An image generation method and system for an urban renewal scene

Abstract

The present application relates to the general image data processing technical field, especially to a kind of city renewal scene image generation method and system, it includes: obtaining multimodal data and preprocessing conversion into uniform data representation format;Call general generation model to generate initial image;Through specification constraint module, compliance detection and correction are carried out;Based on local culture characteristic library, style migration is carried out;Receiving user local modification instruction carries out iterative redrawing.The system includes data acquisition module, initial generation module, specification constraint module, style migration module and local iteration module.The present application realizes that generated image conforms to professional specification, embodies local culture characteristic, and improves man-machine collaborative design efficiency.

Description

Technical Field

[0001] This invention relates to the field of general image data processing technology, and in particular to an image generation method and system for urban renewal scenarios. Background Technology

[0002] In the planning and design of urban renewal projects, rendering is a core step in showcasing design schemes to decision-makers, experts, and the public. Traditionally, this relies on professional designers manually drawing the images or using 3D modeling software to render them frame by frame, which is time-consuming, labor-intensive, and requires a high level of technical skill from the designers. With the rapid development of deep learning technology, image generation technologies based on architectures such as generative adversarial networks and diffusion models have made breakthrough progress in the field of content creation, providing a new technical path for the automatic generation of urban landscape renderings.

[0003] Existing image generation models are primarily trained on natural images or artistic datasets, and their results often prioritize visual aesthetics while lacking an accurate understanding of professional standards and engineering constraints in urban planning. When generating images involving elements such as urban roads, building setbacks, and green space layouts, it is difficult to ensure that the generated content complies with relevant national and local engineering standards. This can lead to professional issues such as insufficient building spacing, non-compliant setback distances, and unreasonable layouts of municipal facilities, resulting in renderings that cannot be directly used for project presentations and expert reviews.

[0004] At the same time, urban renewal projects have significant local characteristics. The urban landscape, cultural traditions, and architectural styles of different regions vary significantly. Although existing general image generation models have a certain style transfer capability, they lack specialized training for specific local cultural characteristics. This makes it difficult to accurately capture and reproduce regional features such as Shanghai style, Jiangnan water towns, and Lingnan architecture, resulting in the generated renderings being out of touch with the cultural context of the project location. Summary of the Invention

[0005] The purpose of this invention is to provide an image generation method and system for urban renewal scenarios to solve the technical problems of generated images not conforming to urban planning professional standards and lacking local cultural characteristics.

[0006] One aspect of the present invention provides an image generation method for urban renewal scenarios, comprising: S1. Obtain the input multimodal data, which includes at least planning text, hand-drawn sketches, block models, and site photos. After preprocessing the input data, convert it into a unified data representation format. S2. Call the general generation model to generate an initial image based on the unified data representation format to obtain an initial urban landscape rendering image; S3. Input the initial urban landscape rendering image into the standard constraint module for compliance detection and correction. The standard constraint module stores road landscape standard atlases and various guideline data. S4. Based on the local cultural characteristics database, style transfer is performed on the corrected images to generate urban renewal scene images that conform to the landscape characteristics of the target region; S5. Receive a user's local modification instruction for the urban renewal scene image, and iteratively redraw the specified area of ​​the image based on the modification instruction.

[0007] In some embodiments, the preprocessing includes at least image size normalization, color space conversion, and noise removal; the unified data representation format is a feature vector of a preset dimension.

[0008] In some embodiments, the general generation model is a deep neural network based on a diffusion model or a generative adversarial network; the resolution of the initial image generation is not lower than a predetermined resolution threshold.

[0009] In some embodiments, the specification constraint module further includes: a building spacing detection unit, a setback distance detection unit, and a municipal facility layout detection unit; the road landscape standard atlas includes vector data of urban road design specifications, greening configuration standards, and building setback requirements.

[0010] In some embodiments, when content that does not meet the specification requirements is detected, the specification constraint module automatically generates a correction vector and feeds the correction vector back to the general generation model for regeneration; the correction process is executed iteratively until the generated image meets all specification requirements.

