A cognitive reconstruction method based on wisdom character image

Abstract

The application relates to the technical field of computer vision, and particularly discloses a cognitive reconstruction method based on intelligent character images, which comprises the following steps: S1, collecting real-time drawing process data of a user and a completed hand-drawn intelligent character image; S2, extracting image feature parameters according to the drawing process data and the hand-drawn intelligent character image, and constructing a virtual intelligent character image based on the image feature parameters; S3, constructing a virtual interactive scene based on the virtual intelligent character image; S4, obtaining an interactive voice input by a user, generating a guide voice based on the current virtual interactive scene, and obtaining a feedback voice input by the user; S5, performing semantic analysis on the feedback voice, determining the matching degree between the currently displayed virtual intelligent character image and the user's expectation, improving the recognition degree of the virtual intelligent character image in the virtual interactive scene in subsequent voice interaction if the matching degree meets preset conditions, and modifying the virtual intelligent character image according to the feedback voice if the matching degree does not meet the preset conditions. The technical scheme of the application can accurately construct a guide image that matches the subconscious of a visitor without breaking the immersion and awe of the visitor.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a cognitive reconstruction method based on the imagery of intelligent human figures. Background Technology

[0002] In the field of psychological counseling and therapy, cognitive behavioral therapy (CBT) is currently recognized as one of the most effective approaches. With the Sinicization of psychological counseling and therapy, as described in the applicant's monograph, *Imagery Therapy: The Sinicization of Cognitive Behavioral Therapy*, the "imagery reconstruction" technique is widely used in interventions for post-traumatic stress, anxiety, and depression. Within the context of Chinese culture, therapists can guide clients to construct an image of a wise person (such as Wang Yangming or Mencius) embodying benevolence, compassion, and wisdom within themselves. This image serves as a projection vehicle for the client's psychological energy and a guide for inner dialogue. In traditional psychological counseling and therapy, therapists typically ask clients to visualize this "wise person" through verbal description or hand-drawn sketches, engaging in a virtual dialogue with this image solely through their imagination, thereby reconstructing negative cognitions.

[0003] However, in actual clinical interventions, it has been found that traditional interactive methods relying on pure imagination and simple hand-drawing have significant limitations. Because most clients lack professional drawing skills and highly focused spatial imagination, their hand-drawn images are often too rudimentary and lack vitality. This makes it difficult for clients to maintain immersion in the image during subsequent psychological interventions. One-way mental reconstruction weakens the interactive experience and significantly reduces the therapeutic effect.

[0004] To address the aforementioned technical issue of weak interaction, researchers attempted to incorporate mature AI image or video generation models. The initial idea was to convert the visitor's description of the intelligent figure into text prompts, then have generative AI directly render a high-fidelity, dynamic virtual intelligent figure, and finally overlay a large-scale voice model for real-time dialogue. However, this technical attempt revealed a critical flaw in psychological intervention scenarios: the output of generative AI is probabilistic and uncontrollable. Images generated solely from text prompts are prone to deviation and inconsistency from the visitor's subconscious expectation of the intelligent figure.

[0005] In typical entertainment or auxiliary design scenarios, if generated images don't meet expectations, this can be resolved by repeatedly modifying and adding prompts, drawing cards, and redrawing. However, in the specific context of psychological cognitive reconstruction, the core function of the image of a wise figure lies in its transcendent solemnity, wisdom, and the unconditional trust the client develops towards it. If the system-generated image doesn't match the client's expectations and is abruptly presented to them, it will instantly shatter the client's psychological identification. Furthermore, the repeated trial-and-error process of having the client or therapist constantly adjust prompts, regenerate, and evaluate the resemblance in order to correct the image completely strips away the sacredness and independent personality of the wise figure. The client will immediately realize that this is merely a digital puppet controlled by code, instantly dispelling their awe and trust in the image of the wise figure, causing their psychological defenses to close again, and rendering the entire cognitive reconstruction therapeutic process a failure.

[0006] In conclusion, there is an urgent need for a new psychological cognitive reconstruction technique that can accurately construct guiding images that resonate with the subconscious of the client without disrupting their sense of immersion and awe. Summary of the Invention

[0007] This invention provides a cognitive reconstruction method based on the imagery of wise figures, which can accurately construct guiding images that resonate with the subconscious of visitors without disrupting their sense of immersion and awe.

