Digital technology grab pattern enhancement method and system

Abstract

The present application relates to the technical field of digital technology grabbing graphics enhancement, and particularly relates to a digital technology grabbing graphics enhancement method and system, the method comprising the following steps: activating one or more semantic contexts of a scene object; for each activated semantic context, dynamically determining an importance degree value according to user attention, user interaction behavior and virtual tour narrative priority; for each activated semantic context, generating an independent enhancement rendering layer, the independent enhancement rendering layer being used to present visual performance of the semantic context; receiving the independent enhancement rendering layer and the importance degree value, fusing the independent enhancement rendering layer according to a semantic area where a pixel is located and the importance degree value, to obtain a final enhancement image; and rendering and outputting the final enhancement image in real time. The above improves the sense of reality of a virtual scene and the sense of immersion of a user.

Description

Technical Field

[0001] This invention relates to the technical field of digital technology for image enhancement, specifically to a method and system for digital technology for image enhancement. Background Technology

[0002] In the field of digital creativity and convergence services, virtual reality content production teams have been committed to providing users with immersive experiences. However, when rendering high-quality 3D scene graphics captured in real time, problems such as blurred textures, color distortion, and lack of detail in the original graphics are often encountered, which seriously affect the realism of the virtual scene and the user's immersion. These challenges are particularly prominent in some complex application scenarios, such as large-scale digital cultural heritage projects. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned shortcomings by proposing a digital technology-based image capture enhancement method and system.

[0004] The present invention adopts the following technical solution:

[0005] A digital technology-based image enhancement method, comprising the following steps:

[0006] Identify scene objects in a virtual scene, obtain semantic information associated with scene objects, and activate one or more semantic contexts of scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with scene objects is non-visual attribute information corresponding to scene objects and used to characterize the semantic meaning of scene objects. Non-visual attribute information includes at least one of functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, texture layer, material layer, environment layer, semantic attribute layer, and interaction layer.

[0007] For each activated semantic context, an importance value is dynamically determined based on user attention, user interaction behavior, and the priority of the virtual tour narrative;

[0008] For each activated semantic context, an independent enhancement rendering layer is generated. The independent enhancement rendering layer is used to present the visual representation of the semantic context. The independent enhancement rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters.

[0009] The system receives independent enhancement rendering layers and their importance values. Based on the semantic region where the pixel is located and its importance value, the system fuses the independent enhancement rendering layers to obtain the final enhanced image.

[0010] The final enhanced image will be rendered and output in real time.

[0011] This technical solution achieves intelligent enhancement of virtual scene graphics by introducing semantic context activation, dynamic determination of importance values, and fusion of independent enhancement rendering layers, thereby improving the realism of virtual scenes and the user's immersion. Especially in complex digital cultural heritage projects, it can better meet the user's information acquisition needs under different narrative modes.

[0012] This application also discloses a digital technology-based image enhancement system, applied to the aforementioned digital technology-based image enhancement method. The system includes:

[0013] The recognition module identifies scene objects in the virtual scene, obtains semantic information associated with the scene objects, and activates one or more semantic contexts of the scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with the scene objects is non-visual attribute information corresponding to the scene objects and used to characterize the semantic meaning of the scene objects. The non-visual attribute information includes at least one of the following: functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, a texture layer, a material layer, an environment layer, a semantic attribute layer, and an interaction layer.

[0014] The module determines an importance value for each activated semantic context based on user attention, user interaction behavior, and the priority of the virtual navigation narrative.

[0015] The generation module generates an independent enhanced rendering layer for each activated semantic context. The independent enhanced rendering layer is used to present the visual representation of the semantic context. The independent enhanced rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters.

[0016] The fusion module receives independent enhanced rendering layers and importance values, and fuses the independent enhanced rendering layers according to the semantic region where the pixel is located and the importance value to obtain the final enhanced image;

[0017] The output module will ultimately render and output the enhanced image in real time.

[0018] Through this technical solution, this application provides a system capable of implementing the aforementioned digital technology-based image enhancement method. This system, through modular design, achieves intelligent enhancement of virtual scene graphics, effectively solving the problems of texture blurring, color distortion, and lack of detail in existing technologies, as well as the conflict of enhancement requirements under multiple semantic contexts, thereby improving the realism of virtual scenes and the user's immersion.

[0019] This application, by introducing a semantic context activation and dynamic importance assessment mechanism, can intelligently identify and enhance key information points in virtual scenes based on user intent and the focus of the navigation narrative. Even areas that are far away or have blurred visual details can be rendered with high quality according to their semantic importance. Furthermore, by generating independent enhancement rendering layers and fusing them based on semantic regions and importance values, this application can finely handle enhancement conflict needs under multiple semantic contexts, avoiding the unintended consequences of simple superposition or averaging parameters. This ensures that the enhancement effects of different semantics can coexist harmoniously in complex scenes, thereby significantly improving the realism of virtual scenes, information delivery efficiency, and user immersion experience.

[0020] To further understand the features and technical content of the present invention, please refer to the following detailed description and drawings of the present invention. However, the drawings provided are for reference and illustration only and are not intended to limit the present invention. Attached Figure Description

[0021] Figure 1 This is a flowchart of a digital technology-based image enhancement method according to the present invention;

[0022] Figure 2 This is a schematic diagram of the structure of a digital technology-based image capture enhancement system according to the present invention. Detailed Implementation

[0023] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated in advance. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0024] This embodiment provides a digital technology-based image capture enhancement method and system, combined with... Figure 1 and Figure 2 As shown.

[0025] refer to Figure 1 A digital technology-based image enhancement method, comprising the following steps:

[0026] Identify scene objects in a virtual scene, obtain semantic information associated with scene objects, and activate one or more semantic contexts of scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with scene objects is non-visual attribute information corresponding to scene objects and used to characterize the semantic meaning of scene objects. Non-visual attribute information includes at least one of functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, texture layer, material layer, environment layer, semantic attribute layer, and interaction layer.

[0027] For each activated semantic context, an importance value is dynamically determined based on user attention, user interaction behavior, and the priority of the virtual tour narrative;

[0028] For each activated semantic context, an independent enhancement rendering layer is generated. The independent enhancement rendering layer is used to present the visual representation of the semantic context. The independent enhancement rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters.

[0029] The system receives independent enhancement rendering layers and their importance values. Based on the semantic region where the pixel is located and its importance value, the system fuses the independent enhancement rendering layers to obtain the final enhanced image.

[0030] The final enhanced image will be rendered and output in real time.

[0031] In this context, "virtual scene" refers to a collection of digital 3D scene data used to house scene objects and support real-time rendering and interaction. To facilitate semantic-driven graphics enhancement processing, virtual scenes can be organized into a multi-layered data structure, including at least: a geometry layer, a texture layer, a material layer, an environment layer, a semantic attribute layer, and an interaction layer. The geometry layer represents the spatial structure data of scene objects, including meshes, surfaces, point clouds, or combinations thereof. The texture layer represents the image texture data of the scene object's surface. The material layer represents the surface material rendering parameters of the scene object, including material properties such as reflection, roughness, metallicity, and transparency. The environment layer represents the rendering environment information, including a set of parameters affecting the overall visual performance of the scene, such as lighting parameters, ambient light, shadows, or fog effects. The semantic attribute layer records semantic attribute data associated with pixels or surface sampling points. Preferably, the semantic attributes can be encoded and stored as a multi-channel texture data structure that can be efficiently accessed by the graphics processor. For example, different semantic attributes or their encodings can be stored using the R, G, B, and A channels of the texture for quick lookup in subsequent real-time rendering stages. Interaction layer: Used to record user interaction behavior, user attention information, and virtual tour narrative clues and narrative priorities to support the dynamic determination of semantic context activation and importance values.

[0032] "Scene objects" refer to any identifiable and interactive entity in a virtual scene, such as buildings, sculptures, landforms, and cultural relics. These scene objects are typically acquired through techniques such as 3D modeling, laser scanning, or photogrammetry, and contain basic visual attributes such as geometry, texture, and material.

[0033] The "Identifying Scene Objects in a Virtual Scene" system can manage scene objects object-orientedly based on the organizational structure of virtual scene data. During the virtual scene loading phase, it reads the set of scene objects in the scene and establishes object description information for each scene object. This object description information includes at least a scene object identifier and corresponding geometric, texture, and material data. Preferably, it can also establish spatial bounding information (such as bounding boxes or bounding spheres) and hierarchical relationship information for scene objects to support subsequent rapid positioning and interaction hit judgment. During the runtime identification phase, the system combines user interaction behavior to locate and confirm scene objects: when a user interacts with the virtual scene through a terminal device by gazing, pointing, clicking, or voice triggering, the system acquires the interaction input and performs object hit judgment to determine the target scene object being interacted with. For example, it can perform intersection detection based on the user's line of sight or pointing ray and the spatial bounding information / geometric surface of the scene object to determine the identifier of the hit scene object, thereby completing the identification of the target scene object. This identification result can be further used to obtain the semantic information associated with the scene object and activate the corresponding semantic context.

[0034] Semantic information refers to non-visual attributes associated with objects in a scene, used to describe the meaning, function, historical background, and cultural value of those objects. For example, a stone might possess semantic information such as "geological sample," "ancient sacrificial relic," or "artistic sculpture." Semantic information can be obtained through methods such as manual annotation, knowledge graph association, or machine learning recognition.