[0011] In some embodiments, the local cultural features database stores regional architectural style characteristics; the style transfer is achieved by separating the style encoder and the content encoder, adjusting the style features while keeping the content structure of the generated image unchanged.

[0012] In some embodiments, the regional architectural style features include at least the building facade material, roof form, color scheme, door and window style, and the morphological features of decorative components; the local cultural feature database stores feature data of Shanghai style, Jiangnan water town style, and Lingnan architectural style.

[0013] In some embodiments, the local modification instructions are generated via an interactive mask interface; the user specifies the area to be modified by drawing a mask area on the generated image. The iterative redrawing is performed only on the masked area, while other areas remain unchanged.

[0014] In some embodiments, the iterative redrawing is based on an attention-based local generation network, which can adjust a specified region while maintaining global consistency.

[0015] Another aspect of the present invention provides an image generation system for urban renewal scenarios, including a data acquisition module with interconnected data, an initial generation module, a standard constraint module, a style transfer module, and a local iteration module; The data acquisition module is used to acquire the input multimodal data; The initial generation module is used to call a general generation model and generate an initial image based on a preprocessed unified data representation format to obtain an initial urban landscape rendering image. The standard constraint module is used to store road landscape standard atlases and various guidelines data, and to perform compliance detection and correction on the initial urban landscape rendering image; The style transfer module is used to store a local cultural characteristics library, perform style transfer on the corrected images, and generate urban renewal scene images that conform to the landscape characteristics of the target region. The local iteration module is used to receive local modification instructions from the user for the urban renewal scene image, and to iteratively redraw the specified area of ​​the image based on the modification instructions.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. By injecting a standard constraint module into the general generative model, it can automatically detect and correct the road, building setbacks and greening layout in the generated images to make them comply with the relevant requirements of national and local engineering specifications. The generated renderings can be directly used for scheme presentations and expert reviews. 2. By establishing a database of local cultural characteristics, it is possible to accurately capture and reproduce the style features of different regions such as Shanghai-style architecture, Jiangnan water towns, and Lingnan architecture, so that the generated renderings are highly integrated with the cultural context of the project location; 3. By supporting multimodal input and local iterative redrawing, designers can directly make targeted modifications and improvements based on the generated results, significantly improving the human-computer collaboration efficiency of the design workflow. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a flowchart of the image generation method for urban renewal scenarios according to the present invention; Figure 2 This is a schematic diagram of the specification constraint module of the present invention. Detailed Implementation

[0019] The following will be based on embodiments of the present invention. Figures 1-2 The technical solutions in the embodiments of the present invention will be clearly and completely described together. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Example 1 In a historical district renewal project in the core area of ​​a city, planning and design personnel need to quickly generate multiple sets of renderings in different styles for scheme comparison and public display. Traditional manual rendering methods are time-consuming and difficult to iterate quickly. This embodiment is designed for this application scenario, constructs a complete urban renewal scene image generation system, and describes its workflow in detail.

[0021] This scenario deployment comprises an urban renewal scene image generation system, consisting of a data acquisition module, an initial generation module, a specification constraint module, a style transfer module, and a local iteration module. The system is deployed on a server cluster equipped with high-performance graphics processing units. The data acquisition module is located at a terminal workstation within the project design office, used to receive raw data in various formats input by planning and design personnel. The initial generation module, specification constraint module, and style transfer module run on a deep learning inference framework on the server side, achieving efficient image generation and processing through distributed computing resources. The local iteration module is deployed as an interactive web service interface, accessible to designers via a browser for real-time editing.

[0022] The data acquisition module further includes four sub-components: a text parsing unit, an image preprocessing unit, a 3D model conversion unit, and a photo enhancement unit.

[0023] The text parsing unit employs a Transformer-based natural language processing model, capable of extracting key spatial parameters from planning texts. These parameters include numerical data such as land use, building height restrictions, floor area ratio requirements, and greening rate indicators, as well as spatial constraints such as road grades and river blue lines. The image preprocessing unit supports importing common image formats, including standard formats like JPEG, PNG, and TIFF. It can perform edge detection and contour extraction on input hand-drawn sketches, converting 2D line drawings into pixel matrices suitable for deep learning models. The 3D model conversion unit integrates interfaces with commonly used 3D modeling software, capable of reading block model files exported from software such as SketchUp, Rhino, and 3ds Max, projecting 3D geometric information into a multi-view 2D view sequence. The photo enhancement unit is equipped with a convolutional neural network-based super-resolution reconstruction model, capable of improving the quality of low-resolution photos taken on-site, removing motion blur and noise interference, while correcting perspective distortion and color temperature deviation.

