected color set, a weighted score of internal contrast and harmony is calculated to obtain a visual harmony score. Visual coordination score Used to measure the visual harmony within a color scheme.

[0078] The system is based on the CIELAB color space and calculates the perceived color difference between the primary color, secondary color, and accent color. This perceptual color difference calculation uses the CIEDE2000 formula, as follows:

[0079]

[0080] Where ΔL', ΔC', and ΔH' represent the differences in lightness, chroma, and hue, respectively. , , For the compensation function, , , For parameter factors. Visual coordination score Finally, the weighted average of the color differences is calculated to reflect the contrast and balance between colors.

[0081] Simulate the color performance under at least two typical lighting conditions (such as a D65 standard light source and a Class A incandescent light source) based on scene lighting parameter data, and calculate the standard color difference of the corrected color set under different conditions. The reciprocal of the mean is used to calculate the environmental stability score. The formula is ,in The standard deviation of the color difference values ​​of the color scheme under different lighting conditions is represented by this score, which reflects the environmental robustness of the color scheme.

[0082] The above scores are linearly combined according to preset weights to generate a comprehensive confidence score. ,in , , These are preset weighting coefficients, which are dynamically adjusted based on the application scenario. The weighting coefficients are determined based on AHP (Analytic Hierarchy Process), with a default value of [value missing]. =0.4、 =0.35、 =0.25, with a weighted sum of 1. This setting was validated through simulation: using 1000 design samples from the COCO dataset, compared to traditional tools (such as Adobe Color), this solution improved semantic consistency score by 25% (from 0.65 to 0.81) and visual harmony by 15% (from 0.70 to 0.80). The simulation environment was a standard light source with a color temperature of 6500K and an illuminance of 500lx. In practical applications, these weighting coefficients can be dynamically adjusted between 0.1 and 0.6 according to user type or scenario requirements.

[0083] When the overall confidence score is lower than the preset threshold, the system will perform targeted optimization for the dimensions with lower scores, such as adjusting the main hue when there is semantic inconsistency, and optimizing the color ratio when there is insufficient coordination.

[0084] The preset threshold can be dynamically set according to the application scenario, with a recommended range of 0.6 to 0.9 and an optimal value of 0.7. For example, in high-quality design scenarios, the threshold is set to 0.8; in rapid generation scenarios, the threshold is set to 0.6.

[0085] Meanwhile, the combination of parameters for personalized adjustments and environmental corrections can be based on changes in user type:

[0086] Variation 1 (Professional Designer): The weighting coefficient combination is as follows =0.5 (semantic consistency) =0.3 (visual coordination) =0.2 (environmental stability), emphasizing the precise communication of design intent.

[0087] Variation 2 (Regular User): The weighting coefficient combination is as follows =0.3、 =0.4、 =0.3, emphasizing visual comfort.

[0088] This application establishes a mapping relationship between quantitative evaluation indicators and specific optimization directions through a multi-dimensional evaluation system, enabling scientific evaluation and precise optimization of color scheme quality. It enhances the transparency and reliability of the color decision-making process, ensuring that the final output color scheme achieves optimal balance in semantic expression, visual aesthetics, and environmental adaptability, providing reliable technical support for high-quality design decisions.

[0089] This application further proposes that when the overall confidence score is lower than a preset threshold, the parameters of the initial color scheme are iteratively optimized, and the specific steps include:

[0090] By comparing semantic consistency scores Visual coordination score Environmental stability score The relative magnitudes of the scores across different dimensions are used to intelligently determine the priority direction for parameter adjustments; for example, when comparing semantic consistency scores... If the score is significantly low, prioritize adjusting the primary hue parameter that best matches the emotion vector; if the visual coordination score is low... If there are shortcomings, the focus should be on optimizing the contrast relationship or brightness gradient between the primary and secondary colors; while if the environmental stability score is insufficient... If the result is unsatisfactory, then focus on fine-tuning the color's adaptability parameters under the target lighting conditions.

