A method for intelligent generation and visual optimization of architectural interior decoration patterns

Abstract

The application discloses a kind of building interior decoration pattern intelligent generation and visual optimization method, it is related to building interior decoration pattern design technical field, collection building interior space three-dimensional structure, material, illumination parameter establishes digitization three-dimensional model, collects user demand and constructs feature dataset, generates feature coding vector by multi-scale feature extraction, inputs pre-training generative adversarial network and generates initial decoration pattern, color coordination, stereoscopic visual effect, texture detail optimization are sequentially carried out, after optimization pattern is mapped to three-dimensional model full visual angle rendering, local adjustment generates final pattern file and outputs parameter and guiding information suitable for different construction technology.The application can generate personalized decoration pattern in line with user demand, improve the adaptability and visual effect of pattern and space, pre-verify application effect, output file suitable for multiple construction technology, improve the efficiency and quality of design and construction.

Description

Technical Field

[0001] This invention relates to the field of architectural interior decoration pattern design technology, and in particular to an intelligent generation and visual optimization method for architectural interior decoration patterns. Background Technology

[0002] Traditional architectural interior decoration pattern design relies heavily on the designer's personal experience and aesthetic sense. The design process involves multiple hand-drawn revisions and adjustments, is time-consuming, and the final result is significantly influenced by the designer's subjective factors. The distinct styles of different designers make it difficult to meet the diverse and personalized needs of users. Furthermore, traditional design methods struggle to fully consider objective factors such as the three-dimensional structure, material properties, and lighting conditions of the architectural interior space. This often results in patterns that clash with the spatial environment and produce poor visual effects in practical applications. For example, patterns applied to large walls may appear too rough upon close inspection, while patterns applied to small spaces may create a sense of oppression due to excessive density of elements. In addition, patterns generated by traditional design methods are mostly two-dimensional, lacking three-dimensional visual expression and failing to create a rich sense of spatial layering and artistic atmosphere.

[0003] With the development of computer and artificial intelligence technologies, digital design methods are increasingly being applied to the field of architectural interior decoration pattern design. Existing digital design tools enable rapid pattern drawing and modification, improving design efficiency, but they still have shortcomings in intelligent pattern generation and visual effect optimization. Most existing tools only provide basic pattern editing functions and cannot automatically generate personalized decorative patterns based on user style preferences, cultural needs, and spatial characteristics. Some pattern generation methods based on generative adversarial networks can generate patterns with specific styles, but the generated patterns often suffer from inaccurate style transfer, content mismatch with requirements, and difficulty in effectively adapting to the structure, materials, and lighting conditions of the architectural interior space. For example, the colors of the generated patterns may clash with the colors of the interior wall materials, and significant color deviations may occur under different lighting conditions.

[0004] Existing technologies also present numerous challenges in adapting decorative patterns to construction processes. Most design tools generate pattern files containing only visual information, lacking parameters and guidance specific to different construction techniques. Construction workers must rely on experience to convert design patterns into construction parameters, a process prone to errors such as dimensional inaccuracies, color deviations, and inaccurate positioning, impacting construction quality. Different construction techniques have varying requirements for pattern resolution, format, and color modes; existing tools struggle to automatically generate pattern files compatible with multiple techniques, increasing pre-construction preparation. Furthermore, current design methods fail to pre-verify the full-view visual effect of patterns in actual space, making it impossible to promptly detect and correct visual distortions and color deviations from different perspectives. This necessitates secondary modifications after construction, extending the construction cycle. Summary of the Invention

[0005] This invention proposes an intelligent generation and visual optimization method for architectural interior decoration patterns to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for intelligent generation and visual optimization of architectural interior decorative patterns includes the following steps:

[0008] Collect three-dimensional structural parameters of the building's interior space, wall and floor material parameters, natural light intensity and angle parameters, and artificial lighting distribution parameters to establish a digital three-dimensional model of the building's interior space;

[0009] Collect user-provided style reference images, color preference data, cultural element requirements, and functional zoning requirements to construct a user requirement feature dataset.

[0010] Multi-scale feature extraction is performed on the user demand feature dataset, extracting style features, color features, cultural element features and functional adaptation features respectively, and generating feature encoding vectors;

[0011] The feature encoding vector is input into a pre-trained generative adversarial network model to generate an initial decorative pattern, which includes a basic texture, color distribution, and element layout.

[0012] The initial decorative pattern is optimized for color coordination by adjusting the hue, saturation, and brightness parameters of each color in the pattern to match the overall color scheme of the building's interior space.

[0013] The initial decorative pattern is enhanced with a stereoscopic visual effect by simulating light and shadow from different angles to generate a depth-perceived visual effect on the surface of the two-dimensional pattern.

[0014] The initial decorative pattern is optimized in terms of texture details, increasing the texture layers and quality of the pattern, so that the pattern visually echoes the materials of the building's interior walls and floors;

[0015] The optimized decorative patterns are mapped onto the digital 3D model of the building's interior space for full-view visual rendering.

[0016] Based on the rendering results, local adjustments are made to the decorative pattern to correct visual distortion and color deviation of the pattern under different viewing angles;

[0017] Generate the final decorative pattern file and output pattern parameters and construction guidance information adapted to different construction techniques.

[0018] Furthermore, it also includes an interactive pattern iterative optimization step based on real-time user feedback. This step receives local modification instructions from the user for the generated pattern, identifies the user's modified area and content, converts the modified content into feature adjustment parameters, inputs them into a generative adversarial network model for local regeneration, generates the modified decorative pattern, and performs visual effect optimization and spatial adaptation verification again.