[0035] "Semantic information associated with scene objects" refers to non-visual attribute information corresponding to scene objects and used to characterize the semantic meaning of scene objects. Non-visual attribute information includes at least one of functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. In this embodiment, the semantic information associated with scene objects is used to indicate "what" a scene object is, "what its purpose is," or "what meaning it represents in a specific semantic interpretation system," and can be selectively invoked or enhanced according to different application needs. For example, when the same scene object is a stone object, its visual form remains consistent, but the associated semantic information can be determined as any one or more of geological samples, ancient sacrificial relics, or artistic sculptures based on preset semantic annotations or external knowledge mapping relationships, thereby giving the scene object different semantic interpretation results.

[0036] "Obtaining semantic information associated with scene objects" can be implemented in the following ways (either individually or in combination): Manual annotation: During the virtual scene construction phase, object identifiers are established for scene objects in the virtual scene. Based on these identifiers, semantic annotations are performed on the scene objects to generate a set of semantic tags or semantic attributes that correspond one-to-one with each scene object. The semantic tags or set of semantic attributes must include at least one of the following: functional attributes, semantic category attributes, historical background attributes, or cultural value attributes, to form a "scene object-semantic attribute" association. Knowledge graph association: Based on the scene object's object identifier, object name, category tag, or externally searchable object index information, the entity retrieval and relationship mapping mechanism in the knowledge base / knowledge graph is invoked to map the scene object to the target knowledge entity. Furthermore, the attribute fields and relationship fields associated with the target knowledge entity are read, and the reading results are converted into semantic information associated with the scene object. Attribute fields may include usage, era, category, historical event association, cultural symbol association, etc., to achieve semantic expansion and consistent management of scene objects. Machine learning recognition and acquisition: During the virtual scene operation or content creation phase, geometric data, texture data, material parameters, or multi-view rendered images of scene objects are used to extract object features and input them into a pre-trained recognition or classification model. The output is semantic category, functional tendency, or theme attribute. The recognition model can be trained based on supervised learning or self-supervised learning, and the output results serve as semantic information associated with scene objects, or as a basis for verifying and supplementing manually labeled / knowledge graph association results. Structured storage and accessibility of semantic information: To ensure that semantic information can be efficiently called upon in subsequent rendering and fusion processes, the semantic information associated with scene objects can be further structured and encoded and written into the semantic attribute layer. In some implementations, semantic attributes can be written into multi-channel texture data according to preset encoding rules, enabling the graphics processor to quickly read the semantic attributes corresponding to pixels during the fragment shading stage, realizing semantic-driven real-time computation and enhancement processing.

[0037] "Semantic context" refers to the specific meaning or focus of attention that a scene object presents under specific user interaction behavior or virtual guided narrative clues. A scene object can be associated with multiple semantic contexts. For example, in the "text interpretation" mode, the semantic context of a stone tablet may focus on the "content of the inscription", while in the "geological investigation" mode, it may focus on the "rock type and degree of weathering".

[0038] The "activated semantic context" is not triggered arbitrarily. Only after meeting at least one triggering condition and passing the activation determination is the corresponding semantic context marked as "activated," and thus enter the generation process of the independent enhanced rendering layer. Specifically, the triggering conditions include at least one of the following: First, interaction triggering, triggered when user interaction with the target scene object or its semantic area is detected, including but not limited to clicking, gazing, pointing, gesture confirmation, or voice commands; Second, narrative triggering, triggered when the virtual tour narrative clues point to a narrative theme or focus and match the semantic information of the scene object, i.e., when the narrative script, explanatory text, or tour progress indicates the semantic content that should be highlighted, the semantic context corresponding to that semantic content is used as an activation candidate; Third, attention triggering, triggered when user attention meets preset conditions, such as the time the user's gaze lingers in the object's semantic area, gaze stability, or number of repeated glances meeting preset conditions, causing the semantic context associated with that area to enter the activation candidate set. Attention triggering can be used in conjunction with interaction triggering and narrative triggering to improve activation reliability. Upon triggering, the system infers the user's exploration intent and the focus of the virtual guided narrative based on the user's multimodal input data and the virtual guided narrative context. It then matches these with the multi-dimensional semantic identifiers of scene objects to obtain candidate semantic contexts. Further filtering based on semantic hierarchy and relevance yields the target semantic context, which is then evaluated for activation confidence. Activation confidence can be obtained from semantic matching scores, user behavior feature scores, narrative theme consistency scores, or a combination thereof. Preferably, when the activation confidence of the target semantic context is not lower than a preset activation condition, the target semantic context is determined to be "activated." When multiple target semantic contexts exist, all with high activation confidence and semantic conflicts occur, conflict resolution rules are first applied to obtain the final candidate semantic context after conflict resolution, which is then determined to be "activated." To ensure the stability and consistency of real-time rendering, once a semantic context is determined to be active, it can remain active for a preset time window. When user interaction focus shifts, narrative theme changes, or the activation confidence level is continuously lower than the preset exit condition, the activation flag of the semantic context is removed or it is downgraded to a candidate state, so as to reduce the computational overhead of irrelevant enhancement rendering layers and avoid visual interference.

[0039] An "independent enhancement rendering layer" refers to a visual enhancement layer generated for a specific semantic context, independent of the base scene rendering. Each independent enhancement rendering layer can include specific texture sharpening, color correction, detail reconstruction, highlighting, and other visual effects to highlight key information within that semantic context. Independent enhancement rendering layers can make the corresponding semantic content more visually prominent or richer in information through methods such as highlighting, texture enhancement, and information annotation. Semantic region mask sub-data: used to identify the effective area of ​​the independent enhancement rendering layer, corresponding to the semantic region division result; Enhancement parameter sub-data: used to describe the enhancement method and intensity of the independent enhancement rendering layer. "Text sharpening parameters" drive adjustments to rendering parameters related to the clarity of text / symbol edges (such as sharpness and contrast); "Texture enhancement parameters" drive adjustments to rendering parameters related to texture detail prominence; "Color correction parameters" drive adjustments to parameters related to color, saturation, and brightness; "Detail reconstruction parameters" and "Highlighting parameters" drive visual enhancement methods such as detail compensation or highlighting. These enhancement parameters can be used as input for subsequent fragment shaders to calculate pixel visual attributes and fusion weights.

[0040] The "Importance Value" is a numerical value that quantifies the importance of each activated semantic context in the current situation. This value comprehensively considers user attention, user interaction behavior, and the priority of the virtual navigation narrative, and serves as a weight parameter for independent enhanced rendering layers during the fusion stage, guiding the subsequent rendering fusion process. During the fusion process, the system obtains one or more independent enhanced rendering layers associated with a pixel based on its semantic region, and performs weighted fusion of the pixel contributions of each independent enhanced rendering layer according to their corresponding importance values. Among them, the independent enhanced rendering layer with a higher importance value has a greater visual contribution weight in the final enhanced image, thereby ensuring that important semantics are given priority and more prominent visual presentation.

[0041] In the step of "fusion of independent enhancement rendering layers", the fusion is preferably performed at the pixel level during the fragment shader stage of the graphics processor. For any pixel, the system determines the set of semantic regions to which the pixel belongs based on the semantic region segmentation result, and determines one or more independent enhancement rendering layers to participate in the fusion accordingly. When a pixel falls into only a single semantic region, the output visual attributes of the pixel are determined by the corresponding independent enhancement rendering layer. When a pixel falls into multiple semantic regions to form a semantically intertwined region, the system reads the semantic region mask sub-data of each independent enhancement rendering layer to obtain the mask contribution at the pixel, and generates a fusion weight by combining the importance value corresponding to each semantic context. Preferably, the fusion weight is normalized so that the independent enhancement rendering layer with a higher importance value occupies a larger visual contribution at the pixel. Subsequently, the system performs linear interpolation or weighted averaging on the pixel visual attributes of each independent enhancement rendering layer at the pixel to obtain the fusion result. When a conflict is detected between the enhancement types corresponding to multiple high-importance semantics, the relevant rendering parameters are coordinated and constrained according to a preset rendering rule set to achieve priority of primary semantics and smooth integration of secondary semantics, thereby outputting a visually continuous and prominent final enhanced image that meets the requirements of real-time rendering output.

[0042] The implementation environment of this application is typically a high-performance graphics rendering system, including a graphics processor, a central processing unit, memory, and various input devices (such as VR headsets, eye trackers, controllers, etc.). This system can process large amounts of graphics data in real time and supports complex rendering pipelines and multi-layer image fusion operations.

[0043] The core of the digital technology-based image enhancement method proposed in this application lies in the intelligent semantic recognition and enhanced rendering of scene objects in virtual scenes.

[0044] First, the process of identifying scene objects in a virtual environment, acquiring their associated semantic information, and activating one or more semantic contexts based on user interaction or virtual tour narrative clues can be implemented in various ways. For example, all scene objects in the virtual environment can be pre-annotated manually, with detailed semantic tags added to each object, such as "architectural structure," "cultural relic," and "natural landscape," further refined into sub-semantics, such as "walls," "roofs," and "bases" under "architectural structure." When a user looks at a scene object through a VR headset or points and clicks on it with a controller, the system can recognize the user's interaction and activate a specific semantic context related to that scene object based on preset rules or the user's current tour mode. For example, when a user clicks on a stone tablet, the system can activate the semantic context of "inscription interpretation" for subsequent enhanced display of the inscription. Another approach is to integrate a natural language processing module to analyze the virtual tour narrative script or user voice commands, extract keywords and themes, and match them with the semantic information of the scene objects to automatically activate the corresponding semantic context. For example, when the guided narrative mentions "ancient sacrificial rituals," the system can automatically activate all semantic contexts in the scene related to "sacrificial relics."