[0091] After clarifying the optimization direction, a hybrid optimization strategy combining heuristic rules and local search is adopted to fine-tune the parameters of the third color vector. The heuristic rules, based on color theory and design experience, predefine a series of adjustment templates for common problems, such as improving harmony by rotating the hue wheel of equal brightness or enhancing semantic expression by adjusting saturation weights. The local search, within the neighborhood of the "high-quality solution" defined by these heuristic rules, systematically perturbs and evaluates at least one parameter of the third color vector—brightness (L value), saturation (C value), hue (h value), or its relative weight in the color scheme—within a preset step size, seeking the parameter combination that achieves local optimization of the overall confidence score.

[0092] This iterative cycle will continue until the overall confidence score reaches or exceeds a preset quality threshold, or the maximum number of iterations is reached. Parameter adjustments and scoring results from each iteration are recorded for the algorithm's self-learning, improving the efficiency of subsequent optimizations.

[0093] This application constructs a highly efficient closed-loop self-calibration system by transforming macroscopic quality assessment feedback into microscopic parameter adjustment instructions. This improves the reliability and quality stability of color scheme output, ensuring a high balance between semantic expression, visual aesthetics, and adaptability to practical application scenarios in the final scheme. It effectively reduces the cost of repeated manual adjustments and achieves an intelligent and automated upgrade to the design decision-making process.

[0094] This application further proposes that, before outputting the final color scheme after iterative optimization, a learning update is also included, specifically comprising the following steps:

[0095] The system records and tracks user actions regarding the initial color scheme output. Specifically, it continuously records user interactions with the system's output color scheme, including directly accepting the scheme, rejecting the scheme, or manually adjusting and saving specific colors within the scheme. These actions are transformed into semantically meaningful feedback signals: acceptance is considered a positive sample, rejection is considered a negative sample, and adjustments are quantified by calculating the color difference before and after the adjustment to determine the user's preference.

[0096] Based on recorded user feedback data, the emotion color mapping model is incrementally updated. This process employs an online learning algorithm to recombine user feedback data with corresponding emotion vectors into new training samples, adjusting model parameters by minimizing the difference between the predicted output and the user's actual preferences. Specifically, for cases where users manually adjust the colors, the system focuses on learning the differences in color attributes between the adjusted colors and the original output colors. This allows it to directly generate color combinations that better match user expectations when encountering similar emotion vectors in the future.

[0097] This application establishes a direct feedback pathway from user experience to model optimization by making users' implicit aesthetic preferences explicit and transforming them into computable learning samples. This enables the color decision system to continuously improve itself, gradually adapting not only to the evolution of individual user aesthetics but also learning more universal color mapping rules from group usage. This enhances the alignment between the system's output color schemes and user expectations, as well as the system's long-term service intelligence and user satisfaction.

[0098] This application further proposes that, along with outputting the final color scheme after iterative optimization, a structured data package is also included. This data package serves as a complete digital twin of the color scheme, containing not only directly applicable color values ​​but also key information such as the scheme's generation logic, quality assessment, and scene adaptation guidance. The structured data package includes:

[0099] Executable color data, decision-making basis data, and scene adaptation guidance data;

[0100] The executable color data includes a mapping table of device-dependent color value codes and device-independent color value codes for each color in the final color scheme. It provides a precise mapping table between device-independent color spaces (such as CIELAB) and commonly used device-dependent color spaces (such as sRGB and CMYK) for each color in the final color scheme, ensuring the consistency of color reproduction on different output devices and driving different output devices to reproduce colors.

[0101] The decision-making data includes the main dimension labels of the emotion vector corresponding to the final color scheme and the comprehensive confidence score. and its sub-scores for each dimension (semantic consistency score) Visual coordination score and environmental stability score This transparently reveals the design intent and quality basis of the scheme, which is used to explain the generation logic and quality assessment of the final color scheme;

[0102] Scene adaptation guidance data includes color fine-tuning suggestions based on scene lighting parameter data for different output media (such as matte paper, display screen) or lighting environments (such as warm light indoors, sunlight), such as saturation compensation values ​​for specific color temperatures.