[0019] Furthermore, it also includes a pattern style transfer and fusion step based on historical generated data. This step retrieves multiple reference patterns with the highest similarity to user needs from the historically generated decorative pattern database, extracts the style features and element features of the reference patterns, and weights and fuses the features of multiple reference patterns to generate a fused feature vector. This vector is then input into a generative adversarial network model to generate decorative patterns with multiple style fusion effects.

[0020] Furthermore, when collecting the three-dimensional structural parameters of the building's interior space, a lidar scanning device is used to perform a full-range scan of the building's interior space to obtain point cloud data of the space. The point cloud data is then denoised, registered, and meshed to generate a high-precision three-dimensional mesh model. At the same time, the dimensions, shapes, and positions of the walls, floors, and ceilings, as well as the positions and dimensions of building components such as doors, windows, beams, and columns, are recorded.

[0021] Furthermore, when performing multi-scale feature extraction on the user demand feature dataset, a convolutional neural network is used to extract low-level visual features and high-level semantic features from the style reference image. Low-level visual features include edge, texture, and shape features, while high-level semantic features include style type, composition method, and element combination features. An autoencoder is used to encode the color preference data to generate a color feature vector. Word embedding technology is used to extract text features from cultural element requirements and functional zoning requirements to generate a text feature vector. All feature vectors are concatenated and normalized to generate a unified feature encoding vector.

[0022] Furthermore, when the feature encoding vector is input into the pre-trained generative adversarial network (GAN) model to generate the initial decorative pattern, the GAN model contains two sub-networks: a generator and a discriminator. The generator receives the feature encoding vector and outputs the initial decorative pattern, while the discriminator receives the generated pattern and real decorative pattern samples and outputs the discrimination result. During training, the model parameters are optimized by minimizing the generator loss function and maximizing the discriminator loss function. The generator loss function is calculated as follows:

[0023] ;

[0024] This represents the total loss value of the generator. Represents the style loss value. Indicates the content loss value. This represents the spatial adaptation loss value. The weighting coefficients representing style loss. The weighting coefficients representing content loss. The weighting coefficients represent the spatial adaptation loss.

[0025] Furthermore, when optimizing the color coordination of the initial decorative pattern, the initial decorative pattern is converted from the RGB color space to the Lab color space. In the Lab color space, the brightness value, a channel value, and b channel value of each color in the pattern are calculated. The parameter range of the main color and auxiliary color is determined according to the overall color tone of the building's interior space. The parameters of each color in the pattern are adjusted to fall within the corresponding parameter range. At the same time, the color difference between adjacent colors is calculated to ensure that the color difference between adjacent colors is within the visually comfortable range.

[0026] Furthermore, when enhancing the stereoscopic visual effect of the initial decorative pattern, the edge contours and texture features of the initial decorative pattern are first extracted. A virtual depth value is assigned to each pixel based on the importance of the feature, with the virtual depth value ranging from 0.1 meters to 1.0 meters. Then, the light intensity and shadow intensity of each pixel are calculated based on the lighting parameters of the building's interior space. The lighting and shadow effects are simulated by adjusting the brightness and contrast of the pixels. The intensity of the stereoscopic visual effect is controlled by the parallax value, which is calculated as follows:

[0027] ;

[0028] Represents pixels disparity value at that location, Indicates the pixel density of the display device. This represents the equivalent focal length of the human eye. The interpupillary distance of the human eye. Represents pixels The corresponding virtual depth value, Represents pixels The angle between the surface normal at that point and the direction of the line of sight.

[0029] Furthermore, when mapping the optimized decorative patterns onto the digital 3D model of the building's interior space for full-view visual effect rendering, the application position and size of the decorative patterns in the 3D model are first determined. The 2D decorative patterns are then texture-mapped to fit onto the corresponding surface of the 3D model. The position and angle of the virtual camera are then set to render the 3D model from multiple different perspectives, generating full-view visual effect images. At the same time, natural lighting at different times and artificial lighting in different modes are simulated to generate visual effect images under different lighting conditions.

[0030] Furthermore, when generating the final decorative pattern file and outputting pattern parameters and construction guidance information adapted to different construction processes, different resolutions and formats of pattern files are generated according to different construction processes. For printing construction processes, a TIFF format file in CMYK color mode is generated; for engraving construction processes, a vector format SVG file is generated; and for spraying construction processes, a PDF file with color separation information is generated. At the same time, the size parameters, color parameters, positioning parameters, and splicing parameters of the pattern are output, as well as the construction steps, precautions, and quality acceptance standards for different construction processes.

[0031] Compared with existing technologies, the beneficial effects of this invention are:

[0032] This invention establishes a digital 3D model by collecting multi-dimensional parameters such as the three-dimensional structure, material, and lighting of the building's interior space. This allows for the comprehensive and accurate acquisition of objective information about the interior space, providing a reliable data foundation for the generation and optimization of decorative patterns. The generated decorative patterns can match the structural characteristics of the building's interior space, adapting to walls, floors, and ceilings of different sizes and shapes, avoiding problems such as pattern stretching, deformation, or proportional imbalance.

[0033] This invention extracts features at multiple scales from a user demand feature dataset, comprehensively capturing users' style preferences, color requirements, cultural elements, and functional needs to generate decorative patterns that meet their personalized requirements. Employing a pre-trained generative adversarial network model for pattern generation ensures the artistic quality and stylistic consistency of the patterns, while also enabling the fusion and transfer of multiple styles to satisfy diverse user aesthetic needs.

[0034] This invention comprehensively enhances the visual effect of decorative patterns through steps such as color coordination optimization, stereoscopic visual effect enhancement, and texture detail optimization. Color coordination optimization ensures that the pattern colors harmonize with the overall color scheme of the interior space, avoiding color clashes. Stereoscopic visual effect enhancement generates a depth-perceived visual effect on the two-dimensional pattern surface, enriching the spatial hierarchy and artistic atmosphere. Texture detail optimization creates a visual echo between the pattern and the interior wall and floor materials, improving the unity of the overall decorative effect.