[0024] The raw multimodal data collected by the data acquisition module is converted into a unified data representation format after preprocessing. The preprocessing process includes three processing stages. First, the image size normalization stage is entered, in which all input images are uniformly adjusted to a standard size of 512 by 512 pixels, while maintaining the aspect ratio and using a bilinear interpolation algorithm to fill in the missing areas.

[0025] Next, the color space conversion stage begins, converting the image data from the RGB color space to the Lab color space, separating the luminance and chrominance channels so that subsequent models can more accurately understand the color distribution characteristics.

[0026] Finally, in the noise removal stage, an adaptive filtering algorithm based on wavelet transform is used to effectively eliminate sensor noise and compression artifacts introduced during image acquisition.

[0027] The preprocessed image data and the structured parameter data extracted from the planning text are encapsulated together into a feature vector of a preset dimension, specifically 2048 dimensions, of which image features occupy 1536 dimensions and text parameters occupy 512 dimensions. The feature vector is constructed using a feature encoder trained with a contrastive learning framework, which ensures that semantically similar input data have high cosine similarity in the feature space.

[0028] The initial generation module calls the general generation model to execute S2. The general generation model employs a deep neural network architecture based on a diffusion model, specifically an improved version of the Latent Diffusion Model. This model has been fine-tuned on a dataset of over five million urban landscape images and is capable of understanding the visual features of urban spatial elements such as building morphology, road networks, and green space layout. The model's text encoder uses a CLIP ViT-L / 14 architecture, receiving structured parameter vectors from the data acquisition module and generating semantic embeddings aligned with the input conditions. The image decoder uses a U-Net network based on calibrated attention, including twelve downsampling stages and twelve upsampling stages, achieving multi-scale feature fusion through residual connections and cross-attention mechanisms.

[0029] The initial generation module employs Classifier-Free Guidance technology to enhance conditional control during the generation process, with a guidance coefficient set to 7.5. This ensures that while maintaining generation diversity, it strictly adheres to the constraints of the input planning parameters. Noise scheduling during the generation process utilizes a cosine decay strategy, gradually reducing the noise level from a maximum of 0.1 to 0.001, with a total of 50 iterations, achieving a balance between generation quality and computational efficiency. The output resolution of the initial image generation is set to 1024 x 1024 pixels. A block rendering strategy is used, dividing the image into four 512 x 512 sub-blocks, generating them separately, and then seamlessly stitching them together to eliminate boundary artifacts. Finally, the output initial urban landscape rendering image includes basic visual elements such as building outlines, linear road networks, green area divisions, and a sky background.

[0030] The compliance and constraint module stores standard road landscape maps and various guidelines, used for compliance checks and corrections of initial urban landscape rendering images. This module includes a building spacing detection unit, a setback distance detection unit, and a municipal facility layout detection unit.

[0031] The building spacing detection unit employs a semantic segmentation-based deep learning model, capable of accurately identifying the bounding boxes and outline polygons of all individual buildings in an image, and then calculating the horizontal spacing between adjacent buildings. The detection model has been specifically trained on an urban building dataset, which includes over 100,000 training samples labeled with building spacing information, and can handle spacing calculations for various building types, including multi-story, high-rise, and super high-rise buildings.

[0032] The setback distance detection unit is based on an instance segmentation network and can accurately extract the spatial distance between the building outline and the land boundary line. By comparing it with the preset building setback requirements, it identifies the distance areas that do not meet the specifications.

[0033] The municipal facility layout detection unit integrates a knowledge graph of urban municipal facility standards, which can identify municipal facilities such as street light poles, transformers, and gas pressure regulating stations in images, and detect whether the safe distance between them and buildings and pedestrian walkways meets the standard requirements.