[0103] These three data modules are not isolated; executable color data, decision-making basis data, and scene adaptation guidance data are interconnected through preset association rules. This allows downstream applications to filter or compare different color schemes based on the decision-making basis data and automatically adjust the executable color data according to the scene adaptation guidance data. For example, the low environmental stability score in the decision-making basis data... It will automatically associate with the compensation parameters for high color temperature environments in the scene adaptation guidance data; downstream applications can screen and compare multiple solutions based on quantitative indicators according to the decision basis data, and can automatically and accurately fine-tune the executable color data according to the scene adaptation guidance data through scripts or design software plugins, thereby seamlessly integrating color decisions into the actual design workflow.

[0104] The structured data packet output mechanism greatly improves the interpretability, comparability, and programmability of color schemes by expanding color schemes from single color value results to composite data entities rich in semantics, quality assessment, and scene intelligence.

[0105] The following is an application example of an AI-assisted design color matching decision-making method—taking the color design of a children's educational app interface as an example:

[0106] A design team needed to develop a math education app for children aged 6-8. The core design requirement was to "create a safe, lively, and focused learning environment." The system first acquired three types of core data: Design semantic text data, which stated "A children's math education app; the main color scheme should reflect safety (e.g., blue tones) and liveliness (e.g., orange accents), avoiding highly saturated and stimulating colors," was preprocessed to retain the core semantic words "children," "safety," "lively," "blue tones," and "orange accents." User visual preference data was generated by accessing visual preference profiles of children aged 6-8, generating a sensitivity weight vector, including brightness sensitivity. Hue sensitivity Saturation sensitivity Contrast preference parameter Scene lighting parameters are collected from typical usage scenario data via sensors; daytime scenes are measured by color temperature. =6500K (D65 standard white light), illuminance Lux=450lx, color temperature for nighttime scenes. =3000K (warm light desk lamp), illuminance Lux=150lx.

[0107] A pre-trained BERT (Bidirectional Encoder-Representation Transformer) model is used to process semantic text, outputting a 128-dimensional sentiment vector. The core dimensions are: safety (0.82), liveliness (0.75), focus (0.68), and stimulation (0.21). This emotion vector is input into the pre-trained emotion color mapping model (a three-layer fully connected neural network) of the color mapping module, and mapped to five CIELAB color basis vector sets: [(L=75, a=-30, b=-20), (L=85, a=25, b=60), (L=90, a=0, b=0), (L=60, a=-10, b=-15), (L=70, a=10, b=30)], corresponding to light blue as the primary color, light orange as the accent color, white as the background color, dark blue as the text color, and light yellow as the emphasis color, respectively.

[0108] The offset ΔCᵢ is calculated based on the sensitivity weight vector, and the basis vector is fine-tuned: the light blue primary color ΔL=+5, Δa=+2, is adjusted to (L=80, a=-28, b=-20); the light orange accent color ΔS=-0.1 is adjusted to (L=85, a=22, b=55). A personalized color set is generated and converted to the CIELAB device-independent space. Based on the lighting parameters, the light blue primary color after correction is (L=82, a=-29, b=-22) for daytime scenes and (L=78, a=-27, b=-18) for nighttime scenes.

[0109] Semantic consistency score Through emotion vectors With color emotion vector The cosine similarity is calculated using the following formula: After substituting the values, we get Visual coordination score Based on the CIELAB distance calculation of primary and secondary colors, we get =0.88; Environmental stability score ,in =0.85, therefore =0.85. Weight =0.4、 =0.35、 =0.25, calculated as follows The value is higher than the preset threshold of 0.85, so no iterative optimization is needed.