[0035] This invention maps optimized decorative patterns onto a digital 3D model for full-view visual rendering, enabling pre-verification of the pattern's application effect in actual space. It also allows for timely detection and correction of visual distortion and color deviations from different perspectives, ensuring visual consistency in practical applications. The generated final pattern file is adaptable to various construction techniques and outputs corresponding construction parameters and guidance information, reducing manual conversion steps during construction and improving efficiency and quality. Attached Figure Description

[0036] Figure 1 A flowchart illustrating the overall logic of an intelligent system for generating interior decorative patterns for buildings.

[0037] Figure 2 Flowchart for multi-dimensional data collection and grid construction of indoor spaces;

[0038] Figure 3 A graph for multi-scale user feature analysis and collaborative training of GAN models;

[0039] Figure 4 Flowchart for optimizing color harmony and stereoscopic vision;

[0040] Figure 5 A flowchart for interactive iterative optimization and construction process document generation. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0043] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0044] Reference Figures 1 to 5 A method for intelligent generation and visual optimization of architectural interior decoration patterns, comprising the following steps:

[0045] Collect three-dimensional structural parameters of the building's interior space, wall and floor material parameters, natural light intensity and angle parameters, and artificial lighting distribution parameters to establish a digital three-dimensional model of the building's interior space;

[0046] Collect user-provided style reference images, color preference data, cultural element requirements, and functional zoning requirements to construct a user requirement feature dataset.

[0047] Multi-scale feature extraction is performed on the user demand feature dataset, extracting style features, color features, cultural element features and functional adaptation features respectively, and generating feature encoding vectors;

[0048] The feature encoding vector is input into a pre-trained generative adversarial network model to generate an initial decorative pattern, which includes a basic texture, color distribution, and element layout.

[0049] The initial decorative pattern is optimized for color coordination by adjusting the hue, saturation, and brightness parameters of each color in the pattern to match the overall color scheme of the building's interior space.

[0050] The initial decorative pattern is enhanced with a stereoscopic visual effect by simulating light and shadow from different angles to generate a depth-perceived visual effect on the surface of the two-dimensional pattern.

[0051] The initial decorative pattern is optimized in terms of texture details, increasing the texture layers and quality of the pattern, so that the pattern visually echoes the materials of the building's interior walls and floors;

[0052] The optimized decorative pattern is mapped onto the digital 3D model of the building's interior space for full-view visual rendering. Based on the rendering results, the decorative pattern is locally adjusted to correct visual distortion and color deviation under different viewpoints.

[0053] Generate the final decorative pattern file and output pattern parameters and construction guidance information adapted to different construction techniques.

[0054] The present invention also includes an interactive pattern iterative optimization step based on real-time user feedback. This step receives the user's local modification instructions for the generated pattern, identifies the user's modification area and modification content, converts the modification content into feature adjustment parameters, inputs them into a generative adversarial network model for local regeneration, generates the modified decorative pattern, and performs visual effect optimization and spatial adaptation verification again.

[0055] The present invention also includes a pattern style transfer and fusion step based on historical generated data. This step retrieves multiple reference patterns with the highest similarity to user needs from the historically generated decorative pattern database, extracts the style features and element features of the reference patterns, weights and fuses the features of multiple reference patterns to generate a fused feature vector, and inputs it into a generative adversarial network model to generate a decorative pattern with multiple style fusion effects.

[0056] In this invention, when collecting three-dimensional structural parameters of the building's interior space, a lidar scanning device is used to perform a full-range scan of the building's interior space to obtain point cloud data of the space. The point cloud data is then denoised, registered, and meshed to generate a high-precision three-dimensional mesh model. At the same time, the dimensions, shapes, and positions of the walls, floors, and ceilings, as well as the positions and dimensions of building components such as doors, windows, beams, and columns, are recorded.

[0057] In this invention, when performing multi-scale feature extraction on the user demand feature dataset, a convolutional neural network is used to extract low-level visual features and high-level semantic features from the style reference image. Low-level visual features include edge, texture, and shape features, while high-level semantic features include style type, composition method, and element combination features. An autoencoder is used to encode color preference data to generate color feature vectors. Word embedding technology is used to extract text features from cultural element requirements and functional zoning requirements to generate text feature vectors. All feature vectors are concatenated and normalized to generate a unified feature encoding vector.

[0058] In this invention, when the feature encoding vector is input into a pre-trained generative adversarial network (GAN) model to generate an initial decorative pattern, the GAN model comprises two sub-networks: a generator and a discriminator. The generator receives the feature encoding vector and outputs the initial decorative pattern, while the discriminator receives samples of the generated pattern and real decorative patterns and outputs a discrimination result. During training, the model parameters are optimized by minimizing the generator loss function and maximizing the discriminator loss function. The generator loss function is calculated as follows:

[0059] ;

[0060] This represents the total loss value of the generator. This represents the style loss value, used to measure the style similarity between the generated pattern and the target style reference image. This represents the content loss value, used to measure the consistency between the generated pattern and the content features requested by the user. This represents the spatial adaptation loss value, used to measure the degree of matching between the generated pattern and the architectural interior space parameters. The weighting coefficients representing style loss range from 0.4 to 0.6. The weighting coefficient representing content loss ranges from 0.2 to 0.4. The weighting coefficients representing spatial adaptation loss range from 0.1 to 0.3. By adjusting the weighting coefficients of each loss term, the style expression, content accuracy, and spatial adaptability of the pattern can be balanced.