[0034] When performing compliance checks, the compliance module first inputs the initial urban landscape rendered image into the semantic segmentation network to generate pixel-level labeled images for categories such as buildings, roads, green spaces, municipal facilities, and land boundaries. Then, it calculates the corresponding compliance index for each category of detected objects. When content that does not meet compliance requirements is detected, the compliance module automatically generates a correction vector. The dimension of the correction vector is consistent with the latent space dimension of the general generative model, and it is mapped to visual-level correction guidance through a pre-trained decoder.

[0035] The correction vector is fed back to the cross-attention layer of the general generative model, guiding the model to adjust the visual features of the corresponding regions during the regeneration process. This correction process adopts an iterative execution mechanism, and after each iteration, the specification compliance check is performed again until the generated image meets all specification requirements or the maximum number of iterations is reached. The maximum number of iterations is set to three. After three iterations, regions that still do not fully meet the specifications will be marked by local masking to prompt planning and design personnel for manual confirmation.

[0036] The style transfer module stores a local cultural feature library, which is used to perform style transfer on the corrected images to generate urban renewal scene images that conform to the landscape characteristics of the target region. The local cultural feature library stores the architectural style features of each region using feature embedding, with the feature dimension set to 256. Discriminative style representations are extracted from massive architectural images through a contrastive learning framework.

[0037] The style transfer module comprises a style encoder, a content encoder, and a fusion generator. The style encoder uses the first five layers of a VGG-19 network as a feature extractor, capable of separating style features such as texture, color, and brushstrokes from the reference architectural image. The content encoder uses a ResNet-50 network, capable of extracting semantic content features such as architectural structure and spatial layout from the original image. The fusion generator achieves the separation and recombination of style and content based on adaptive instance normalization. This technique learns affine transformation parameters to transfer the statistical properties of style features to the conditional distribution of content features, maintaining the content structure of the generated image while adjusting the style features.

[0038] The local cultural characteristics database stores feature data on three regional styles: Shanghai style, Jiangnan water town style, and Lingnan architectural style. Shanghai style features are extracted from image data of the Bund architectural complex and lane houses, with feature parameters covering the geometric features and textural statistical characteristics of visual elements such as exposed red brick walls, roughcast cement walls, Shikumen door frames, dormer windows, and old-style Western-style houses. Jiangnan water town style features are extracted from image data of historical districts in Suzhou and Shaoxing, with feature parameters covering the spatial distribution patterns and color matching patterns of elements such as whitewashed walls and black tile roofs, wooden lattice doors and windows, waterfront corridors, and gable walls. Lingnan architectural style features are extracted from image data of wok-ear houses and arcade streets, with feature parameters covering the morphological features and structural details of elements such as wok-ear gables, Manchurian windows, sliding doors, and colonnaded arcades. During execution, the style transfer module first retrieves the corresponding style feature vector from the local cultural characteristic database based on the user-specified regional style type. Then, it inputs the style feature vector along with the content features output by the content encoder into the fusion generator to generate an urban renewal scene image that conforms to the target regional landscape characteristics. It should be noted that the output resolution of the fusion generator is consistent with the input image, and an adversarial training strategy is employed to ensure that the generated image achieves photorealistic visual quality.

[0039] The local iteration module receives user-defined modification commands for urban renewal scene images and iteratively redraws specified areas of the image based on these commands, achieving a human-computer collaborative design loop. The module provides a web-based interactive mask editing interface, allowing users to draw mask areas on the generated image using a drawing tool. The system supports three mask shape selection tools: circular, rectangular, and freehand. The edges of the mask areas are feathered with a feather radius of 8 pixels to ensure a natural transition between the redrawn area and surrounding pixels. Users can specify their modification intentions within the mask area using text descriptions or reference images. Text descriptions are encoded into semantic vectors using a pre-trained language model, while reference images have style features extracted using a style encoder.