[0045] Secondly, in the step of dynamically determining an importance value for each activated semantic context based on user attention, user interaction behavior, and the priority of the virtual tour narrative, rule-based or machine learning methods can be employed. For example, a set of priority rules can be established: semantic contexts directly interacted with by the user (such as clicks or gazes) have the highest priority, followed by semantic contexts explicitly mentioned in the current virtual tour narrative, and finally areas with high user attention (such as eye-tracking data). Each priority can correspond to a preset initial importance value. When multiple semantic contexts are activated simultaneously, the system can comprehensively evaluate them according to these rules and dynamically adjust their importance values. For example, if a user is gazing at a stone tablet, and the current tour narrative is explaining the history of the tablet, then the importance values ​​of semantic contexts related to "inscription interpretation" and "historical background" will be significantly increased. Another implementation approach is that the system can train a machine learning model that takes user eye-tracking data, interaction logs, and tour narrative text as input features and outputs the importance value of each activated semantic context. The model can learn from historical data users' preferences for different semantics in different contexts, thereby more intelligently allocating importance.

[0046] Furthermore, in the step of generating an independent enhancement rendering layer for each activated semantic context, which is used to present the visual representation of the semantic context, multi-layer rendering techniques in the graphics rendering pipeline can be utilized. For example, for the activated "inscription interpretation" semantic context, the system can generate a dedicated rendering layer that sharpens, enhances contrast, and corrects the color of the inscription area to ensure the text is clearly legible. For the "geological survey" semantic context, another rendering layer can be generated, which may focus on highlighting the cracks, textures, and mineral colors of the rocks, and may even apply normal maps or displacement maps to enhance surface details. These independent enhancement rendering layers can employ different rendering techniques and parameters; for example, post-processing shaders can be used for sharpening and color adjustment, or texture details can be enhanced by modifying the material properties of scene objects. Each rendering layer is independent of the base scene rendering and only generates visual representations for its corresponding semantic context area.

[0047] Finally, the fusion process is crucial in the steps of receiving independent enhancement rendering layers and their importance values, and then fusing these layers based on the semantic region and importance value of each pixel to obtain the final enhanced image. For example, the system can implement the fusion logic in the fragment shader of the graphics processor. For each pixel on the screen, it first determines which semantic region(s) the pixel belongs to. If a pixel belongs to only one active semantic region, its final visual attributes will be primarily determined by the corresponding independent enhancement rendering layer and its importance value. If a pixel belongs to multiple active semantic regions (i.e., semantically intertwined regions), the system needs to perform weighted fusion based on the importance values ​​of these semantics and a preset fusion rule. For example, linear interpolation can be used to weight the visual contribution of different independent enhancement rendering layers to the pixel, with the weights determined by their respective importance values. The higher the importance value of a semantic layer, the greater its influence on the final pixel's visual attributes. For example, if a pixel is located in both the "text sharpening" and "geological texture enhancement" semantic regions, and the importance value of "text sharpening" is higher than that of "geological texture enhancement", then the final visual attributes of the pixel will be more inclined to the effect of text sharpening, but at the same time, the enhancement effect of geological texture will be incorporated in a soft way to avoid a harsh superposition.

[0048] The step of rendering the final enhanced image in real time involves presenting the blended image to the user's display device, such as a VR headset or monitor. This process requires ensuring a high frame rate and low latency to provide a smooth and immersive user experience.

[0049] This application further proposes steps for obtaining the final enhanced image, including:

[0050] The scene object surface is divided into semantic regions, and semantic attributes are assigned to pixels according to the divided semantic regions;

[0051] Store semantic attributes as multichannel texture data;

[0052] In the fragment shader of the graphics processor, semantic attributes are read from multi-channel texture data, and preliminary visual attributes of pixels are calculated and generated in real time based on semantic attributes and importance values.

[0053] When multiple highly important semantics intertwine on the same pixel, the rendering parameters on which the initial visual attributes of the pixel depend are adjusted collaboratively according to the preset rendering rule set. In particular, when text sharpening and geological texture enhancement conflict on the same pixel, the semantics with higher importance are processed first. By increasing the rendering parameters related to text sharpening and decreasing the rendering parameters related to geological texture, the geological texture enhancement effect is integrated in a soft way.

[0054] The final visual attributes of the pixels are regenerated based on the collaboratively adjusted rendering parameters, and the independent enhanced rendering layers are fused according to the final visual attributes of the pixels to obtain the final enhanced image.

[0055] Specifically, semantic region segmentation of scene object surfaces refers to the logical division of scene object surfaces in a virtual scene based on the semantic information they carry. For example, a building model can be divided into different semantic regions such as walls, windows, and roofs. Assigning semantic attributes to pixels based on the segmented semantic regions means that during the rendering process, each pixel is associated with specific attributes of its respective semantic region. These attributes can include semantic type, importance, interaction state, etc.

[0056] Storing semantic attributes as multi-channel texture data can be understood as encoding the semantic attributes assigned to pixels and storing them in a texture data structure that can be efficiently accessed by the graphics processor. For example, different semantic attributes or their encodings can be stored in the R, G, B, and A channels of the texture, with the aim of providing fast semantic information retrieval capabilities for subsequent real-time rendering.

[0057] In practical applications, in the fragment shader of a graphics processor, reading semantic attributes from multi-channel texture data and calculating and generating the preliminary visual attributes of a pixel in real time based on the semantic attributes and importance values ​​refers to using the parallel computing capabilities of the graphics processor to quickly obtain the semantic attributes of each pixel to be rendered during the fragment shading stage of the rendering pipeline, and combining the importance value of the semantic context corresponding to the pixel to preliminarily determine the visual performance of the pixel, such as color, brightness, and transparency.

[0058] Furthermore, when multiple highly important semantic elements intertwine on the same pixel, the coordinated adjustment of rendering parameters that determine the initial visual attributes of the pixel, based on a predefined set of rendering rules, means that during the rendering process, if multiple semantic contexts of high importance are detected on the same pixel (e.g., important text information superimposed on an important geological texture), the system will intelligently coordinate and modify the rendering parameters (such as color, transparency, sharpness, and blur) that affect the final visual appearance of the pixel according to a set of predefined rendering rules. For example, when text sharpening and geological texture enhancement conflict on the same pixel, if the importance of the text semantics is higher than that of the geological texture semantics, the rendering parameters related to text sharpening will be increased first to ensure clear and readable text, while the rendering parameters related to geological texture will be decreased so that the enhancement effect of the geological texture blends into the background in a soft and non-interfering way, avoiding visual conflict and confusion.

[0059] Therefore, based on the collaboratively adjusted rendering parameters, the final visual attributes of pixels are regenerated. The independent enhanced rendering layers are then fused according to these final visual attributes to obtain the final enhanced image. This means that after the rendering parameters are collaboratively adjusted, the system uses these optimized parameters to recalculate and determine the final visual representation of each pixel, ensuring a harmonious and clear visual effect even in semantically conflicting regions. Ultimately, these pixels with precise visual attributes are used to fuse the various independent enhanced rendering layers, thereby generating a high-quality final enhanced image.

[0060] This application's solution provides rich semantic context information for each pixel by introducing refined semantic region segmentation and multi-channel texture data storage. In the fragment shader of the graphics processor, the visual attributes of the pixel can be initially calculated in real time based on these semantic attributes and dynamically determined importance values. More importantly, when the system detects multiple highly important semantics intertwined at the same pixel, it no longer uses simple superposition or averaging, but instead adjusts the rendering parameters affecting the pixel's visual performance collaboratively according to a preset set of rendering rules. This collaborative adjustment mechanism intelligently balances the importance of different semantics. For example, when text sharpening and geological texture enhancement conflict, it dynamically adjusts rendering parameters to ensure the clarity of important information (such as text) while subtly integrating secondary information (such as geological texture), effectively solving the visual confusion and information ambiguity problems that may occur when traditional fusion methods handle complex semantic conflicts. Finally, the final visual attributes of the pixel are regenerated based on these collaboratively adjusted rendering parameters, ensuring the visual harmony and accurate information delivery of the final enhanced image.

[0061] In some preferred embodiments, "segmenting the scene object surface into semantic regions and assigning semantic attributes to pixels based on the segmented semantic regions" includes the following steps: First, acquiring multi-source surface description data of the scene object surface, wherein the multi-source surface description data includes at least multispectral data and geometric / depth data corresponding to the scene object surface; wherein, the multispectral data is used to characterize the spectral response characteristics of the scene object surface in different bands, and the geometric / depth data is used to characterize the spatial location and depth level of points on the scene object surface. Second, performing initial segmentation of candidate semantic regions on the scene object surface based on the spectral feature differences of the multispectral data, specifically: extracting spectral feature vectors from the sampling points / texture sampling points of the scene object surface and calculating the spectral difference degree between adjacent sampling points; when the spectral difference degree meets the preset boundary judgment condition, marking the corresponding position as the semantic region boundary, thereby obtaining the boundary set of candidate semantic regions. Furthermore, for candidate semantic regions that overlap or intertwine, depth information of intertwined regions is extracted by combining geometric / depth data, and the intertwined regions are separated based on "spectral feature differences + depth information" to obtain multiple independent semantic regions. Preferably, an independent geometric proxy is generated for each separated semantic region, so that different semantic regions have independent spatial carrying units during subsequent rendering and attribute mapping, thereby avoiding semantic crosstalk between intertwined regions. Then, the above semantic regions are mapped from object space to texture space or screen space to achieve pixel-level semantic assignment: a corresponding semantic region mask is generated for each semantic region (or its geometric proxy), and semantic attribute encoding is written in pixels in the off-screen buffer or semantic attribute map; when a pixel falls into a semantic region mask, the semantic type, level identifier, interaction state identifier, or other semantic attributes corresponding to the semantic region are written for the pixel; when a pixel falls into multiple semantic region masks at the same time, the primary semantic attribute is determined and the secondary semantic attribute is recorded according to the preset priority rules or conflict resolution rules, or multiple semantic attributes are encoded and stored separately in a multi-channel manner to support the subsequent collaborative rendering of semantically intertwined regions. Finally, the pixel semantic attributes are stored in the form of multi-channel texture data. Preferably, different semantic attributes or their encodings are stored in different channels of the texture, so that the graphics processor can read the pixel semantic attributes in real time during the fragment shader stage and calculate the visual attributes and fusion weights of the pixel in combination with the importance value.