[0061] In this invention, when optimizing the color coordination of the initial decorative pattern, the initial decorative pattern is converted from the RGB color space to the Lab color space. In the Lab color space, the brightness value, a-channel value, and b-channel value of each color in the pattern are calculated. The parameter range of the main color and auxiliary color is determined according to the overall color tone of the building's interior space. The parameters of each color in the pattern are adjusted to fall within the corresponding parameter range. At the same time, the color difference between adjacent colors is calculated so that the color difference between adjacent colors is within the visually comfortable range.

[0062] In this invention, when enhancing the stereoscopic visual effect of the initial decorative pattern, the edge contour and texture features of the initial decorative pattern are first extracted. A virtual depth value is assigned to each pixel based on the importance of the features, with the virtual depth value ranging from 0.1 meters to 1.0 meters. Then, the light intensity and shadow intensity of each pixel are calculated based on the lighting parameters of the building's interior space. The lighting and shadow effects are simulated by adjusting the brightness and contrast of the pixels. The intensity of the stereoscopic visual effect is controlled by a parallax value, which is calculated as follows:

[0063] ;

[0064] Represents pixels The disparity value at that location, in pixels. This indicates the pixel density of the display device, measured in pixels per millimeter. This represents the equivalent focal length of the human eye, and is taken as 17 millimeters. This represents the interpupillary distance of the human eye, with a value of 65 millimeters. Represents pixels The corresponding virtual depth value, in meters. Represents pixels The angle between the surface normal and the viewing direction is measured in radians. By calculating the disparity value of each pixel and adjusting the horizontal offset of the pixels, a slight difference is created between the patterns seen by the left and right eyes, thus producing a stereoscopic visual effect.

[0065] In this invention, when mapping the optimized decorative pattern onto a digital 3D model of the building's interior space for full-view visual effect rendering, the application position and size of the decorative pattern in the 3D model are first determined. The 2D decorative pattern is then texture-mapped to fit onto the corresponding surface of the 3D model. The position and angle of the virtual camera are then set, and the 3D model is rendered from multiple different perspectives to generate a full-view visual effect image. At the same time, natural lighting at different times and artificial lighting in different modes are simulated to generate visual effect images under different lighting conditions.

[0066] In this invention, when generating the final decorative pattern file and outputting pattern parameters and construction guidance information adapted to different construction processes, pattern files with different resolutions and formats are generated according to different construction processes. For printing construction processes, a TIFF format file in CMYK color mode is generated; for engraving construction processes, a vector format SVG file is generated; and for spraying construction processes, a PDF file with color separation information is generated. At the same time, the size parameters, color parameters, positioning parameters, and splicing parameters of the pattern are output, as well as construction steps, precautions, and quality acceptance standards for different construction processes.

[0067] The present invention will be further illustrated below through two embodiments:

[0068] Example 1

[0069] This embodiment is applied to the design and generation of decorative patterns for residential living room walls. First, architectural interior space parameters are collected. A lidar scanning device is used to perform a 360-degree scan of the living room, covering all walls, floors, ceilings, and architectural components, acquiring raw point cloud data of the space. The raw point cloud data is preprocessed to remove noise and outliers generated during the scanning process. An iterative nearest-point algorithm is used to register the multi-view scan data, generating a complete spatial point cloud model. The registered point cloud model is then meshed to generate a high-precision 3D mesh model. Simultaneously, the length, width, and height dimensions of the living room, as well as the location, shape, and size information of architectural components such as the TV background wall, sofa background wall, windows, doors, beams, and columns, are recorded. Material parameters for the walls, floors, and ceilings are collected. The walls are made of matte latex paint, the floors of solid wood flooring, and the ceilings of plasterboard. Natural lighting parameters are collected, recording the orientation and size of the living room windows, as well as the intensity and angle of natural light at different times. Artificial lighting parameters are collected, recording the location, quantity, and lighting parameters of ceiling lights, downlights, and light strips. Based on all the above parameters, a digital 3D model of the living room is created.

[0070] The process involves collecting users' decorating needs. Users provided three reference images of Scandinavian-style interior design, specifying color preferences such as light blue, off-white, and light gray. Their cultural element requirements included simple geometric elements and natural textures. Functionally, the living room was designated for daily relaxation and entertaining, and the overall decorative style needed to be simple and bright, avoiding overly complex elements. The user-provided reference images, color preference data, cultural element requirements, and functional zoning requirements were integrated to construct a user requirement feature dataset.

[0071] Multi-scale feature extraction is performed on the user demand feature dataset. A convolutional neural network is used to extract features from style reference images, sequentially extracting low-level visual features such as edges, textures, and shapes, as well as high-level semantic features such as style type, composition, and element combinations. An autoencoder is used to encode user color preference data, converting discrete color samples into continuous color feature vectors. Word embedding technology is used to extract features from text content related to cultural element requirements and functional zoning requirements, converting text information into numerical text feature vectors. The extracted visual feature vectors, color feature vectors, and text feature vectors are concatenated and then normalized to generate a feature encoding vector with a unified dimension.

[0072] The feature encoding vector is input into a pre-trained generative adversarial network (GAN) model to generate an initial decorative pattern. The generator of the GAN model receives the feature encoding vector and, through multiple convolutional and deconvolutional operations, outputs an initial decorative pattern with a resolution of 4096×4096. The initial decorative pattern includes basic geometric line textures, a gradient color distribution of light blue and off-white, and a symmetrical element layout. A discriminator distinguishes the generated initial pattern from authentic Nordic-style decorative pattern samples and outputs the discrimination result. During training, the model continuously optimizes its parameters to improve the quality of the generated patterns.