[0040] The local iterative module employs an attention-based local generative network to redraw the masked region. This network takes the complete image and the masked region as dual inputs. First, a global context encoder extracts global features from the complete image, serving as constraints during the redrawing process. Then, a region attention module generates fine-grained features within the masked region. This module uses a hierarchical attention mechanism, enabling refined adjustments to a specified region while maintaining global consistency. The redrawing process employs an iterative optimization strategy. Each iteration generates a candidate result and calculates its feature consistency score with the original unmasked region. Feedback from the consistency score guides the generation direction of subsequent iterations. The number of iterations is set to 20. Finally, the output is a locally redrawn result updated only for the masked region, while other regions remain unchanged, achieving seamless image fusion.

[0041] To better understand this embodiment, the workflow of the system in this embodiment is now described: First, the data acquisition module receives raw data such as planning text, hand-drawn sketches, block models, and site photos input by planning and design personnel, and converts them into feature vectors in a unified format after preprocessing.

[0042] Then, the initial generation module calls the general generation model to generate the initial urban landscape rendering image based on the feature vector.

[0043] Next, the standard constraint module performs compliance checks on the initial image and automatically corrects any content that does not conform to the road, building setback, and green space layout standards.

[0044] Subsequently, the style transfer module extracts corresponding style features from the local cultural characteristics database based on the regional style type specified by the user, performs style transfer processing on the corrected image, and generates urban renewal scene images that conform to the landscape characteristics of the target region.

[0045] Finally, the local iteration module receives the user's local modification instructions, specifies the modification area through an interactive mask interface, and uses a local generative network based on an attention mechanism to iteratively redraw the specified area, thus realizing a human-computer collaborative design loop.

[0046] In a specific application scenario, a historical and cultural district renewal project needs to generate three sets of renderings in different styles for the project presentation.

[0047] The planning and design personnel first imported the project's planning text and existing site photos through the data acquisition module. The system automatically extracted key parameters such as a plot ratio of 2.5, a building height limit of 24 meters, and a greening rate of 35%. Subsequently, the initial generation module generated initial urban landscape rendering images based on these parameters. The regulatory constraint module detected that the setback distance of some buildings was less than 18 meters, automatically generating correction vectors that were fed back to the generation model for regeneration, ultimately producing image results that met regulatory requirements. Following this, the planning and design personnel selected three different architectural styles—Shanghai style, Jiangnan water town style, and Lingnan style—for style transfer, generating three sets of renderings with different regional characteristics for comparison.

[0048] During the design refinement phase, the planning and design team proposed modifications to the exterior details of a certain historical building. Through the local iteration module, a mask area was drawn on the generated image, specifying that the gable style of the building be modified to a more traditional form. The system automatically completed the local redrawing while maintaining harmony with the surrounding buildings and environment.

[0049] Example 2 Based on Example 1, this example further optimizes the detection accuracy and correction efficiency of the specification constraint module, and provides a solution for the multi-plot linkage generation needs in complex urban renewal scenarios.

[0050] In this embodiment, the normative constraint module introduces a building relationship reasoning unit based on a graph neural network. This unit can understand the spatial adjacency relationships and visual corridor constraints between individual buildings. The unit represents each individual building as a graph node and the spatial relationships between buildings as graph edges, calculating the influence weights of different relationship types through a graph attention mechanism.

[0051] When performing spacing detection, the building relationship reasoning unit considers not only the direct spacing between adjacent buildings, but also the overall form of the building group and the permeability requirements of visual corridors. For example, when it detects that the direct spacing between two buildings meets the code requirements, but the visual corridor formed by the two is blocked by a third building, the system will mark this situation as a potential problem and generate corresponding correction suggestions. This graph-based reasoning method significantly improves the code constraint module's ability to understand complex urban spatial relationships.

[0052] This embodiment also optimizes the generation strategy of the correction vector by introducing a multi-objective optimization framework that considers multiple regulatory constraints simultaneously. When the regulatory constraint module detects multiple non-compliant elements in the image, the multi-objective optimization framework analyzes the severity and remediation difficulty of various violations, generating a comprehensive correction vector to guide the generation model for global optimization. Severity assessment employs a deep learning-based scoring model, trained under the guidance of urban planning and engineering regulations experts, which accurately quantifies the technical risks of various violations. Remediation difficulty assessment considers the local impact of the correction operation, prioritizing correction schemes with less impact on surrounding areas.