[0062] In some preferred embodiments, suppose a user is observing an ancient stone wall in a virtual historical site tour scenario. This stone wall is inscribed with important historical inscriptions (text sharpening semantics), while the wall itself possesses complex weathered geological textures (geological texture enhancement semantics). When the user focuses on this stone wall through eye movement or interactive behavior, the system recognizes that these two highly important semantic contexts—text sharpening and geological texture enhancement—are intertwined within the same pixel area.

[0063] According to the scheme of this application, the surface of the stone wall is first divided into semantic regions, separating the inscribed area from the non-inscribed area, and each pixel is assigned a corresponding semantic attribute. These semantic attributes are stored as multi-channel texture data. In the fragment shader of the graphics processor, these semantic attributes and their respective importance values ​​are read in real time. Assume that at this time the system determines that the semantic importance of text sharpening is higher than that of geological texture enhancement (e.g., the user explicitly clicked on the inscription or the navigation narrative emphasized the inscription content).

[0064] At this point, the system collaboratively adjusts the rendering parameters affecting the final visual appearance of the pixel based on a preset set of rendering rules. Specifically, rendering parameters related to text sharpening (such as contrast and edge sharpness) are increased to ensure the inscription is clearly visible and easy to read on the screen. Simultaneously, rendering parameters related to geological texture enhancement (such as texture detail intensity and sharpness) are appropriately reduced, allowing the enhanced geological texture to blend into the background smoothly and seamlessly, avoiding visual conflict with the clarity of the inscription. Finally, based on these collaboratively adjusted rendering parameters, the final visual attributes of the pixel are regenerated and integrated into the final enhanced image. In this way, users can clearly read the inscription content and experience the ancient and realistic geological texture of the stone wall without the text being blurred by the texture or the texture being abruptly covered by the text, thus providing a high-quality and information-rich visual experience.

[0065] This application further proposes a step for dynamically determining an importance value, including:

[0066] Collect user eye movement data and smooth the user eye movement data to obtain smoothed eye movement data;

[0067] Identify areas on scene objects where visual details are blurred due to physical factors, and pre-associate these blurred areas with potentially conflicting semantics to obtain association results;

[0068] Based on the correlation results, when the user's gaze falls on an area with blurred visual details, micro-eye movement pattern features are extracted from smooth eye movement data, and the eye movement behaviors corresponding to the micro-eye movement pattern features are classified into different sub-modes.

[0069] Map submodals to specific user intent semantics and dynamically adjust the priority of the mapping based on the narrative theme of the current virtual tour;

[0070] Based on the semantics of user intent and the priority of the adjusted mapping, assign importance values ​​to the semantic contexts activated in regions where visual details are blurred.

[0071] Specifically, collecting user eye-tracking data refers to acquiring raw eye movement trajectory information such as fixation points, saccades, and gaze movements in a virtual scene in real time using eye-tracking devices, such as infrared eye trackers or camera-based eye-tracking systems. Smoothing the user eye-tracking data aims to eliminate noise and jitter that may exist in the raw data. Algorithms such as moving averages, Gaussian filtering, or Kalman filtering are used to obtain more stable and reliable smoothed eye-tracking data for subsequent pattern recognition.

[0072] Identifying areas where visual details are blurred due to physical factors can be understood as using image processing techniques or pre-defined scene metadata to identify areas in the scene where visual information is unclear or difficult to discern due to factors such as distance, lighting, material, and rendering effects. Examples include a distant sculpture, an inscription in shadow, or a worn textured surface. Pre-associating these areas with potentially conflicting semantics aims to address the fact that multiple potential semantic pieces of information may exist within these blurred areas, and these semantic pieces of information may compete with or be ambiguous. For example, a blurred mural may simultaneously contain both historical events and artistic styles. This pre-association provides a foundation for subsequent semantic disambiguation.

[0073] In practical applications, based on the association results, when a user's gaze falls on an area with blurred visual details, extracting micro-eye movement pattern features from smoothed eye movement data refers to the micro-features that differ from typical saccades in the user's eye movement behavior when observing a blurred area. These features include pupil dilation, prolonged fixation time, reduced saccade amplitude, and repetitive saccades. These micro-features can be extracted and quantified using specific algorithms. The eye movement behaviors corresponding to these micro-eye movement pattern features are then classified into different sub-patterns. For example, continuous micro-saccades may indicate that the user is attempting to discern details, while pupil dilation may be associated with increased cognitive load or interest. These sub-patterns reflect the different cognitive strategies and intentions of users when faced with ambiguous information.

[0074] Furthermore, mapping sub-patterns to specific user intent semantics refers to associating identified microscopic eye-tracking sub-patterns with potential user intent semantics such as exploration, recognition, understanding, and comparison, based on a pre-defined rule base or machine learning model. For example, repeated scanning accompanied by pupil dilation might be mapped to the intent semantic of "attempting to recognize details." The priority of these mappings is dynamically adjusted according to the narrative theme of the current virtual tour, aiming to ensure that the allocation of importance values ​​remains consistent with the overall goals of the tour. For instance, in a history-themed tour, the intent semantic priority of "recognizing historical inscriptions" would be higher than that of "appreciating artistic styles."

[0075] Therefore, based on the user's semantic intent and the priority of the adjusted mapping, importance values ​​are assigned to the semantic context activated in areas with blurred visual details. Specifically, when the user expresses semantic intent highly relevant to the navigation topic, the corresponding semantic context in that area will be assigned a higher importance value, resulting in a more significant visual representation in subsequent enhanced rendering.

[0076] The solution proposed in this application, by introducing the analysis of user eye-tracking data, especially micro-eye-tracking patterns, and the identification and pre-association of blurred areas of visual details in the scene, can gain a deeper understanding of the user's true focus and potential intentions when faced with complex or uncertain visual information.

[0077] In some preferred embodiments, imagine a user viewing an ancient mural during a virtual museum tour. Due to its age and dim lighting, details of the text and patterns in some areas of the mural are very blurry.

[0078] If the system only considers the user's overall gaze time or click behavior on the mural, it may only be able to determine that the user is interested in the mural as a whole, but it cannot know whether the user specifically wants to see the blurry inscriptions on the mural or wants to understand the overall artistic style of the mural.

[0079] The proposed solution first identifies the blurred inscription areas on the mural as areas of visual detail obscuration. When the user's gaze falls on these areas, an eye-tracking device collects the user's eye movement data. If the system detects that the user exhibits continuous microscopic eye movement patterns such as small saccades and pupil dilation in these blurred areas, these patterns are categorized as a sub-mode of "attempting to recognize details." Combined with the current virtual tour's narrative theme of "historical research," this sub-mode of "attempting to recognize details" is mapped to the user's semantic intent of "interpreting historical inscriptions" and given higher priority.

[0080] Based on this, the system assigns a higher importance value to the semantic context of the "historical inscription" activated within the blurred inscription area. Subsequently, when generating and blending independent enhanced rendering layers, the visual performance of the inscription area (e.g., through sharpening, contrast enhancement, or highlighting) will be significantly enhanced, while the enhancement effects on other non-inscription areas of the mural may be relatively softer. This precisely satisfies the user's deeper intention to try to identify the blurred inscription and provides a more targeted visual enhancement experience.

[0081] In some embodiments described above in this application, the step of dynamically determining an importance value includes:

[0082] Identify conflicts between user attention, user interaction behavior, and the narrative priority of virtual guided tours;

[0083] Based on the pre-set conflict resolution rules, the conflict is coordinated to arrive at a coordination result;

[0084] The importance level is dynamically determined based on the coordination results.

[0085] Specifically, identifying conflicts among user attention, user interaction behavior, and virtual tour narrative priority refers to the system comparing and analyzing the importance of the semantic contexts indicated by these three key factors. For example, when user eye-tracking data shows high attention to a non-narrative core detail in a scene, while the virtual tour's narrative priority points to another main object, and the user's recent interactions are concentrated in a third area, a potential conflict between these factors is identified. This conflict may manifest as assigning high importance to different semantic contexts, or as significantly different assessments of the importance of the same semantic context.

[0086] The process of resolving conflicts according to pre-defined rules and achieving a coordinated result can be understood as follows: once a conflict is identified, the system resolves the contradictions based on a pre-set set of rules. These rules may include priority ranking (e.g., narrative priority is higher than user attention in a specific context), weighted averaging (dynamically adjusting the weights of factors based on context), or decision-making based on machine learning models (training the model to learn how to make the best decision in different conflict situations). For example, when user attention conflicts with narrative priority, the rules might stipulate that the weight of user attention increases when the user stays in a certain area for a long time, while the weight of narrative priority increases at key narrative nodes. By applying these rules, the system can effectively coordinate conflicts, thereby arriving at a comprehensive and contradictory coordinated result that reflects the true importance of each semantic context in the current situation.