[0073] The initial decorative pattern underwent color coordination optimization by converting it from the RGB color space to the Lab color space. In the Lab color space, the brightness, a-channel value, and b-channel value of each pixel in the pattern were calculated. Based on the overall color scheme of the living room, the primary color was determined to be off-white, with light blue and light gray as secondary colors. The parameter ranges for the brightness, a-channel, and b-channel values ​​of the primary and secondary colors were set respectively. The parameters of each color in the pattern were adjusted to ensure all colors fall within their corresponding parameter ranges. Simultaneously, the color difference between adjacent colors in the pattern was calculated to ensure that the color difference between adjacent colors is within a visually comfortable range.

[0074] To enhance the stereoscopic visual effect of the initial decorative pattern, the edge contours and texture features of the initial decorative pattern are first extracted. A virtual depth value is assigned to each pixel based on the importance of the feature, with the virtual depth value ranging from 0.1 meters to 1.0 meter. Higher virtual depth values ​​are assigned to the edges of geometric lines, and lower virtual depth values ​​are assigned to the background of the pattern. Based on the lighting parameters of the living room, the light intensity and shadow intensity of each pixel under natural and artificial lighting are calculated. By adjusting the brightness and contrast of the pixels, realistic lighting and shadow effects are simulated. Based on the virtual depth value of each pixel and the angle between the surface normal and the viewing direction, the corresponding parallax value is calculated. The horizontal offset of the pixel is adjusted according to the parallax value to create a slight difference in the pattern seen by the left and right eyes, thereby generating a stereoscopic visual effect with depth perception on the surface of the two-dimensional pattern.

[0075] The initial decorative pattern underwent texture detail optimization, with subtle wood grain textures added to the geometric lines, mirroring the grain of the solid wood flooring. This increased textural depth ensured rich detail across different viewing distances. The contrast and transparency of the textures were adjusted to harmonize with the pattern's base colors and element layout, creating a visual harmony with the latex paint on the living room walls and the solid wood flooring.

[0076] The optimized decorative patterns are mapped onto the digital 3D model of the living room, determining their application to the TV and sofa background walls. The patterns are scaled and cropped according to the actual wall dimensions to ensure a perfect fit. The position and angle of a virtual camera are set to render the 3D model from multiple perspectives, including front, side, and obliquely above, generating a full-view visual effect image. Simultaneously, natural light at different times of day (morning, noon, afternoon) and artificial lighting with and without lights are simulated to generate visual effect images under various lighting conditions.

[0077] Based on the rendering results, local adjustments were made to the decorative patterns. It was found that the light blue portion of the pattern appeared too bright in the area near the window when natural sunlight was strong at midday. The brightness parameters of the pattern in this area were adjusted to reduce the color brightness. A slight visual distortion was found in the corner area of ​​the sofa background wall when viewed from an oblique angle. The proportions and positions of the elements in the pattern in this area were adjusted to correct the visual distortion.

[0078] The system receives real-time user feedback, indicating a user's request to add a small amount of plant leaf elements to the blank areas of the pattern. It identifies the user-specified modification area and content, converts the request to add plant leaf elements into feature adjustment parameters, inputs them into a generative adversarial network model for local regeneration, and generates a modified decorative pattern containing plant leaf elements. The modified pattern is then further optimized for color harmony, enhanced stereoscopic visual effects, and refined texture details. Finally, it is mapped onto a 3D model for spatial adaptation verification, confirming that the modified pattern is consistent with the overall style and spatial environment.

[0079] From a database of historically generated decorative patterns, the five reference patterns with the highest similarity to the user's Nordic style preference are retrieved. The stylistic features and natural texture elements of these reference patterns are extracted. The features of multiple reference patterns are weighted and fused to generate a fused feature vector. This vector is then input into a generative adversarial network model to generate decorative patterns that incorporate various Nordic style elements for the user to choose from.

[0080] The system generates the final decorative pattern file, producing corresponding file formats based on different construction processes. For wall printing, it generates a TIFF file in CMYK color mode with a resolution of 300 dpi. For wall spraying, it generates a PDF file with color separation information. It also outputs the pattern's size, color, positioning, and splicing parameters, as well as the construction steps, precautions, and quality acceptance standards for both printing and spraying processes.

[0081] This embodiment comprehensively collects spatial, material, and lighting parameters of a residential living room to establish a precise digital 3D model, ensuring that the generated decorative patterns perfectly match the living room's spatial structure and environmental characteristics. Multi-scale feature extraction accurately captures the user's personalized needs, and combined with generative adversarial networks, achieves high-quality pattern generation, satisfying the user's aesthetic requirements for Nordic style and natural elements. Multiple visual optimization steps comprehensively enhance the pattern's color harmony, three-dimensional visual effect, and texture, allowing the pattern to seamlessly integrate with the overall interior environment. Full-view rendering verification and interactive iterative optimization promptly correct visual issues, ensuring the final application effect. File output adapted to various construction techniques and detailed construction instructions effectively reduce errors during construction, improving accuracy and efficiency.

[0082] Example 2

[0083] This embodiment is applied to the design and generation of floor decoration patterns in a commercial exhibition hall. First, the interior space parameters of the building are collected. A LiDAR scanning device is used to perform a 360-degree scan of the entire exhibition hall, acquiring raw point cloud data. The raw point cloud data is then denoised to remove irrelevant point cloud data such as personnel and equipment mixed in during the scanning process. A multi-view geometry algorithm is used to register the scan data from different locations, generating a complete point cloud model covering the entire exhibition hall. The registered point cloud model is then triangulated to generate a high-precision 3D mesh model, simultaneously recording the overall length, width, and height dimensions of the exhibition hall, as well as the division of various display areas, and the location and dimensions of architectural components such as columns, entrances / exits, and stairs. Material parameters for the floor, walls, and ceiling are collected. The floor is made of polished tiles, the walls are made of white latex paint, and the ceiling is made of aluminum composite panels. Natural lighting parameters are collected, recording the location and size of the skylights in the exhibition hall, as well as the range and intensity variations of natural light within the hall. Artificial lighting parameters are collected, recording the location, angle, and lighting parameters of track lights, wall washers, and light strips. Based on the above parameters, a digital 3D model of the exhibition hall was created.