[0008] To solve the above-mentioned technical problems, this application provides the following technical solution: A cognitive reconstruction method based on the imagery of intelligent figures includes: S1. Real-time collection of user drawing process data and completed hand-drawn smart character images; S2. Extract imagery feature parameters from the painting process data and hand-drawn intelligent character images, and construct virtual intelligent character images based on the imagery feature parameters; S3. Construct a virtual interactive scene based on the virtual intelligent character image, and place the virtual intelligent character image in the virtual interactive scene; wherein, the virtual intelligent character image is displayed in the virtual interactive scene in an initial outline state; S4. Obtain the interactive voice input from the user, generate guiding voice based on the current virtual interactive scenario, and output it through a virtual intelligent character image; The system obtains user input feedback voice, inputs the feedback voice into a preset large language model to generate response content, generates response voice based on the response content, and outputs it through a virtual intelligent character. S5. Perform semantic analysis on the feedback voice to determine the matching degree between the currently displayed virtual intelligent character and the user's expectations; if the matching degree meets the preset conditions, improve the recognition of the virtual intelligent character in the virtual interaction scene in subsequent voice interactions; if the matching degree does not meet the preset conditions, modify the virtual intelligent character based on the feedback voice.

[0009] Furthermore, in step S1, data on the communication behavior between the user and the psychologist is also collected during the drawing process.

[0010] Furthermore, step S2 specifically includes: The line density and geometric closure features of hand-drawn intelligent character images are extracted, and a static refinement index representing the complexity of the image is calculated. Extract the frequency of pen pauses, the number of stroke redraws, and the variance of pen speed from the drawing process data to calculate the first dynamic confidence index that characterizes the continuity of drawing. By analyzing the number of interactions between users and psychologists in the communication behavior data, a second dynamic confidence index representing the confidence level in drawing was obtained. The image feature parameters are constructed by combining the static refinement index, the first dynamic confidence index, and the second dynamic confidence index. The image feature parameters are input into a preset inverse realism mapping function to calculate the target rendering realism parameter. The inverse realism mapping function is configured such that: the higher the static refinement index and the higher the first and second dynamic confidence indices, the lower the output target rendering realism parameter; the lower the static refinement index and the lower the first and second dynamic confidence indices, the higher the output target rendering realism parameter. Based on the target rendering realism parameters, a virtual intelligent character image is constructed on the basis of a hand-drawn intelligent character image. Specifically, when the target rendering realism parameters are lower than a first preset threshold, an abstract and conceptual virtual intelligent character image is constructed; when the target rendering realism parameters are higher than a second preset threshold, a high-fidelity and detailed virtual intelligent character image is constructed.

[0011] Furthermore, in step S3, constructing a virtual interactive scene based on a virtual intelligent character specifically includes: The initial global atmosphere parameters of the virtual interactive scene are set according to the target rendering realism parameters; the initial global atmosphere parameters include at least volumetric fog density and depth blur; wherein, volumetric fog density is inversely proportional to the target rendering realism parameters, and depth blur is inversely proportional to the target rendering realism parameters; The character style is extracted from hand-drawn intelligent character images, and the corresponding set of scene elements is matched in a preset historical scene library; the set of scene elements includes natural landscape entities and architectural structure entities. Based on the virtual intelligent character image, the matched set of scene elements is stylized and rendered to construct a virtual interactive scene that is consistent with the rendering style of the virtual intelligent character image.

[0012] Furthermore, in step S3, placing the virtual intelligent character image within a virtual interactive scene specifically includes: The virtual intelligent character is deployed at a distant coordinate point in the virtual interactive scene, far from the virtual viewpoint; and the initial orientation vector of the virtual intelligent character is adjusted so that it faces away from the virtual viewpoint. Based on the entity attributes in the scene element set, semi-transparent occlusion entities are deployed on the visual path between the virtual viewpoint and the distant coordinate point; the semi-transparent occlusion entities are configured as partition models with non-uniform transparency fields or hollow texture mapping to perform local occlusion processing on the virtual intelligent character image. By using dynamic light sources in a virtual interactive scene to project light and shadow onto a semi-transparent obscuring entity, a virtual intelligent character image is projected onto the semi-transparent obscuring entity in real time.

[0013] Furthermore, in step S4, generating and outputting guiding voice based on the current virtual interaction scenario specifically includes: Extract environmental style feature words and rendering realism feature words from the current virtual interaction scene; By inputting environmental style features, rendering realism features, and user interaction speech into a large language model, guiding speech containing environmental metaphors is generated.

[0014] Furthermore, in step S5, semantic analysis is performed on the feedback voice to determine the matching degree between the currently displayed virtual intelligent character and the user's expectations, specifically including: Extract the user's feedback voice in response to the guided voice input, perform natural language processing on the feedback voice, and input it into a large language model to obtain the user's satisfaction evaluation value for the current virtual interaction scenario; Based on the shared rendering relationship between virtual interactive scenes and virtual intelligent character images, the satisfaction evaluation value of virtual interactive scenes is mapped to the matching degree between the current virtual intelligent character image and the user's expectations.