[0087] In practical applications, dynamically determining the importance value based on the coordination results means that the system uses the aforementioned coordination results as input and assigns a final importance value to each activated semantic context. This importance value is after conflict resolution, thus more accurately reflecting a comprehensive consideration of the current user intent, the system's navigation goals, and the user's actual behavior. For example, if the coordination results indicate that a certain semantic context should be given high priority after considering all factors, then its importance value will be set to a higher level, and vice versa.

[0088] This application's solution effectively compensates for the shortcomings of simply aggregating user attention, user interaction behavior, and virtual tour narrative priority by introducing a conflict identification and resolution mechanism. Specifically, when these factors disagree on the importance assessment of the same or different semantic contexts, traditional direct aggregation methods may fail to yield optimal or consistent importance values, and could even lead to decision biases in the system's enhanced rendering. By first identifying these potential conflicts, the system can clearly locate the situations requiring special handling. Subsequently, with the help of preset conflict resolution rules, the system can intelligently coordinate these conflicts according to the specific context (e.g., the user's current mood, tour stage, characteristics of scene objects, etc.), ensuring that a logically consistent and context-aware importance value is assigned to each activated semantic context under multiple influencing factors. This avoids the confusion or inaccuracy of enhanced rendering effects caused by contradictions between factors, allowing subsequent enhanced rendering layer fusion processes to be based on more reliable input.

[0089] In some preferred embodiments, suppose a user is viewing an ancient painting in a virtual museum tour scenario. At this time, the system identifies the following: User attention: Eye-tracking data shows that the user's gaze lingers on a signature in the lower right corner of the painting for an extended period, indicating a high level of interest in this detail. User interaction behavior: The user has just clicked the "Background Story" button next to the painting but has not yet finished reading it, indicating interest in the overall background information of the painting. Virtual tour narrative priority: The current tour's narrative focus is on introducing the painting's overall artistic style and historical background, rather than a specific detail or signature.

[0090] In this scenario, a conflict arises between user attention (signature) and the narrative priority of the virtual tour (overall art style), while user interaction (background story) also points to a broader semantic meaning. The proposed solution first identifies these conflicts. Then, based on pre-defined conflict resolution rules—for example, rules might stipulate that when a user actively interacts and the narrative priority is high, the weight of narrative priority and interaction behavior will temporarily increase, while the weight of user attention (if it's only brief attention to detail) will relatively decrease. Through coordination, the system arrives at a result where the semantic context related to the overall art style and background story should be prioritized for enhancement, while the user's attention to the signature should be considered in a subtle way. Based on this coordination result, the system dynamically determines the importance value of each semantic context; for example, setting the importance value of the overall art style and background story to high, and the importance value of the signature to medium-high, thereby guiding subsequent enhancement rendering and ensuring that the final enhanced image not only conforms to the main tour theme but also appropriately responds to the user's detailed interests.

[0091] The steps to activate one or more semantic contexts of a scene object include:

[0092] The scene object surface is divided into semantic regions, and multi-dimensional semantic labels are attached to the divided semantic regions;

[0093] Collect user multimodal input data and obtain the virtual tour narrative context;

[0094] Based on user multimodal input data and the virtual guided narrative context, infer the user's exploration intent and the focus of the guided narrative;

[0095] By matching user exploration intent, navigation narrative focus, and multi-dimensional semantic identifiers of scene objects, candidate semantic contexts are obtained.

[0096] Based on the granularity of the user's exploration intent and the focus of the navigation narrative, the granularity of the candidate semantic context is adaptively adjusted to obtain the granularized candidate semantic context.

[0097] Based on the hierarchical relationship and relevance of semantics, target semantic contexts related to the current user's exploration intent and the focus of the navigation narrative are selected from the candidate semantic contexts after granular adjustment;

[0098] Based on the consistency between the target semantic context and the user's multimodal input data and the virtual tour narrative context, the activation confidence of the target semantic context is evaluated.

[0099] When multiple target semantic contexts have high activation confidence and semantic conflicts exist, conflict resolution rules are used to process the target semantic contexts to obtain the final candidate semantic contexts after conflict resolution.

[0100] Based on the final candidate semantic context after conflict resolution, activate one or more semantic contexts of the scene object.

[0101] Specifically, semantic region segmentation of scene objects refers to visually or logically subdividing scene objects in a virtual scene. For example, an architectural scene object can be divided into different semantic regions such as walls, windows, doors, and roofs. Adding multi-dimensional semantic labels to these segmented semantic regions can be understood as assigning labels to these subdivided regions containing information such as semantic type (e.g., "architectural structure," "decorative element"), hierarchy (e.g., "windows" under "overall building"), spatial location (e.g., three-dimensional coordinates or relative position), and semantic associations determined based on spatial adjacency (e.g., "windows" and "walls" are adjacent). The purpose is to provide a refined data foundation for subsequent semantic context matching and activation.

[0102] This includes collecting multimodal user input data, which can include, but is not limited to, various forms of input such as voice commands, gesture recognition, eye-tracking data, head posture, and keyboard and mouse operations. Obtaining the narrative context of the virtual tour refers to acquiring information such as the current progress, theme, content already explained, and content to be explained. Its purpose is to comprehensively understand the user's current intentions and the overall direction of the tour.

[0103] In practical applications, based on user multimodal input data and the context of the virtual guided narrative, the user's exploration intent and the focus of the guided narrative can be inferred. For example, when a user gazes at a sculpture and issues the voice command "What is this?", it can be inferred that the user's exploration intent is to learn about the sculpture's background information, and the focus of the guided narrative may be on the sculpture's historical or artistic value. The aim is to accurately grasp the user's focus and interests in the virtual environment.

[0104] Matching user exploration intent, narrative focus, and multi-dimensional semantic identifiers of scene objects to obtain candidate semantic contexts involves comparing the inferred user intent and narrative focus with pre-attached semantic identifiers on scene objects to identify all potentially relevant semantic regions and their associated information. For example, if the user intent is to understand the "structure of ancient buildings," then all regions with the semantic identifier "architectural structure" will be matched.

[0105] Furthermore, based on the granularity of the user's exploration intent and the focus of the navigation narrative, the granularity of the candidate semantic context is adaptively adjusted to obtain a granularized candidate semantic context. For example, if the user's intent is to understand the "overall architectural style" macroscopically, the granularity of the candidate semantic context is adjusted to "the entire building"; if the user's intent is to understand the "details of the carvings on the windows" microscopically, the granularity is adjusted to "a specific area of ​​the window". The purpose is to ensure that the activated semantic context matches the granularity of the user's focus.

[0106] Based on the hierarchical relationship and relevance of semantics, the target semantic context that is relevant to the current user's exploration intent and the focus of the navigation narrative is selected from the candidate semantic contexts after the granularity is adjusted. This means that after adjusting the granularity, further refinement is carried out by utilizing the hierarchy (e.g., "building" contains "window") and relevance (e.g., "window" is related to "wall") between semantics to ensure that the final selected semantic context is the most relevant and meaningful.

[0107] Specifically, assessing the activation confidence of the target semantic context based on its matching consistency with user multimodal input data and the virtual navigation narrative context involves quantifying the degree of fit between the target semantic context and the user input and narrative context. For example, a higher matching degree corresponds to a higher confidence level. The purpose is to provide a quantitative basis for subsequent activation decisions.

[0108] When multiple target semantic contexts have high activation confidence and semantic conflicts exist, conflict resolution rules are used to process the target semantic contexts, resulting in the final candidate semantic context after conflict resolution. For example, when a user is simultaneously interested in the artistic style and material composition of a sculpture, and these two aspects may conflict visually, priority judgment and processing are required based on rules. The purpose is to resolve ambiguity and visual confusion during multiple semantic activations.

[0109] Therefore, activating one or more semantic contexts of a scene object based on the final candidate semantic contexts after conflict resolution refers to determining which semantic regions need to be activated and enhanced in rendering based on the final candidate semantic contexts after screening and conflict handling. The purpose is to ensure that the activated semantic contexts are accurate, conflict-free, and highly consistent with the user's intent.

[0110] This application's solution lays the foundation for subsequent semantic recognition by finely dividing the surface of scene objects into semantic regions and attaching multi-dimensional semantic labels. By collecting user multimodal input data and virtual navigation narrative context, it can comprehensively and accurately infer the user's exploration intention and the focus of the navigation narrative. Furthermore, the inferred user intention and narrative focus are matched with the multi-dimensional semantic labels of scene objects, and the granularity of candidate semantic contexts is adaptively adjusted according to the granularity of user attention, ensuring that the activated semantic contexts precisely match the user's specific needs, avoiding information overload or insufficient information. Filtering based on semantic hierarchy and relevance, and evaluating the activation confidence of target semantic contexts, ensures the accuracy and relevance of the selected semantic contexts. When multiple high-confidence semantic contexts exist and conflicts exist, they are handled through preset conflict resolution rules, ultimately achieving precise and intelligent activation of one or more semantic contexts of scene objects.

[0111] In some embodiments of this application described above, the steps of dividing the surface of a scene object into semantic regions and attaching multi-dimensional semantic identifiers to the divided semantic regions include:

[0112] Acquire multispectral data of the surface of scene objects;

[0113] By utilizing the differences in spectral features in multispectral data, the boundaries of overlapping or interwoven semantic regions on the surface of scene objects can be identified;

[0114] Based on the boundaries of the overlapping or intertwined semantic regions, extract the depth information of the overlapping or intertwined semantic regions;

[0115] Based on spectral feature differences and depth information, overlapping or intertwined semantic regions are separated, and independent geometric proxies are generated for each separated semantic region.