[0084] The process involves collecting users' decoration needs. Users provided four reference images in a modern, technological style, specifying a color preference for cool colors such as dark blue, light gray, and silver-gray. Their cultural element requirements included technological lines, abstract geometric shapes, and circuit textures. Functional zoning requirements stipulated that the floor decoration patterns should guide pedestrian flow, differentiate different product display areas, and align with the showroom's technological theme. The user-provided reference images, color preference data, cultural element requirements, and functional zoning requirements were integrated to construct a user requirement feature dataset.

[0085] Multi-scale feature extraction is performed on the user demand feature dataset. A convolutional neural network is used to extract hierarchical features from style reference images. First, low-level visual features such as edges, corners, and textures are extracted. Then, high-level semantic features such as style semantics, compositional logic, and element combination rules are extracted through high-level convolutional layers. An autoencoder is used to encode users' cool color preference data, generating color feature vectors that represent color distribution and matching patterns. Word embedding technology is used to extract textual features from cultural element requirements and functional requirements for pedestrian flow guidance and area division, converting textual descriptions into numerical textual feature vectors. All feature vectors are concatenated and normalized to map feature values ​​to a unified numerical range, generating the final feature encoding vector.

[0086] The feature encoding vector is input into a pre-trained generative adversarial network model to generate an initial decorative pattern. After receiving the feature encoding vector, the generator performs multiple convolutional, pooling, and deconvolution operations to output an initial decorative pattern with a resolution of 8192×8192. The initial decorative pattern includes smooth, technologically-inspired lines and textures, a gradient color distribution of deep blue and silver-gray, and a streamlined element layout, with the lines generally following the flow of people in the exhibition hall. A discriminator compares the generated initial pattern with samples of real technological-style decorative patterns and outputs the discrimination result. The model continuously iterates and optimizes parameters to improve the stylistic accuracy and content plausibility of the generated pattern.

[0087] The initial decorative pattern underwent color coordination optimization by converting it from the RGB color space to the Lab color space and calculating the brightness, a-channel, and b-channel values ​​for each pixel. Based on the overall cool color scheme of the exhibition hall, the primary color was determined to be silver-gray, with dark blue and light gray as secondary colors, and parameter ranges were set for each color. The parameters of all colors in the pattern were adjusted to fall within their respective ranges. Simultaneously, the color difference between adjacent colors in the pattern was calculated to ensure it remained within a visually comfortable range, preventing visual fatigue when the pattern was applied to large areas of the floor.

[0088] The initial decorative pattern was enhanced with a 3D visual effect. The technological lines and geometric elements of the initial decorative pattern were extracted, and virtual depth values ​​were assigned based on the importance and functional attributes of the elements. The virtual depth values ​​ranged from 0.1 meters to 1.0 meters, with higher virtual depth values ​​assigned to pedestrian guidance lines and area boundary lines, and lower virtual depth values ​​assigned to the background area of ​​the pattern. Based on the artificial lighting parameters of the exhibition hall, the light intensity and shadow intensity of each pixel under track spotlight illumination were calculated. The brightness and contrast of the pixels were adjusted to simulate the light and shadow effects produced by the spotlights. Based on the virtual depth value of each pixel and the angle between the surface normal and the viewing direction, the corresponding parallax value was calculated. The horizontal offset of the pixels was adjusted according to the parallax value to give the pattern a distinct 3D visual effect and enhance the visual impact of the pedestrian guidance lines.

[0089] The initial decorative pattern underwent texture detail optimization, with subtle circuit textures added to the areas featuring technological lines, ensuring the circuit textures align with the direction of the technological lines. The texture layers were increased to clearly display pedestrian flow guidance lines and area divisions when viewed from a distance, while revealing rich circuit details upon close inspection. The gloss level of the texture was adjusted to match the material characteristics of the polished floor tiles and to echo the technological theme of the exhibition hall.

[0090] The optimized decorative patterns are mapped onto the digital 3D model of the exhibition hall. The patterns are scaled and stitched according to the actual size and shape of the hall floor to ensure a perfect fit. Virtual camera positions and angles are set to simulate visitor perspectives, rendering the 3D model from multiple viewpoints, including entrances, various exhibition areas, and passageways, generating a full-view visual effect image. Simultaneously, both natural daylight and artificial nighttime lighting conditions are simulated to generate visual effect images under different lighting conditions.

[0091] Based on the rendering results, local adjustments were made to the decorative patterns. It was found that the pedestrian guidance lines at the corner of the exhibition hall exhibited visual distortion when viewed from an oblique angle. The curvature and width of the lines in this area were adjusted to correct the visual distortion. It was also found that the boundary lines of some exhibition areas appeared darker under spotlights. The brightness parameters of the lines in these areas were adjusted to improve their visibility.

[0092] The system receives real-time user feedback, indicating a user's request to adjust the density of the technical lines, increasing line density in certain areas. It identifies the user-specified adjustment area and parameters, converting the line density adjustment request into feature adjustment parameters, which are then input into a generative adversarial network model for local regeneration, generating the adjusted decorative pattern. The adjusted pattern is further optimized for color harmony, enhanced stereoscopic visual effects, and refined texture details, and then mapped onto a 3D model for spatial adaptation and pedestrian guidance effect verification.

[0093] From a database of historically generated decorative patterns, six reference patterns with the highest similarity to the user's modern technological style requirements are retrieved. Technological line features and circuit texture features of these reference patterns are extracted. The features of multiple reference patterns are weighted and fused to generate a fused feature vector. This vector is then input into a generative adversarial network model to generate decorative patterns that incorporate various technological style elements, allowing users to compare and choose.