[0053] To address the need for multi-plot collaborative generation scenarios, this embodiment extends the multi-input coordination function of the data acquisition module. In urban renewal projects, the same design scheme often involves the coordinated development of multiple adjacent plots, requiring the generation of an overall rendering reflecting the spatial relationships between the plots. The multi-input coordination function allows users to simultaneously import planning data from multiple plots. The system integrates the data from multiple plots into a unified input representation through coordinate matching and boundary stitching techniques. During the generation process, the general generation model employs a cross-plot attention mechanism, enabling it to understand the functional complementarity and spatial transition logic between different plots, generating a coherent and smooth overall urban landscape image.

[0054] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0055] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for generating images of urban renewal scenarios, characterized in that, include: S1. Obtain the input multimodal data, which includes at least planning text, hand-drawn sketches, block models, and site photos. After preprocessing the input data, convert it into a unified data representation format. S2. Call the general generation model to generate an initial image based on the unified data representation format to obtain an initial urban landscape rendering image; S3. Input the initial urban landscape rendering image into the standard constraint module for compliance detection and correction. The standard constraint module stores road landscape standard atlases and various guideline data. S4. Based on the local cultural characteristics database, style transfer is performed on the corrected images to generate urban renewal scene images that conform to the landscape characteristics of the target region; S5. Receive a user's local modification instruction for the urban renewal scene image, and iteratively redraw the specified area of ​​the image based on the modification instruction.

2. The method according to claim 1, characterized in that, The preprocessing includes at least image size normalization, color space conversion, and noise removal; the unified data representation format is a feature vector of a preset dimension.

3. The method according to claim 1, characterized in that, The general generation model is a deep neural network based on a diffusion model or a generative adversarial network; the resolution of the initial image generation is not lower than a predetermined resolution threshold.

4. The method according to claim 1, characterized in that, The specification constraint module also includes: a building spacing detection unit, a setback distance detection unit, and a municipal facility layout detection unit; the road landscape standard atlas includes vector data of urban road design specifications, greening configuration standards, and building setback requirements.

5. The method according to claim 1, characterized in that, When content that does not meet the specifications is detected, the specification constraint module automatically generates a correction vector and feeds the correction vector back to the general generation model for regeneration. The correction process is performed iteratively until the generated image meets all specification requirements.

6. The method according to claim 1, characterized in that, The local cultural features database stores regional architectural style characteristics; the style transfer is achieved by separating the style encoder and the content encoder, adjusting the style features while keeping the content structure of the generated image unchanged.

7. The method according to claim 6, characterized in that, The regional architectural style features include at least the building facade materials, roof forms, color schemes, door and window styles, and the morphological features of decorative components; the local cultural characteristics database stores feature data on the Shanghai style, the Jiangnan water town style, and the Lingnan architectural style.

8. The method according to claim 1, characterized in that, The local modification instructions are generated through an interactive mask interface; the user specifies the area to be modified by drawing a mask area on the generated image. The iterative redrawing is performed only on the masked area, while other areas remain unchanged.

9. The method according to claim 8, characterized in that, The iterative redrawing is based on an attention-based local generation network, which can adjust a specified region while maintaining global consistency.

10. An image generation system for urban renewal scenarios, characterized in that, It includes a data acquisition module for data interconnection, an initial generation module, a specification constraint module, a style transfer module, and a local iteration module; The data acquisition module is used to acquire the input multimodal data; The initial generation module is used to call a general generation model and generate an initial image based on a preprocessed unified data representation format to obtain an initial urban landscape rendering image. The standard constraint module is used to store road landscape standard atlases and various guidelines data, and to perform compliance detection and correction on the initial urban landscape rendering image; The style transfer module is used to store a local cultural characteristics library, perform style transfer on the corrected images, and generate urban renewal scene images that conform to the landscape characteristics of the target region. The local iteration module is used to receive local modification instructions from the user for the urban renewal scene image, and to iteratively redraw the specified area of ​​the image based on the modification instructions.

Images