[0015] Furthermore, in step S5, if the matching degree does not meet the preset conditions, the virtual intelligent character image is modified according to the feedback voice, specifically including: Semantic elements representing users' evaluation of virtual interaction scenarios are extracted from feedback voice, and these semantic elements are correlated with environmental style feature words and rendering realism feature words in the current virtual interaction scenario to determine the rendering dimension labels to be corrected. The rendering dimension labels include at least the realism-abstract dimension, the cool-warm color dimension, and the hard-soft texture dimension. Calculate the offset vector of semantic elements under each rendering dimension label, and use the offset vector to adjust the virtual intelligent character image while keeping the current outline topology unchanged.

[0016] Furthermore, in step S5, improving the display and recognition of the virtual intelligent character in the virtual interaction scene during subsequent voice interaction specifically includes: The system dynamically adjusts the spatial distance between the virtual viewpoint and the virtual intelligent character, gradually reducing the Euclidean distance between the virtual viewpoint and the distant coordinate point to make the user visually closer to the virtual intelligent character. Based on changes in the position of the virtual viewpoint, the system gradually adjusts the transparency parameter of the semi-transparent occluding entity or removes the semi-transparent occluding entity to remove partial occlusion of the virtual intelligent character. When the distance between the virtual viewpoint and the virtual intelligent character is less than a preset interaction distance threshold, the system modifies the initial orientation vector of the virtual intelligent character so that the virtual intelligent character faces the virtual viewpoint and outputs a response voice with its face towards the user.

[0017] This solution extracts static precision and dynamic certainty indices and uses an inverse mapping function for reverse output: For users with high static precision and dynamic certainty, an abstract, conceptualized intelligent character is matched, fully respecting the subconscious of users with high projection ability, leaving blank spaces for them to fill in the details that meet their inner expectations; for users who draw poorly or hesitate repeatedly, a high-fidelity, realistic, and detailed intelligent character is matched, providing a clear visual image for users who lack concrete drawing ability. By creating intelligent character images tailored to each individual, it is possible to ensure that the generated intelligent character image meets the user's psychological expectations to the greatest extent possible. Attached Figure Description

[0018] Figure 1 This is a flowchart of an embodiment of a cognitive reconstruction method based on the imagery of intelligent figures. Detailed Implementation

[0019] The following detailed description illustrates the specific implementation method: Example 1 like Figure 1 As shown in this embodiment, a cognitive reconstruction method based on the imagery of a wise person includes the following steps: S1. Real-time collection of user drawing process data and completed hand-drawn intelligent character image; also, during the drawing process, collection of communication behavior data between the user and the psychologist. Communication behavior data includes the number of conversations or eye contact between the user and the psychologist; in this embodiment, the user's facial orientation is analyzed, and when the user's face is turned towards the psychologist during the drawing process, it is considered eye contact.

[0020] S2. Extract imagery feature parameters from the painting process data and hand-drawn intelligent character images, and construct a virtual intelligent character image based on the imagery feature parameters; specifically including: This paper extracts line density and geometric closure features from hand-drawn intelligent character images to calculate a static precision index representing the complexity of the image. In this embodiment, the acquired hand-drawn intelligent character images are preprocessed by grayscale and binarization to separate line areas from background areas. The line density feature is calculated, which is the proportion of the total number of pixels belonging to line areas to the total number of pixels in the image. This proportion is used as the line density feature representing the complexity of the lines in the image. All closed contours in the image are extracted using an edge detection algorithm, and the number, average perimeter, and area of ​​closed contours are calculated. The number of closed contours is compared with a preset complexity mapping table, or the ratio of the total area of ​​all closed contours to the total area of ​​the image is calculated as the geometric closure feature. After normalizing the line density and geometric closure features, a weighted sum is obtained to obtain a static precision index that comprehensively reflects the degree of line filling and the complexity of graphic closure in the image. The higher the index value, the denser the lines and the more complex the geometric closure in the hand-drawn intelligent character image, and the clearer the image of the intelligent character for the user. Conversely, the lower the index value, the simpler and more open the image, and the more ambiguous the image of the intelligent character for the user.