[0094] The final decorative pattern file is generated, with corresponding file formats based on different construction techniques. For ground carving, a vector SVG file is generated. For ground printing, a CMYK color mode TIFF file is generated with a resolution of 200 dpi. The output also includes the pattern's size, color, positioning, and splicing parameters, as well as the construction steps, precautions, and quality acceptance standards for both carving and printing techniques.

[0095] This embodiment addresses the large space and functional requirements of commercial showrooms by achieving precise adaptation of large-area floor decorative patterns through high-precision spatial parameter acquisition and 3D model creation. Multi-scale feature extraction accurately captures the user's technological style requirements and pedestrian flow guidance functionalities, resulting in patterns that not only possess excellent artistic effects but also effectively guide pedestrian flow and delineate display areas. Enhanced stereoscopic visual effects and optimized texture details elevate the visual impact and quality of the patterns, perfectly aligning with the showroom's technological theme. Full-view rendering verification allows for testing the pattern effects from the visitor's actual perspective, enabling timely identification and correction of issues. File output adaptable to various construction techniques and detailed construction instructions ensure the accuracy and consistency of large-area floor construction, enhancing the overall decorative quality of the commercial showroom.

[0096] Figure 1 This paper summarizes the overall workflow of the intelligent generation and visual optimization method. The system begins with a comprehensive scan of the interior space using LiDAR to construct a digital 3D model, simultaneously integrating user style, color, and functional requirements. A multimodal feature extraction module deeply analyzes these requirements, generating feature encoding vectors. A generative adversarial network model uses this as its core to construct the initial pattern. Subsequently, the system enters the core optimization phase, significantly improving the pattern's texture and spatial fit through color harmony adjustments, stereoscopic visual enhancement, and texture detail rendering. Finally, the optimized scheme is mapped back to the 3D model for full-view rendering and distortion correction. The system outputs professional pattern files and guidance information covering various construction techniques, achieving closed-loop management from requirements gathering to construction implementation.

[0097] Figure 2This document details the core process of constructing a high-precision digital 3D model of a building. The system uses LiDAR to scan the target interior space from multiple points, capturing massive amounts of raw point cloud data. Through a series of data preprocessing algorithms, measurement noise is removed, spatial point cloud registration is performed, and unstructured, scattered point clouds are meshed. This process not only accurately recreates the dimensions and shapes of walls, floors, and ceilings but also automatically identifies and locates the positions of key building components such as doors, windows, beams, and columns. The resulting high-precision mesh model serves as the physical foundation for subsequent spatial adaptation of decorative patterns, texture mapping, and lighting simulation, ensuring complete dimensional accuracy between the virtual patterns and the real interior structure, preventing errors during later construction.

[0098] Figure 3 The algorithmic chain from requirement analysis to model training is described in detail. The system deconstructs user requirements into multi-scale features such as low-level visual, high-level semantic, color preferences, and functional zoning, and transforms heterogeneous data into a unified encoding vector through a deep neural network. After receiving this vector, the GAN model enters adversarial training mode. The generator is responsible for synthesizing patterns based on the vector, while the discriminator provides supervised feedback by comparing with a real sample library. During training, the system dynamically adjusts the weight coefficients by calculating three core loss functions: style, content, and spatial adaptation, thereby achieving a precise balance between aesthetic style, user intent, and physical space. This mechanism ensures that the generated decorative patterns not only conform to artistic aesthetics but also have extremely high feasibility for implementation in interior spaces.

[0099] Figure 4 This demonstrates the core technical logic for visual secondary processing of initially generated patterns. In the color optimization stage, the system converts the pattern space to the Lab color model. Based on a preset overall indoor hue threshold, it performs automated parameter mapping on hue, brightness, and saturation, and calculates color difference values ​​to ensure the pattern's color layout conforms to the visual comfort zone, avoiding color clashes. In the stereoscopic enhancement stage, the system uses edge contour detection to extract texture depth information. Through depth value calculation and parallax simulation, it reconstructs multi-level visual offsets on a two-dimensional plane. By calculating surface normal shadows based on an indoor lighting model, it imbues the pattern with a realistic depth perception, allowing it to present a lifelike decorative depth from different viewing angles, effectively breaking away from the flatness of traditional wall decorations.

[0100] Figure 5This system presents an optimized closed loop from user interaction feedback to the final delivery of standardized construction documents. During the interaction phase, users can directly make local corrections to the preview image. This instruction is instantly converted into feature parameters, driving the model to regenerate in the local area, avoiding the efficiency loss caused by redoing the entire image. Once visual confirmation of no distortion, the system automatically refines the image according to the pre-selected construction process. For printing processes, it outputs TIFF files in CMYK color mode; for engraving processes, it converts to SVG vector paths; and for spraying processes, it generates PDF drawings containing color separation information. Simultaneously, the system automatically extracts all positioning and splicing parameters and construction precautions, automatically forming a structured acceptance standard document, truly achieving a seamless transition from intelligent design schemes to construction guidance documents.