[0110] The system outputs the final color scheme after iterative optimization, along with a structured data package. This package contains executable color data (e.g., light blue primary color sRGB:#E6F2FF, CMYK:15,5,0,0), decision-making data (emotional vector main dimension labels "Safety: 0.82" and "Lively: 0.75", with a comprehensive score of 0.90), and scene adaptation guidance data (recommending a 5% reduction in light orange saturation for nighttime scenes). In practical applications, the design cycle has been shortened from 2-3 days to 2 hours. Testing with children shows that visual comfort scores have improved from 3.2 / 5 to 4.6 / 5, and color deviation ΔE < 5 under different lighting conditions, solving the problem of poor scene adaptability in traditional solutions.

[0111] Through the above embodiments, this application realizes end-to-end intelligent decision-making from abstract design requirements to implementation solutions, effectively improving the semantic accuracy, scenario robustness and user satisfaction of color schemes.

[0112] Through the above technical solutions, this application not only provides designers with a basis for a deeper understanding of the solutions, but also provides a solid data foundation for realizing cross-platform and cross-media automated color management and precise control, enhancing the practical value and integration capability of the entire color decision-making system in complex and dynamic design environments.

[0113] Example 2:

[0114] like Figure 4 As shown, the AI-assisted design color matching decision system uses the aforementioned AI-assisted design color matching decision method, including:

[0115] The data acquisition module is used to acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data;

[0116] The feature extraction module is used to convert semantic text data into pre-numbered emotion vectors using machine learning algorithms, and to map the emotion vectors into a set of color basis vectors using a pre-trained emotion color mapping model.

[0117] The personalized adjustment module is used to adjust the parameters of the color basis vector set based on user visual preference data to generate a personalized color set.

[0118] The environment adaptation module is used to perform environmental adaptive correction on the personalized color set based on scene lighting parameter data, and generate the corrected color set.

[0119] The evaluation module is used to perform multi-dimensional intelligent evaluation of the corrected color set, calculate the comprehensive confidence score, and generate an initial color scheme based on the corrected color set.

[0120] The optimization iteration module is used to start a parallel optimization thread when the overall confidence score is lower than a preset threshold, to iteratively optimize the parameters of at least two initial color schemes, and select the final color scheme based on the game result of the corresponding overall confidence scores.

[0121] The output module is used to output the final color scheme after iterative optimization.

[0122] The hardware architecture of this application consists of a data acquisition layer, a computing processing layer, and an output layer. The data acquisition layer includes a user interaction terminal (such as a computer or tablet computer) and an optional integrated light sensor; the computing processing layer is implemented by a server system equipped with a GPU accelerator; and the output layer includes a color calibration display and a network transmission interface.

[0123] The data acquisition module receives design semantic text input through a graphical interface provided by the user interaction terminal and retrieves pre-stored visual preference data from the user configuration archive. When physical environment adaptation is required, this module can automatically collect scene lighting parameters through a connected Bluetooth / Wi-Fi light sensor, or receive environmental parameters manually input by the user.

[0124] The feature extraction module is deployed on a computing server and loaded with a pre-trained natural language processing model (such as a BERT variant), specifically responsible for converting the input text description into a 128-dimensional emotion vector. The color mapping module integrates an emotion color mapping model (using a fully connected neural network structure) trained on a large number of design samples, mapping the emotion vector into a set of color basis vectors containing 5-8 primary colors.

[0125] The personalized adjustment module reads the user's visual preference configuration file (containing parameters such as brightness sensitivity and hue preference) and uses a linear transformation algorithm based on the weight matrix to perform personalized correction on the basis vector set. The environment adaptation module integrates the CIECAM02 color appearance model, calculates the visual performance of colors under different lighting conditions in real time based on lighting parameters, and performs color adaptation transformations. The personalized adjustment module supports multiple variations to adapt to different needs:

[0126] Variation A (High Contrast Preference): In user visual preference data, the contrast preference parameter =0.9, the system automatically increases the difference in brightness between primary and secondary colors, for example, adjusting the brightness difference of the primary color vector from ΔL=10 to ΔL=20.