[0021] The frequency of pen stroke pauses, the number of stroke redraws, and the variance of pen stroke speed are extracted from the drawing process data to calculate a first dynamic confidence index characterizing the continuity of drawing. In this embodiment, the user draws using a tablet computer with a stylus, and the tablet computer records the user's handwriting data in real time. The frequency of pen stroke pauses is statistically analyzed by monitoring trajectory segments at sampling points whose speed is lower than a preset speed threshold and whose duration exceeds a preset time threshold. The proportion of the occurrence of such segments to the total duration of the entire drawing process is used as the pause frequency. Erasing the original lines and redrawing them is determined as a redrawing action, and the total number of redraws is accumulated. The variance of pen stroke speed is calculated based on the statistical variance of the instantaneous speed values ​​at each sampling point throughout the drawing process to reflect the degree of fluctuation in pen stroke speed. After normalizing the frequency of pen stroke pauses, the number of stroke redraws, and the variance of pen stroke speed, a weighted sum is obtained to obtain a first dynamic confidence index that comprehensively characterizes the continuity and hesitation of the user's operation during the drawing process. The higher the value of this indicator, the stronger the continuity of the drawing process and the less hesitation and correction there is; conversely, the lower the value, the more pauses and repeated modifications there are in the drawing process.

[0022] By analyzing the number of interactions between users and psychologists in the communication behavior data, a second dynamic confidence index representing the confidence level in drawing was obtained; the higher the index value, the more hesitation, uncertainty or non-autonomous behavior there is in the drawing process.

[0023] The image feature parameters are constructed by combining the static precision index, the first dynamic confidence index, and the second dynamic confidence index; the image feature parameters are input into the preset inverse realism mapping function to calculate the target rendering realism parameters. The reverse realism mapping function is configured such that: the higher the static refinement index and the higher the first dynamic confidence index and the second dynamic confidence index, the lower the output target rendering realism parameter; the lower the static refinement index and the lower the first dynamic confidence index and the second dynamic confidence index, the higher the output target rendering realism parameter. Specifically, the static precision index, the first dynamic confidence index, and the second dynamic confidence index are normalized to ensure they all fall within the same numerical range (0 to 1 in this embodiment) to eliminate the influence of dimensions. A reverse realism mapping function is constructed, configured to output a target rendering realism parameter that is negatively correlated with the static precision index, and also negatively correlated with both the first and second dynamic confidence indices. This mapping function uses a weighted summation followed by inversion, where the target rendering realism parameter is set to a preset maximum value minus the weighted sum of the static precision index, the first dynamic confidence index, and the second dynamic confidence index; or it uses a reciprocal form, where the target rendering realism parameter equals a preset constant divided by the sum of the weighted sum of the three indices plus the smoothing factor. This embodiment uses the weighted summation followed by inversion form. The weight coefficients in the reverse realism mapping function can be preset empirically. When a user's hand-drawn image has dense lines and the drawing process is coherent and confident, the output target rendering realism parameter is low, and the virtual intelligent character image constructed subsequently tends to be abstract; conversely, when the hand-drawn image has sparse lines and the drawing process is hesitant and disjointed, the output target rendering realism parameter is high, and the virtual intelligent character image constructed tends to be highly realistic.

[0024] For example, in this embodiment, the weight coefficients of the preset static refinement index, the first dynamic confidence index, and the second dynamic confidence index are w1, w2, and w3, respectively. The preset values ​​for w1, w2, and w3 are 0.4, 0.3, and 0.3, respectively. The preset maximum value M is set to 1. The formula for calculating the target rendering realism parameter R is:

[0025] Where S is the normalized static precision index, D1 is the normalized first dynamic confidence index, and D2 is the normalized second dynamic confidence index.