[0116] Each independent geometric agent is attached with a multi-dimensional semantic identifier, which includes semantic type, level, spatial location, and semantic relevance determined based on spatial adjacency.

[0117] Check for conflicts in the additional multi-dimensional semantic identifiers;

[0118] When a conflict is detected, the multi-dimensional semantic identifier of the conflict is adjusted according to the preset conflict resolution rules to ensure the independence and non-conflictability of the multi-dimensional semantic identifier.

[0119] Acquiring multispectral data of scene object surfaces refers to collecting the reflection or emission spectrum information of scene object surfaces in a virtual scene using multispectral sensors. This data contains light intensity information at different wavelengths, revealing the physical and chemical properties of the scene object surface materials. By utilizing the differences in spectral features in the multispectral data, the boundaries of overlapping or interwoven semantic regions on the scene object surface can be identified. For example, regions of different materials or in different states will exhibit unique characteristics in their spectral responses, and these differences can be used by algorithms to accurately define the physical boundaries of semantic regions.

[0120] Furthermore, depth information of the overlapping or intertwined semantic regions is extracted based on their boundaries. Depth information is crucial for distinguishing semantic regions that may visually overlap but are actually at different depths. Depth information can be obtained from various sources, such as depth sensors, stereo vision algorithms, or geometric models of the scene. Based on spectral feature differences and depth information, overlapping or intertwined semantic regions can be effectively separated. After separation, independent geometric proxies are generated for each separated semantic region. These proxies can be simplified 3D models or 2D planar representations, used for subsequent semantic labeling and processing.

[0121] Each independent geometric agent is attached with a multi-dimensional semantic identifier, which includes semantic type, hierarchy, spatial location, and semantic relevance determined based on spatial adjacency. Semantic type can refer to the category of an object (e.g., "rock," "vegetation," "water"), hierarchy indicates its position in the semantic structure (e.g., "peak" under "mountain range"), spatial location provides precise geometric positioning, and semantic relevance describes the relationship between this region and other semantic regions (e.g., the adjacency relationship between "river" and "riverbank"). These multi-dimensional identifiers together construct a comprehensive understanding of the surface semantics of scene objects.

[0122] Building upon this, it is necessary to check for conflicts in the additional multi-dimensional semantic identifiers. Conflicts may arise from data acquisition errors, semantic parsing ambiguities, or inconsistencies between different semantic sources. When a conflict is detected, the multi-dimensional semantic identifiers are adjusted according to pre-defined conflict resolution rules to ensure their independence and conflict-free nature. These rules can resolve ambiguities based on priority, confidence level, or contextual information, thereby guaranteeing the accuracy and consistency of the semantic identifiers.

[0123] This application further proposes the following steps for evaluating the activation confidence of the target semantic context:

[0124] Based on user multimodal input data and virtual tour narrative context, feature matching is performed on the target semantic context to obtain the initial activation confidence.

[0125] When the initial activation confidence of multiple target semantic contexts is at a medium level and there is semantic ambiguity, analyze the physical interleaving characteristics of the corresponding regions of the target semantic contexts;

[0126] Based on the physical interweaving characteristics, and combined with the user's historical behavior patterns and preferences for virtual tour narrative themes, the initial activation confidence is corrected, and the corrected activation confidence is output.

[0127] Specifically, "initial activation confidence" refers to a quantitative value obtained through preliminary feature matching, reflecting the degree of matching between the target semantic context and the user's exploration intent and the focus of the navigation narrative. When this value is at a moderate level, it indicates that the matching result is ambiguous, and there may be multiple semantic contexts that are somewhat related but cannot be clearly distinguished. "Semantic ambiguity" can be understood as two or more semantic contexts having similar activation probabilities given user input and narrative background, making it difficult for the system to determine which semantic context the user is truly interested in. In practical applications, "analyzing the physical interweaving characteristics of the regions corresponding to the target semantic contexts" refers to in-depth analysis of visual regions in the virtual scene related to these target semantic contexts. For example, checking whether these regions overlap spatially, are closely adjacent, or have visual occlusion relationships. This analysis helps to understand the complexity of semantic contexts in physical space, providing an objective basis for subsequent confidence correction. Furthermore, "combining user historical behavior patterns" refers to considering information such as the user's habits, preferences, and exploration paths when interacting with the virtual environment in the past, such as whether the user frequently focuses on specific types of objects or areas. "Preference for virtual tour narrative themes" refers to the core content or storyline emphasized in the current virtual tour. For example, if the tour theme is "ancient architecture," then semantic context related to architecture may be given higher weight. Therefore, "correcting the initial activation confidence" means adjusting the original moderate level of activation confidence after considering physical interweaving characteristics, user historical behavior patterns, and virtual tour narrative theme preferences, to more accurately reflect actual user intent and scene relevance. The corrected confidence will be output as "corrected activation confidence" and used for subsequent semantic context activation decisions.

[0128] This application's solution effectively addresses the potential inaccuracy of initial assessments by incorporating analysis of the physical interweaving characteristics of the target semantic context, user historical behavior patterns, and virtual tour narrative theme preferences when initial activation confidence is at a moderate level and semantic ambiguity exists. Specifically, physical interweaving characteristics provide objective information about the spatial relationships between semantic contexts within a scene, helping to distinguish semantics that are visually difficult to separate. User historical behavior patterns reflect the user's potential interests from a personalized perspective, enabling the system to make judgments based on the user's long-term preferences. Preferences for virtual tour narrative themes ensure the consistency between the activated semantic context and the current tour's main theme from a global perspective. By comprehensively considering these multi-dimensional information, the system can more comprehensively and accurately understand user intent and scene context, thereby finely refining the initial activation confidence and avoiding activation errors or omissions due to insufficient matching of a single element.

[0129] In some preferred embodiments, suppose a user is taking a guided tour in a virtual museum, with the current narrative theme being "Ancient Egyptian Civilization." The user expresses interest in a display case through gestures and voice input (user multimodal input data). This display case contains a "Pharaoh Statue" and a "Hieroglyphic Scroll," which are physically adjacent, and the user's gaze frequently shifts between them. Initial system analysis reveals that the initial activation confidence levels of both the "Pharaoh Statue" and the "Hieroglyphic Scroll" are at a moderate level, indicating semantic ambiguity. At this point, the solution proposed in this application will be applied. First, the system analyzes the physical interweaving characteristics of the corresponding areas of the "Pharaoh Statue" and the "Hieroglyphic Scroll," finding that they are closely juxtaposed in the display case and visually difficult to completely separate. Second, the system considers the user's historical behavior patterns; for example, if the user has repeatedly clicked on or inquired about exhibits related to "text" during previous tours, it indicates a high preference for textual content. Simultaneously, the system will consider the current virtual tour narrative theme's preferences; for example, "Ancient Egyptian Civilization" might focus more on its unique cultural symbols and writing system. Incorporating this information, the system will adjust the initial activation confidence. Even if the "Pharaoh Statue" might be more visually prominent, considering the user's historical preference for the writing system and the tour's emphasis on cultural symbols, the system might increase the activation confidence of the "Hieroglyphic Scroll" and correspondingly decrease the confidence of the "Pharaoh Statue," or establish a stronger connection between the two, ensuring that the final activated semantic context better aligns with the user's deeper exploration intentions and the tour's narrative focus. Ultimately, the system will output the adjusted activation confidence and activate the semantic context of the "Hieroglyphic Scroll" accordingly, for example, displaying detailed information about the history and interpretation of hieroglyphs, rather than simply focusing on the statue's artistic value.

[0130] This application further proposes steps for conflict handling of the target semantic context according to preset conflict resolution rules, including:

[0131] Collect user physiological data, including heart rate variability and skin conductance response;

[0132] The intensity and nature of the conflict are quantified based on the semantic type differences, activation confidence differences, visual complexity of the conflict area, and the narrative focus of the current virtual tour, among the target semantic contexts with semantic conflicts.

[0133] Infer the user's current emotional state based on the user's physiological data;

[0134] Based on the intensity and nature of the conflict and the user's current emotional state, conflict resolution rules are dynamically selected from a preset conflict resolution strategy library;

[0135] Conflicts are resolved according to conflict resolution rules.

[0136] Specifically, collecting user physiological data refers to acquiring data on the user's physiological responses in a virtual environment through sensors. This physiological data can include heart rate variability and conductance of skin (CSE). Heart rate variability reflects the activity state of the autonomic nervous system and can be used to assess a user's stress level, concentration, or relaxation level. Conductance of skin (CSE) is related to sweat gland activity and can sensitively reflect a user's emotional arousal level and psychological tension. This physiological data is used to objectively assess the user's internal state when facing semantic conflict.

[0137] Furthermore, quantifying the intensity and nature of conflict refers to conducting a multi-dimensional analysis of conflicts between different semantic contexts. Specifically, this includes analyzing the semantic type differences between target semantic contexts where semantic conflict exists, such as whether the conflict is at the content level or the visual presentation level; activation confidence differences, i.e., the degree of certainty that different conflicting semantics are activated; visual complexity of the conflict area, such as whether the conflict occurs in a visually dense or sparse area; and the narrative focus of the current virtual tour, i.e., which theme or information the current tour emphasizes. By integrating these factors, a precise numerical assessment of the severity and specific manifestations of the conflict can be achieved.