[0127] Variant B (Low-light environment correction): In adaptive environmental correction, the color temperature is adjusted accordingly. In a warm light environment of 3000K, the color adaptation transformation function increases the saturation compensation value by +5% to ensure that colors are not distorted under low light conditions. These variations are implemented through heuristic rules, such as the rule corresponding to variation A, 'If...' Then the brightness difference is linearly amplified.

[0128] The evaluation module employs a multi-task neural network architecture to simultaneously calculate the semantic consistency, visual harmony, and environmental stability scores of the color scheme, and generates a comprehensive confidence score through weighted fusion. The optimization iteration module implements a policy gradient-based reinforcement learning algorithm, automatically adjusting parallel optimization threads when the comprehensive confidence score falls below a preset threshold. The output module ultimately generates a color scheme containing standard color values ​​(PANTONE, sRGB, CMYK) and pushes the results to design software plugins or mobile applications via a RESTful API.

[0129] In one embodiment, when a user inputs a design requirement for "summer refreshing beverage packaging" on a tablet, the system automatically calls the user's preferred high-saturation configuration and detects the ambient light temperature as 5500K using the tablet's camera. The system generates a color scheme primarily based on blue-green tones on the server side, performs environmental adaptation correction, and pushes it to the user interface, providing both printing and screen display color specifications. This application, through a hardware and software co-engineering architecture, achieves fully automated processing from design intent input to multi-scenario adapted output, improving the efficiency and cross-media consistency of color design, and providing the design industry with an end-to-end intelligent color solution.

[0130] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. An AI-assisted color matching decision-making method for design, characterized by: Includes the following steps: Acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data; The designed semantic text data is converted into a pre-defined emotion vector using a machine learning algorithm, and the emotion vector is mapped to a set of color basis vectors using a pre-trained emotion color mapping model. Based on the user's visual preference data, the color basis vector set is adjusted with personalized parameters to generate a personalized color set; Based on the scene lighting parameter data, the personalized color set is subjected to environmental adaptive correction to generate a corrected color set; The corrected color set is subjected to multi-dimensional intelligent evaluation, a comprehensive confidence score is calculated, and an initial color scheme is generated based on the corrected color set. When the overall confidence score is lower than a preset threshold, a parallel optimization thread is started to iteratively optimize the parameters of at least two initial color schemes, and the final color scheme is selected based on the game result of the corresponding overall confidence scores. Output the final color scheme after iterative optimization.

2. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: The specific steps for converting the designed semantic text data into a pre-defined sentiment vector using a machine learning algorithm include: The designed semantic text data is input into a pre-trained text sentiment analysis model; The text sentiment analysis model converts text sequences into contextual semantic vectors through an embedding layer; Based on the context semantic vector, a K-dimensional, normalized sentiment probability distribution is generated through the fully connected output layer of the text sentiment analysis model, which serves as the sentiment vector; where K is the predetermined number of bits.

3. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: The pre-trained emotion color mapping model specifically includes: Supervised learning initialization is performed using an expert-annotated sample database; The user's actual interaction with the final color scheme is used as an implicit feedback label, which, together with the corresponding emotion vector, forms a new training sample. The emotion color mapping model is incrementally learned periodically using accumulated new training samples to dynamically adjust the model parameters.

4. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: The personalized color set is generated by adjusting the personalized parameters of the color basis vector set based on the user's visual preference data. The steps of adjusting the personalized parameters specifically include: The user visual preference data is represented as a sensitivity weight vector, which includes brightness sensitivity, hue sensitivity, saturation sensitivity and contrast preference parameters; The offset of each first color vector in the color basis vector set is calculated based on the sensitivity weight vector; The personalized color set is generated by transforming and redistributing the weights of the first color vector based on the offset.

5. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: Based on the scene lighting parameter data, environmental adaptive correction is performed on the personalized color set to generate a corrected color set. The environmental adaptive correction steps specifically include: Convert the second color vector in the personalized color set to a device-independent color space; Based on the color temperature and illuminance parameters in the scene lighting parameters, the second color vector is corrected using a color adaptation transformation function.

6. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: The corrected color set is subjected to multi-dimensional intelligent evaluation, and the multi-dimensional evaluation in calculating the comprehensive confidence score includes: Based on the corrected color set, a color emotion vector is calculated using the emotion color mapping model. And through the emotion vector With the aforementioned color emotion vector Perform cosine similarity calculation to generate a semantic consistency score. The formula is ; Based on the CIELAB color space distances of the primary, secondary, and accent colors in the corrected color set, a weighted score for contrast and harmony within the CIELAB color space distance is calculated to obtain a visual harmony score. ; Based on the scene lighting parameter data, simulate the color performance under at least two typical lighting conditions, and calculate the standard color difference of the corrected color set under different conditions. The reciprocal of the mean is used to calculate the environmental stability score. The formula is ,in This represents the standard deviation of the color difference values ​​of the initial color scheme under different lighting conditions; The above scores are linearly combined according to preset weights to generate the comprehensive confidence score. ,in , , The preset weighting coefficients, and .

7. The artificial intelligence-assisted design color matching decision-making method according to claim 6, characterized in that: When the overall confidence score is lower than a preset threshold, the method further includes iterative optimization of the parameters of the initial color scheme, specifically including: Based on the semantic consistency score The visual coordination score and the environmental stability score The relative magnitudes of the scores across each dimension determine the priority adjustment direction; The third color vector parameters of the corrected color set are fine-tuned by combining heuristic rules with local search. The fine-tuning includes adjusting at least one of the brightness, saturation, hue, or weight of the third color vector.

8. The artificial intelligence-assisted design color matching decision-making method according to claim 7, characterized in that: Before outputting the final color scheme after iterative optimization, the process includes learning and updating, specifically the following steps: Record and track the user's acceptance or adjustment of the output initial color scheme; The emotion color mapping model is incrementally updated based on the recorded user feedback data.

9. The artificial intelligence-assisted design color matching decision-making method according to claim 1, characterized in that: In addition to outputting the final color scheme after iterative optimization, the system also outputs a structured data packet, which includes: Executable color data, decision-making basis data, and scene adaptation guidance data; The executable color data includes a mapping table of device-dependent color value codes and device-independent color value codes for each color in the final color scheme; The decision-making data includes the main dimension labels of the emotion vector corresponding to the final color scheme, the comprehensive confidence score and its sub-scores for each dimension; The scene adaptation guidance data includes color fine-tuning suggestions based on the scene lighting parameter data, generated for different output media or lighting environments. The executable color data, the decision-making basis data, and the scene adaptation guidance data are interconnected through preset association rules.

10. An AI-assisted color matching decision-making system for design, characterized by: The artificial intelligence-assisted design color matching decision-making method as described in any one of claims 1 to 9 includes: The data acquisition module is used to acquire user-input design semantic text data, user visual preference data, and scene lighting parameter data; The feature extraction module is used to convert the designed semantic text data into a pre-defined emotion vector using a machine learning algorithm, and to map the emotion vector into a set of color basis vectors using a pre-trained emotion color mapping model. The personalization adjustment module is used to adjust the personalized parameters of the color basis vector set based on the user's visual preference data to generate a personalized color set. An environment adaptation module is used to perform environment adaptive correction on the personalized color set based on the scene lighting parameter data, and generate a corrected color set. The evaluation module is used to perform multi-dimensional intelligent evaluation on the corrected color set, calculate the comprehensive confidence score, and generate an initial color scheme based on the corrected color set. The optimization iteration module is used to start a parallel optimization thread when the overall confidence score is lower than a preset threshold, to iteratively optimize the parameters of at least two initial color schemes, and to select the final color scheme based on the game result of the corresponding overall confidence scores. The output module is used to output the final color scheme after iterative optimization.

Images