[0026] Based on the target rendering realism parameters, the corresponding material texture resolution, facial mesh topology density, and lighting rendering engine type are matched from a preset image resource library to construct a virtual intelligent character image based on a hand-drawn intelligent character image. In this embodiment, when the target rendering realism parameters are lower than a first preset threshold, a low-poly mesh and non-realistic rendering technology are used to construct an abstract conceptual virtual intelligent character image; when the target rendering realism parameters are higher than a second preset threshold, a high-precision topological mesh and physically based rendering technology are used to construct a high-fidelity and detailed virtual intelligent character image. In other embodiments, existing image generation models can also be used to construct either an abstract conceptual virtual intelligent character image or a high-fidelity and detailed virtual intelligent character image. In this embodiment, when the target rendering realism parameters are between the first and second preset thresholds, a medium-precision topological mesh and cartoon rendering technology are used to construct a virtual intelligent character image with a semi-realistic style. In this state, the virtual intelligent character possesses clearly identifiable facial features and body contours to meet basic visual guidance while retaining appropriate artistic white space and non-realistic color blocks to balance the visitor's intermediate state between the concrete and the abstract. The first and second preset thresholds are empirically set by technicians based on specific psychological interaction scenarios and the perceptual sensitivity of the target user group. The first preset threshold ranges from 0.2 to 0.4, and the second preset threshold ranges from 0.6 to 0.8. In this embodiment, the first preset threshold is set to 0.3, and the second preset threshold is set to 0.7. Taking a visitor A who paints coherently, confidently, and with strong painting skills as an example, the collected and calculated normalized static precision index is 0.8, the first dynamic confidence index is 0.9, and the second dynamic confidence index is 0.85. Substituting the above data into the inverse realism mapping function, which is a weighted sum and then inverted, we calculate the target rendering realism parameter R as 1.0 minus the weighted sum of 0.845, resulting in 0.155. Because this parameter is low, the system will subsequently construct an abstract, conceptualized virtual intelligent character for it, leaving ample white space. Conversely, taking a visitor B who hesitates in their drawing and whose lines are sparse as an example, the collected and calculated normalized static refinement index is 0.2, the first dynamic confidence index is 0.3, and the second dynamic confidence index is 0.2. Substituting these values ​​into the formula, we calculate a weighted sum of 0.23, resulting in a target rendering realism parameter R of 0.77. This parameter is high, allowing us to construct a high-fidelity virtual intelligent character for them.

[0027] S3. Construct a virtual interactive scene based on the virtual intelligent character image, and place the virtual intelligent character image in the virtual interactive scene; wherein, the virtual intelligent character image is displayed in the virtual interactive scene in an initial outline state.

[0028] Specifically, the construction of virtual interactive scenarios based on virtual intelligent character images includes: The character style is extracted from the hand-drawn intelligent character image, and the corresponding scene element set is matched in the preset historical scene library; the scene element set includes natural landscape entities, architectural structure entities and meteorological particle systems.

[0029] Based on the virtual intelligent character image, the matched set of scene elements is stylized and rendered to construct a virtual interactive scene that is consistent with the rendering style of the virtual intelligent character image.

[0030] Placing virtual intelligent avatars in virtual interactive scenarios specifically includes: The virtual intelligent character is deployed at a distant coordinate point in the virtual interactive scene, far from the virtual viewpoint; and the initial orientation vector of the virtual intelligent character is adjusted so that it faces away from the virtual viewpoint, in order to build a sense of intelligence based on its back view at the visual level. Based on the entity attributes in the scene element set, semi-transparent occlusion entities are deployed on the visual path between the virtual viewpoint and the distant coordinate point; the semi-transparent occlusion entities are configured as partition models with non-uniform transparency fields or hollow texture mapping to perform local occlusion processing on the virtual intelligent character image. By using dynamic light sources in a virtual interactive scene to project light and shadow onto a semi-transparent obscuring entity, a virtual intelligent character image is projected onto the semi-transparent obscuring entity in real time.

[0031] S4. Obtain the user's input interactive voice, generate guidance content based on the current virtual interaction scenario, convert it into guidance voice, and output it through a virtual intelligent character image; specifically, this includes: extracting environmental style feature words and rendering realism feature words from the current virtual interaction scenario; inputting the environmental style tags, rendering realism tags, and the user's interactive voice into a large language model to generate guidance voice containing environmental metaphors.

[0032] The system obtains user input feedback voice, inputs the feedback voice into a preset large language model to generate response content, generates response voice based on the response content, and outputs the response voice through a virtual intelligent character; S5. Perform semantic analysis on the feedback voice to determine the matching degree between the currently displayed virtual intelligent character and the user's expectations. If the matching degree meets the preset conditions, improve the recognizability of the virtual intelligent character in the virtual interaction scene in subsequent voice interactions. If the matching degree does not meet the preset conditions, modify the virtual intelligent character based on the feedback voice. In this embodiment, only one modification is performed. After modification, the recognizability of the virtual intelligent character in the virtual interaction scene is improved in subsequent voice interactions. In other embodiments, multiple rounds of dialogue modification can also be set.

[0033] This includes performing semantic analysis on the feedback voice to determine the degree of match between the currently displayed virtual intelligent character and the user's expectations, specifically including: The user's feedback voice in response to the guided voice input is extracted, and natural language processing is performed on the feedback voice. The feedback voice is then input into a large language model to obtain the user's satisfaction assessment value for the current virtual interaction scenario. In this embodiment, if the feedback voice does not contain a satisfaction assessment value for the current virtual interaction scenario, it is considered as satisfactory.