[0138] Inferring a user's current emotional state based on physiological data involves using collected physiological data such as heart rate variability and skin conductance response, combined with machine learning models or pre-defined physiological-emotion mapping rules, to identify whether the user is currently in a positive, negative, calm, tense, or confused emotional state. For example, high heart rate variability may indicate that the user is relaxed or focused, while low heart rate variability may be associated with stress or fatigue; a sudden increase in skin conductance response may indicate that the user is emotionally stimulated or surprised.

[0139] In practical applications, conflict resolution rules are dynamically selected from a pre-defined conflict resolution strategy library based on the intensity and nature of the conflict and the user's current emotional state. This library contains a collection of predefined conflict handling schemes, each tailored to a specific combination of conflict intensity, nature, and user emotional state. Dynamic selection means the system does not employ a single, fixed rule, but intelligently matches and activates the most appropriate resolution strategy based on real-time assessments of the conflict situation and the user's emotions. For example, when the conflict intensity is high and the user is emotionally stressed, a quick and explicit resolution rule might be chosen; while when the conflict intensity is low and the user is calm, a gentler, more gradual resolution rule might be selected.

[0140] In some preferred embodiments, suppose a user is visiting an ancient mural in a virtual museum. The mural simultaneously contains archaeological semantics regarding its creation date and art historical semantics regarding its artistic style, and both semantic contexts have high activation confidence, potentially causing a conflict in visual enhancements within the same area (e.g., dating information requires sharp text, while artistic style requires soft textures). In this case, the system first collects the user's heart rate variability and conductance per skin response (CPS). If low CPS and elevated CPS are detected, this may indicate confusion or mild anxiety regarding the current conflict. Simultaneously, the system quantifies the conflict: semantic type difference (text vs. texture), activation confidence difference (both high), visual complexity of the conflicting area (rich mural detail), and current narrative focus (e.g., the current tour emphasizes art appreciation rather than archaeological dating). Based on this information, the system infers that the user may be in a state of mild stress or confusion. Therefore, the system dynamically selects a "flexible integration" strategy from its conflict resolution strategy library. This strategy may instruct the system to visually prioritize the presentation of art style semantics more relevant to the current narrative focus (art appreciation), while seamlessly integrating historical information in a non-obtrusive manner (e.g., by reducing transparency or using gradient effects) to avoid jarring visual clashes. This dynamic and user-emotion-aware approach ensures that users receive a smooth and personalized virtual tour experience without being disturbed.

[0141] This application further proposes steps for handling the conflict, including:

[0142] Collect data on the intensity and nature of the conflict;

[0143] Collect data on the user's current emotional state;

[0144] The intensity and nature of the conflict data are normalized to obtain normalized conflict data.

[0145] Normalize the user's current emotional state data to obtain normalized emotional data;

[0146] Normalized conflict data and normalized sentiment data are used as inputs. Through interpolation, a smooth transition is achieved between adjacent rules in a pre-defined conflict resolution strategy library to generate intermediate resolution rules.

[0147] Conflicts are resolved according to intermediate resolution rules.

[0148] Specifically, collecting data on the intensity and nature of conflict refers to obtaining specific numerical values ​​of the conflict intensity and nature quantified in the above steps. These data reflect the severity and type of semantic conflict. Simultaneously, collecting data on the user's current emotional state refers to obtaining information about the user's emotional state inferred from the above steps, such as whether the user is calm, curious, confused, or anxious. This data forms the basis for subsequent refined conflict handling.

[0149] The process involves normalizing the intensity and nature of conflict data to obtain normalized conflict data, and normalizing the user's current emotional state data to obtain normalized emotional data. The purpose is to unify data of different dimensions or ranges into standardized intervals, such as [0,1] or [-1,1]. Normalization helps eliminate differences in dimensions between data points, ensuring the comparability of conflict intensity, nature, and emotional state in subsequent interpolation calculations. It also prevents certain data from dominating the calculation results due to excessively large numerical ranges, thereby improving the accuracy and fairness of the interpolation results.

[0150] In practical applications, normalized conflict data and normalized sentiment data are used as input. Interpolation methods are employed to smoothly transition between adjacent rules in a pre-defined conflict resolution strategy library, generating intermediate resolution rules. Interpolation can be understood as a mathematical technique used to estimate the value of unknown data points between known data points. For example, linear interpolation, polynomial interpolation, or spline interpolation can be used. When the normalized conflict data and sentiment data fall within a certain region between pre-defined rules, the system no longer simply selects the closest pre-defined rule. Instead, based on the relative positions of these data points and adjacent pre-defined rules, it uses interpolation to generate a new rule—the intermediate resolution rule—that lies between these adjacent rules. This intermediate resolution rule integrates the characteristics of adjacent rules, more accurately reflecting the current complex conflict situation and the user's emotional state.

[0151] This application's solution normalizes conflict intensity and nature data, as well as user emotional state data, and uses interpolation to generate intermediate resolution rules between adjacent rules in a pre-defined conflict resolution strategy library. This effectively solves the rigidity and inaccuracy problems that may exist in rule selection in traditional methods. Normalization ensures fair participation of different types of data, while interpolation allows conflict processing to move beyond discrete pre-defined rules and generate a highly customized and smoothly transitioning resolution rule based on real-time, continuously changing conflict situations and user emotions. Therefore, the system can provide more refined, flexible conflict resolution solutions that are highly matched to the user's current state, avoiding visual or experiential inconsistencies that may result from abrupt rule switching.

[0152] In some preferred embodiments, it is assumed that in a virtual scene, a user is observing an ancient mural containing intricate textual descriptions and exquisite geological textures. At this point, the system detects a conflict between the semantic contexts of "text sharpening" and "geological texture enhancement" within the same pixel area. Through the steps described above, the system quantifies the intensity and nature of the conflict and infers that the user's current emotional state is "slightly curious but slightly tired." A pre-defined conflict resolution strategy library may contain two adjacent rules: Rule A (high-intensity text sharpening, low-intensity geological texture integration) and Rule B (medium-intensity text sharpening, medium-intensity geological texture integration).

[0153] First, the system collects data on the intensity and nature of the conflict (e.g., text sharpening demand intensity is 0.8, geological texture enhancement demand intensity is 0.6), and also collects data on the user's current emotional state (e.g., curiosity level is 0.7, fatigue level is 0.4). Next, this data is normalized to obtain normalized conflict data and normalized emotional data. For example, the normalized conflict intensity might be 0.75, and the normalized emotional state might be 0.55 (between the typical emotional states corresponding to rule A and rule B).

[0154] Subsequently, the system uses this normalized data as input and generates an intermediate resolution rule between rule A and rule B through interpolation methods (such as linear interpolation). This intermediate resolution rule might instruct: for text sharpening, use an intensity slightly lower than rule A but slightly higher than rule B; for geological texture enhancement, use a blending method higher than rule A but slightly lower than rule B. For example, the intermediate resolution rule might specifically be: increase text sharpening rendering parameters by 70%, decrease geological texture-related rendering parameters by 30%, and use a soft edge blending algorithm. Finally, based on this generated intermediate resolution rule, the rendering parameters on which the initial visual attributes of pixels depend are collaboratively adjusted to obtain the final enhanced image. This approach achieves a more balanced visual fusion effect for text sharpening and geological texture enhancement, better matching the user's current mood and scene needs, avoiding abrupt trade-offs and improving the subtlety of the user experience.

[0155] refer to Figure 2 This application further proposes a digital technology-based image enhancement system, applied to the aforementioned digital technology-based image enhancement method. The system includes:

[0156] The recognition module identifies scene objects in the virtual scene, obtains semantic information associated with the scene objects, and activates one or more semantic contexts of the scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with the scene objects is non-visual attribute information corresponding to the scene objects and used to characterize the semantic meaning of the scene objects. The non-visual attribute information includes at least one of the following: functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, a texture layer, a material layer, an environment layer, a semantic attribute layer, and an interaction layer.

[0157] The module determines an importance value for each activated semantic context based on user attention, user interaction behavior, and the priority of the virtual navigation narrative.

[0158] The generation module generates an independent enhanced rendering layer for each activated semantic context. The independent enhanced rendering layer is used to present the visual representation of the semantic context. The independent enhanced rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters.

[0159] The fusion module receives independent enhanced rendering layers and importance values, and fuses the independent enhanced rendering layers according to the semantic region where the pixel is located and the importance value to obtain the final enhanced image;

[0160] The output module will ultimately render and output the enhanced image in real time.

[0161] Specifically, the recognition module is configured to perform the recognition steps in the method described above. This module is responsible for analyzing scene objects in the virtual scene to obtain semantic information associated with these objects. Furthermore, the recognition module can activate one or more semantic contexts of scene objects based on user interaction behaviors, such as user clicks, gazes, or voice commands, or based on preset virtual navigation narrative cues. For example, when a user clicks on a historical building, the recognition module can activate semantic contexts related to the building's historical background, architectural style, or cultural significance.

[0162] The determination module is configured to perform the determination steps in the above method. Its function is to dynamically calculate and assign an importance value to each semantic context activated by the identified module, taking into account the user's attention level, user interaction behavior, and the narrative priority of the current virtual tour. This importance value reflects the level of importance of a specific semantic context in the current situation; for example, areas that the user focuses on for a long time or content highlighted in the tour will have their importance value increased accordingly.