[0101] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent generation and visual optimization of architectural interior decoration patterns, characterized in that, Includes the following steps: Collect three-dimensional structural parameters of the building's interior space, wall and floor material parameters, natural light intensity and angle parameters, and artificial lighting distribution parameters to establish a digital three-dimensional model of the building's interior space; Collect user-provided style reference images, color preference data, cultural element requirements, and functional zoning requirements to construct a user requirement feature dataset. Multi-scale feature extraction is performed on the user demand feature dataset, extracting style features, color features, cultural element features and functional adaptation features respectively, and generating feature encoding vectors; The feature encoding vector is input into a pre-trained generative adversarial network model to generate an initial decorative pattern, which includes a basic texture, color distribution, and element layout. The initial decorative pattern is optimized for color coordination by adjusting the hue, saturation, and brightness parameters of each color in the pattern to match the overall color scheme of the building's interior space. The initial decorative pattern is enhanced with a stereoscopic visual effect by simulating light and shadow from different angles to generate a depth-perceived visual effect on the surface of the two-dimensional pattern. The initial decorative pattern is optimized in terms of texture details, increasing the texture layers and quality of the pattern, so that the pattern visually echoes the materials of the building's interior walls and floors; The optimized decorative patterns are mapped onto the digital 3D model of the building's interior space for full-view visual rendering. Based on the rendering results, local adjustments are made to the decorative pattern to correct visual distortion and color deviation of the pattern under different viewing angles; Generate the final decorative pattern file and output pattern parameters and construction guidance information adapted to different construction techniques.

2. The method for intelligent generation and visual optimization of architectural interior design patterns as claimed in claim 1, wherein, It also includes an interactive pattern iterative optimization step based on real-time user feedback. This step receives local modification instructions from the user for the generated pattern, identifies the user's modified area and content, converts the modified content into feature adjustment parameters, inputs them into a generative adversarial network model for local regeneration, generates the modified decorative pattern, and performs visual effect optimization and spatial adaptation verification again.

3. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, It also includes a pattern style transfer and fusion step based on historical generated data. This step retrieves multiple reference patterns with the highest similarity to user needs from the historically generated decorative pattern database, extracts the style features and element features of the reference patterns, and weights and fuses the features of multiple reference patterns to generate a fused feature vector. This vector is then input into a generative adversarial network model to generate decorative patterns with multiple style fusion effects.

4. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When collecting three-dimensional structural parameters of the building's interior space, a lidar scanning device is used to scan the interior space from all directions to obtain point cloud data. The point cloud data is then denoised, registered, and meshed to generate a high-precision three-dimensional mesh model. At the same time, the dimensions, shapes, and locations of the walls, floors, and ceilings, as well as the locations and dimensions of building components such as doors, windows, beams, and columns, are recorded.

5. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When performing multi-scale feature extraction on the user demand feature dataset, a convolutional neural network is used to extract low-level visual features and high-level semantic features from style reference images. Low-level visual features include edge, texture, and shape features, while high-level semantic features include style type, composition method, and element combination features. An autoencoder is used to encode color preference data to generate color feature vectors. Word embedding technology is used to extract text features from cultural element requirements and functional zoning requirements to generate text feature vectors. All feature vectors are concatenated and normalized to generate a unified feature encoding vector.

6. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When the feature encoding vector is input into a pre-trained generative adversarial network (GAN) model to generate an initial decorative pattern, the GAN model consists of two sub-networks: a generator and a discriminator. The generator receives the feature encoding vector and outputs the initial decorative pattern, while the discriminator receives the generated pattern and real decorative pattern samples and outputs the discrimination result. During training, the model parameters are optimized by minimizing the generator loss function and maximizing the discriminator loss function. The generator loss function is calculated as follows: ； This represents the total loss value of the generator. Represents the style loss value. Indicates the content loss value. This represents the spatial adaptation loss value. The weighting coefficients representing style loss. The weighting coefficients representing content loss. The weighting coefficients represent the spatial adaptation loss.

7. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When optimizing the color coordination of the initial decorative pattern, the initial decorative pattern is converted from the RGB color space to the Lab color space. In the Lab color space, the brightness value, a channel value, and b channel value of each color in the pattern are calculated. The parameter range of the main color and auxiliary color is determined according to the overall color tone of the building's interior space. The parameters of each color in the pattern are adjusted to fall within the corresponding parameter range. At the same time, the color difference between adjacent colors is calculated to ensure that the color difference between adjacent colors is within the visually comfortable range.

8. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When enhancing the stereoscopic visual effect of the initial decorative pattern, the edge contour and texture features of the initial decorative pattern are first extracted. A virtual depth value is assigned to each pixel based on the importance of the feature, with the virtual depth value ranging from 0.1 meters to 1.0 meters. Then, the light intensity and shadow intensity of each pixel are calculated based on the lighting parameters of the building's interior space. The lighting and shadow effects are simulated by adjusting the brightness and contrast of the pixels. The intensity of the stereoscopic visual effect is controlled by the parallax value, which is calculated as follows: ； Represents pixels disparity value at that location, Indicates the pixel density of the display device. This represents the equivalent focal length of the human eye. The interpupillary distance of the human eye. Represents pixels The corresponding virtual depth value, Represents pixels The angle between the surface normal at that point and the direction of the line of sight.

9. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When mapping the optimized decorative pattern onto the digital 3D model of the building's interior space for full-view visual rendering, the application position and size of the decorative pattern in the 3D model are first determined. The 2D decorative pattern is then texture-mapped to fit the corresponding surface of the 3D model. The position and angle of the virtual camera are then set to render the 3D model from multiple different perspectives, generating full-view visual effect images. At the same time, natural lighting at different times and artificial lighting in different modes are simulated to generate visual effect images under different lighting conditions.

10. The method for intelligent generation and visual optimization of architectural interior decoration patterns according to claim 1, characterized in that, When generating the final decorative pattern file and outputting pattern parameters and construction guidance information adapted to different construction processes, different resolutions and formats of pattern files are generated according to different construction processes. For printing construction processes, a TIFF format file in CMYK color mode is generated; for engraving construction processes, a vector format SVG file is generated; and for spraying construction processes, a PDF file with color separation information is generated. At the same time, the size parameters, color parameters, positioning parameters, and splicing parameters of the pattern are output, as well as construction steps, precautions, and quality acceptance standards for different construction processes.

Images