[0034] Based on the shared rendering relationship between virtual interactive scenes and virtual intelligent character images, the satisfaction evaluation value of virtual interactive scenes is directly mapped to the matching degree between the current virtual intelligent character image and the user's expectations.

[0035] If the matching degree does not meet the preset conditions, the virtual intelligent character image will be modified according to the feedback voice, specifically including: Semantic elements representing users' evaluation of virtual interaction scenarios are extracted from feedback voice, and these semantic elements are correlated with environmental style feature words and rendering realism feature words in the current virtual interaction scenario to determine the rendering dimension labels to be corrected. The rendering dimension labels include at least the realism-abstract dimension, the cool-warm color dimension, and the hard-soft texture dimension. The offset vectors of semantic elements under each rendering dimension label are calculated. These offset vectors are then used to adjust the virtual intelligent character's appearance while maintaining its current outline topology. This embodiment pre-defines a set of rendering dimension labels, including realistic-abstract dimensions, cool-warm color tones, and hard-soft texture dimensions. A corresponding semantic embedding vector baseline is pre-constructed for each dimension label. For example, in the realistic-abstract dimension, words like "realistic," "delicate," and "realistic" are mapped to the realistic end, while words like "simple," "imagery," and "abstract" are mapped to the abstract end. When user feedback voice is obtained and converted to text, this text is input into a pre-trained large language model. The model is guided by prompts to extract evaluative semantic elements related to each rendering dimension label and outputs a quantified tendency value for each dimension. This value can be in interval scaling form; in this embodiment, it ranges from 0 to 10, where 0 represents realistic and 10 represents abstract. The existing rendering parameter values ​​of the current virtual intelligent character in each dimension are used as the baseline vector. The difference between the quantized tendency value output by the large language model and the baseline vector is calculated to obtain the offset vector under each rendering dimension label. Each component of the offset vector represents the direction and magnitude of adjustment required in that dimension. Positive values ​​indicate an offset towards abstract, warm, or soft colors, while negative values ​​indicate an offset in the opposite direction.

[0036] To improve the display and recognition of virtual intelligent characters in subsequent voice interactions, specifically including: The system dynamically adjusts the spatial distance between the virtual viewpoint and the virtual intelligent character, gradually reducing the Euclidean distance between the virtual viewpoint and the distant coordinate point to make the user visually closer to the virtual intelligent character. Based on changes in the position of the virtual viewpoint, the system gradually adjusts the transparency parameter of the semi-transparent occluding entity or removes the semi-transparent occluding entity to remove partial occlusion of the virtual intelligent character. When the distance between the virtual viewpoint and the virtual intelligent character is less than a preset interaction distance threshold, the system modifies the initial orientation vector of the virtual intelligent character so that the virtual intelligent character faces the virtual viewpoint and outputs a response voice with its face towards the user.

[0037] In this embodiment, the virtual intelligent avatar turns around during the next voice interaction, and then gradually brings the user closer to the virtual intelligent avatar visually in subsequent voice interactions. Through spatial approximation from a virtual viewpoint, the gradual dissolution of semi-transparent occlusion, and the changing orientation of the virtual intelligent avatar, a non-verbal interaction metaphor is constructed between the user and the virtual intelligent avatar, moving from back to front and from far to near. This allows the user to gain recognition and acceptance of the virtual intelligent avatar at a psychological level, establishing trust in the virtual intelligent avatar.

[0038] During its development, this invention addressed the shortcomings of current image or video generation models in the field of psychological intervention. Specifically, images generated solely based on prompts are prone to discrepancies with the user's subconscious expectations, leading to a loss of trust in the guiding imagery and undermining subsequent reconstruction. Compared to existing technologies, this solution extracts static precision and dynamic certainty indices and uses an inverse mapping function for reverse output: for users who draw well and execute crisp strokes (i.e., high static precision and dynamic certainty), the system matches an abstract, conceptualized low-polygon mesh model. This fully respects the subconscious of users with high projective abilities, leaving blank spaces for them to fill in the details that match their inner expectations; for users who draw poorly and hesitate repeatedly (i.e., low indices), the system matches a high-fidelity, realistic, and detailed physically based rendering model, providing a clear visual image for users lacking concrete imagery. By creating intelligent character images tailored to each individual, this approach maximizes the chances of the generated intelligent character image matching the user's psychological expectations.