[0163] In practical applications, the generation module is configured to perform the generation steps described above. This module independently creates an enhanced rendering layer for each active semantic context with an importance value. Each independent enhanced rendering layer is specifically designed to present the visual representation of its corresponding semantic context, for example, through highlighting, texture enhancement, and information annotation, making the semantic context more visually prominent or richer in information.

[0164] Furthermore, the fusion module is configured to perform the fusion step in the above method. This module receives the importance values ​​of each independent enhanced rendering layer generated by the generation module and its corresponding semantic context. The fusion module intelligently fuses these independent enhanced rendering layers based on the semantic region of each pixel and the corresponding importance value of that region, thereby generating a unified final enhanced image. The fusion process weights and coordinates the visual representation of different semantic regions according to their importance values.

[0165] Finally, the output module is configured to perform the output steps in the above method. This module is responsible for rendering the final enhanced image generated by the fusion module in real time and outputting it to the user's display device, ensuring that the user can see the enhanced virtual scene instantly.

[0166] This system provides an efficient and scalable architecture to support the actual deployment and operation of the aforementioned digital technology-based image enhancement methods, thereby overcoming the abstraction problems that may exist in pure method descriptions in practical applications.

[0167] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A digital technology-based image enhancement method, characterized in that, The method includes the following steps: Identify scene objects in a virtual scene, obtain semantic information associated with scene objects, and activate one or more semantic contexts of scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with scene objects is non-visual attribute information corresponding to scene objects and used to characterize the semantic meaning of scene objects. Non-visual attribute information includes at least one of functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, texture layer, material layer, environment layer, semantic attribute layer, and interaction layer. For each activated semantic context, an importance value is dynamically determined based on user attention, user interaction behavior, and the priority of the virtual tour narrative; For each activated semantic context, an independent enhancement rendering layer is generated. The independent enhancement rendering layer is used to present the visual representation of the semantic context. The independent enhancement rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters. The system receives independent enhancement rendering layers and their importance values. Based on the semantic region where the pixel is located and its importance value, the system fuses the independent enhancement rendering layers to obtain the final enhanced image. The final enhanced image will be rendered and output in real time. The steps to obtain the final enhanced image include: The scene object surface is divided into semantic regions, and semantic attributes are assigned to pixels according to the divided semantic regions; Store semantic attributes as multichannel texture data; In the fragment shader of the graphics processor, semantic attributes are read from multi-channel texture data, and preliminary visual attributes of pixels are calculated and generated in real time based on semantic attributes and importance values. When multiple highly important semantics intertwine on the same pixel, the rendering parameters on which the initial visual attributes of the pixel depend are adjusted collaboratively according to the preset rendering rule set. In particular, when text sharpening and geological texture enhancement conflict on the same pixel, the semantics with higher importance are processed first. By increasing the rendering parameters related to text sharpening and decreasing the rendering parameters related to geological texture, the geological texture enhancement effect is integrated in a soft way. The final visual attributes of pixels are regenerated based on the collaboratively adjusted rendering parameters, and the independent enhanced rendering layers are fused according to the final visual attributes of the pixels to obtain the final enhanced image. The steps for dynamically determining an importance value include: Collect user eye movement data and smooth the user eye movement data to obtain smoothed eye movement data; Identify areas on scene objects where visual details are blurred due to physical factors, and pre-associate these blurred areas with potentially conflicting semantics to obtain association results; Based on the correlation results, when the user's gaze falls on an area with blurred visual details, micro-eye movement pattern features are extracted from smooth eye movement data, and the eye movement behaviors corresponding to the micro-eye movement pattern features are classified into different sub-modes. Map submodals to specific user intent semantics and dynamically adjust the priority of the mapping based on the narrative theme of the current virtual tour; Assign importance values ​​to the semantic context activated in regions with blurred visual details, based on the semantic meaning of the user intent and the priority of the adjusted mapping. The steps for dynamically determining an importance value include: Identify conflicts between user attention, user interaction behavior, and the narrative priority of virtual guided tours; Based on the pre-set conflict resolution rules, the conflict is coordinated to arrive at a coordination result; The importance level is dynamically determined based on the coordination results.

2. The digital technology-based image enhancement method as described in claim 1, characterized in that, The steps to activate one or more semantic contexts of a scene object include: The scene object surface is divided into semantic regions, and multi-dimensional semantic labels are attached to the divided semantic regions; Collect user multimodal input data and obtain the virtual tour narrative context; Based on user multimodal input data and the virtual guided narrative context, infer the user's exploration intent and the focus of the guided narrative; By matching user exploration intent, navigation narrative focus, and multi-dimensional semantic identifiers of scene objects, candidate semantic contexts are obtained. Based on the granularity of the user's exploration intent and the focus of the navigation narrative, the granularity of the candidate semantic context is adaptively adjusted to obtain the granularized candidate semantic context. Based on the hierarchical relationship and relevance of semantics, target semantic contexts related to the current user's exploration intent and the focus of the navigation narrative are selected from the candidate semantic contexts after granular adjustment; Based on the consistency between the target semantic context and the user's multimodal input data and the virtual tour narrative context, the activation confidence of the target semantic context is evaluated. When multiple target semantic contexts have high activation confidence and semantic conflicts exist, conflict resolution rules are used to process the target semantic contexts to obtain the final candidate semantic contexts after conflict resolution. Based on the final candidate semantic context after conflict resolution, activate one or more semantic contexts of the scene object.

3. The digital technology-based image enhancement method as described in claim 2, characterized in that, The steps of dividing the surface of scene objects into semantic regions and attaching multi-dimensional semantic labels to the divided semantic regions include: Acquire multispectral data of the surface of scene objects; By utilizing the differences in spectral features in multispectral data, the boundaries of overlapping or interwoven semantic regions on the surface of scene objects can be identified; Based on the boundaries of the overlapping or intertwined semantic regions, extract the depth information of the overlapping or intertwined semantic regions; Based on spectral feature differences and depth information, overlapping or intertwined semantic regions are separated, and independent geometric proxies are generated for each separated semantic region. Each independent geometric agent is attached with a multi-dimensional semantic identifier, which includes semantic type, level, spatial location, and semantic relevance determined based on spatial adjacency. Check for conflicts in the additional multi-dimensional semantic identifiers; When a conflict is detected, the multi-dimensional semantic identifier of the conflict is adjusted according to the preset conflict resolution rules to ensure the independence and non-conflictability of the multi-dimensional semantic identifier.

4. The digital technology-based image enhancement method as described in claim 2, characterized in that, The steps for evaluating activation confidence in the target semantic context include: Based on user multimodal input data and virtual tour narrative context, feature matching is performed on the target semantic context to obtain the initial activation confidence. When the initial activation confidence of multiple target semantic contexts is at a medium level and there is semantic ambiguity, analyze the physical interleaving characteristics of the corresponding regions of the target semantic contexts; Based on the physical interweaving characteristics, and combined with the user's historical behavior patterns and preferences for virtual tour narrative themes, the initial activation confidence is corrected, and the corrected activation confidence is output.

5. A digital technology-based image enhancement method as described in claim 2, characterized in that, The steps for handling conflicts in the target semantic context according to preset conflict resolution rules include: Collect user physiological data, including heart rate variability and skin conductance response; The intensity and nature of the conflict are quantified based on the semantic type differences, activation confidence differences, visual complexity of the conflict area, and the narrative focus of the current virtual tour, among the target semantic contexts with semantic conflicts. Infer the user's current emotional state based on the user's physiological data; Based on the intensity and nature of the conflict and the user's current emotional state, conflict resolution rules are dynamically selected from a preset conflict resolution strategy library; Conflicts are resolved according to conflict resolution rules.

6. The digital technology-based image enhancement method as described in claim 5, characterized in that, The steps to resolve a conflict include: Collect data on the intensity and nature of the conflict; Collect data on the user's current emotional state; The intensity and nature of the conflict data are normalized to obtain normalized conflict data. Normalize the user's current emotional state data to obtain normalized emotional data; Normalized conflict data and normalized sentiment data are used as inputs. Through interpolation, a smooth transition is achieved between adjacent rules in a pre-defined conflict resolution strategy library to generate intermediate resolution rules. Conflicts are resolved according to intermediate resolution rules.

7. A digital technology-based image enhancement system, used to execute a digital technology-based image enhancement method as described in any one of claims 1-6, characterized in that, The system includes: The recognition module identifies scene objects in the virtual scene, obtains semantic information associated with the scene objects, and activates one or more semantic contexts of the scene objects based on user interaction behavior or virtual tour narrative clues. The semantic information associated with the scene objects is non-visual attribute information corresponding to the scene objects and used to characterize the semantic meaning of the scene objects. The non-visual attribute information includes at least one of the following: functional attributes, semantic category attributes, historical background attributes, or cultural value attributes. The virtual scene includes a geometric layer, a texture layer, a material layer, an environment layer, a semantic attribute layer, and an interaction layer. The module determines an importance value for each activated semantic context based on user attention, user interaction behavior, and the priority of the virtual navigation narrative. The generation module generates an independent enhanced rendering layer for each activated semantic context. The independent enhanced rendering layer is used to present the visual representation of the semantic context. The independent enhanced rendering layer includes semantic region mask sub-data and enhancement parameter sub-data. The enhancement parameter sub-data includes at least one of the following: text sharpening parameters, texture enhancement parameters, color correction parameters, detail reconstruction parameters, or highlight emphasis parameters. The fusion module receives independent enhanced rendering layers and importance values, and fuses the independent enhanced rendering layers according to the semantic region where the pixel is located and the importance value to obtain the final enhanced image; The output module will ultimately render and output the enhanced image in real time.

Images