[0039] This solution does not immediately present the full image of the intelligent character to the user after generation. Instead, it first places the intelligent character in a misty, partially obscured area, revealing only its outline, and extracts visual feature words from the interactive scene. A large language model is used to integrate scene elements with the user's confusion words, generating guiding voice with environmental metaphors. By analyzing the user's unintentional evaluations of the scene description, the system subtly infers the user's affinity for the intelligent character image. This provides the system with a valuable opportunity for mid-process adjustments and parameter drift compensation, avoiding abrupt presentations when the image does not match, and ensuring that the intelligent character image meets the user's expectations when finally seen clearly. This solution also artificially increases the distance between the user and the intelligent character image, placing it in the distance, with its back to the user, and partially obscured by a screen, to establish a sense of compassion and wisdom. As the affinity is achieved in step S5, the user feels that through their own actions and positive feedback, they have prompted the intelligent character to turn towards them and approach the intelligent character. This dynamic interaction, from a silhouette to a close-up face-to-face encounter, transforms abstract psychological identification into concrete spatial changes, allowing users to deeply feel that they have been recognized by the virtual intelligent character, and greatly establishing users' deep trust in the virtual intelligent character's image.

[0040] In summary, this solution employs a logical interlocking mechanism among its various technical methods, from the reverse realism mapping based on dynamic confidence in steps S1 and S2, to the implicit satisfaction verification using environmental metaphors in steps S3 and S4, and finally to the spatial distance reduction and face-to-face interaction triggered in step S5. The target rendering realism determined in step S2 directly dictates the scene construction strategy in step S3; conversely, the specific environmental visual features constructed in step S3 serve as the basis for generating metaphorical prompts through the fusion of the large language model in step S4; and ultimately, the implicit semantic analysis results in step S5 not only provide a fault-tolerant calibration for the initial generation effect in step S2 but also form the basis for the transition from long-distance back-to-back interaction to close-range face-to-face interaction in step S5.

[0041] Example 2 This example illustrates specific scenarios for clinical interventions in depression and workplace anxiety.

[0042] Client's basic information: Client A recently experienced a major setback at work, falling into deep self-doubt and anxiety, and was assessed as having moderate depression by psychological scales. The therapist and client decided to use Confucius as the core figure of wisdom to assist in cognitive restructuring.

[0043] Step S1: The therapist guides client A to draw their image of Confucius on a pressure-sensitive digital drawing board. During the drawing process, the system collects real-time data on pen pressure, coordinates, and timestamps. At the same time, client A shows obvious hesitation, stopping several times and asking the therapist, "I can't draw his face well, the lines feel all messed up, is this okay?" The system simultaneously collects data on this interaction.

[0044] Step S2: The system analyzes the drawing data of visitor A: Due to his poor drawing skills, the extracted line density and geometric closure features are extremely irregular, and the calculated static precision index is extremely low; at the same time, because he frequently pauses his brushstrokes, uses undo and redraw extensively, and has communicated (asked for help) with the psychologist as many as 8 times, the calculated first dynamic confidence index, which represents the continuity of drawing, and the second dynamic confidence index, which represents self-confidence, are both at extremely low levels.

[0045] The system inputs the aforementioned low indicators into the inverse realism mapping function, determining that the visitor's current visualization ability is extremely weak and their inner state is filled with uncertainty. Based on the inverse compensation logic, the system outputs high-level target rendering realism parameters and automatically matches high-precision topological meshes and physically based rendering techniques from the resource library. Based on the visitor's rough sketch outline, it generates a realistic 3D virtual image of "Confucius" with extremely lifelike skin texture and flowing beard.

[0046] Step S3: The system extracts the confusion keywords "confusion, frustration, and unclear future" from the visitor A's communication data, and matches "ancient academy in autumn rain" as a set of scene elements in the historical scene library.

[0047] Based on high-fidelity rendering parameters, the system sets extremely high volumetric fog density and deep ambient light occlusion intensity. In the virtual 3D space, a realistic version of "Confucius" is deployed at a distant coordinate point far from the virtual viewpoint, with its initial orientation vector set to face away from visitor A (i.e., from behind). Simultaneously, a semi-transparent wooden screen model with openwork carvings (a semi-transparent occlusion entity) is deployed between the visitor's viewpoint and the wise figure. A flickering lamp in the scene projects the wise figure's silhouette onto the screen, presenting a hazy, compassionate, and wise impression.

[0048] Step S4: The system extracts the physical atmosphere tags of the current scene (autumn rain, slight chill, solitary lamp). The large language model combines the visitor's confused keywords to generate a guiding voice with environmental metaphors, which is then output in a calm tone by a wise figure behind the screen: "Listen to the sound of the autumn rain hitting the withered leaves in this courtyard. Is your self-doubt as chilling as this autumn rain, making you feel surrounded by enemies?" Upon hearing this, visitor A felt a strong sense of empathy and responded into the microphone: "Yes, I feel extremely cold. No matter how hard I try, it seems like I can't escape this rain. I feel completely lost." Step S5, Hidden Verification and Dynamic Visual Transition: The system performs semantic analysis on visitor A's feedback speech, extracting emotional feature vectors (frustration, agreement) and strong response semantic vectors to environmental metaphors (autumn rain / chilly weather). Calculations show that the semantic resonance between the two is extremely high, leading the system to determine that the visitor has subconsciously accepted the physical atmosphere of the current scene. This implies that the currently displayed virtual intelligent character matches the visitor's expectations according to preset conditions.

[0049] Triggering dynamic evolution: In the next voice interaction, the system activates the viewpoint displacement compensation algorithm, and the virtual viewpoint begins to automatically close the distance and bypass the semi-transparent wooden screen; at the same time, the system smoothly reduces the volumetric fog density, and the academy scene gradually becomes clear.

[0050] As the visitor "approaches," the virtual intelligent character triggers a skeletal animation, slowly turning from its back view to face the visitor A directly, establishing a direct eye contact axis. At this moment, the large language model generates a response voice, which is output by the intelligent character: "Though the rain is cold, it can only soak your clothes, but it cannot extinguish the light of benevolence in your heart. With such a bright heart, what more needs to be said?" After experiencing the complete interactive metaphor of "distant background view, environmental empathy, bypassing the obstruction, and the wise figure turning around to make eye contact," visitor A deeply felt the shock of being accepted and recognized by the image of the wise figure, instantly letting down their psychological defenses and entering a state of deep cognitive reconstruction.

[0051] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A cognitive reconstruction method based on the imagery of intelligent figures, characterized in that, include: S1. Real-time collection of user drawing process data and completed hand-drawn smart character images; S2. Extract imagery feature parameters from the painting process data and hand-drawn intelligent character images, and construct virtual intelligent character images based on the imagery feature parameters; S3. Construct a virtual interactive scene based on the virtual intelligent character image, and place the virtual intelligent character image in the virtual interactive scene; wherein, the virtual intelligent character image is displayed in the virtual interactive scene in an initial outline state; S4. Obtain the interactive voice input from the user, generate guiding voice based on the current virtual interactive scenario, and output it through a virtual intelligent character image; The system obtains user input feedback voice, inputs the feedback voice into a preset large language model to generate response content, generates response voice based on the response content, and outputs it through a virtual intelligent character. S5. Perform semantic analysis on the feedback voice to determine the matching degree between the currently displayed virtual intelligent character and the user's expectations; if the matching degree meets the preset conditions, improve the recognition of the virtual intelligent character in the virtual interaction scene in subsequent voice interactions; if the matching degree does not meet the preset conditions, modify the virtual intelligent character based on the feedback voice.

2. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 1, characterized in that: In step S1, data on the interaction between the user and the psychologist is also collected during the drawing process.

3. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 2, characterized in that: Step S2 specifically includes: The line density and geometric closure features of hand-drawn intelligent character images are extracted, and a static refinement index representing the complexity of the image is calculated. Extract the frequency of pen pauses, the number of stroke redraws, and the variance of pen speed from the drawing process data to calculate the first dynamic confidence index that characterizes the continuity of drawing. By analyzing the number of interactions between users and psychologists in the communication behavior data, a second dynamic confidence index representing the confidence level in drawing was obtained. The image feature parameters are constructed by combining the static refinement index, the first dynamic confidence index, and the second dynamic confidence index. The image feature parameters are input into a preset inverse realism mapping function to calculate the target rendering realism parameter. The inverse realism mapping function is configured such that: the higher the static refinement index and the higher the first and second dynamic confidence indices, the lower the output target rendering realism parameter; the lower the static refinement index and the lower the first and second dynamic confidence indices, the higher the output target rendering realism parameter. Based on the target rendering realism parameters, a virtual intelligent character image is constructed on the basis of a hand-drawn intelligent character image. Specifically, when the target rendering realism parameters are lower than a first preset threshold, an abstract and conceptual virtual intelligent character image is constructed; when the target rendering realism parameters are higher than a second preset threshold, a high-fidelity and detailed virtual intelligent character image is constructed.

4. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 3, characterized in that: In step S3, constructing a virtual interactive scene based on a virtual intelligent character specifically includes: The initial global atmosphere parameters of the virtual interactive scene are set according to the target rendering realism parameters; the initial global atmosphere parameters include at least volumetric fog density and depth blur; wherein, volumetric fog density is inversely proportional to the target rendering realism parameters, and depth blur is inversely proportional to the target rendering realism parameters; The character style is extracted from hand-drawn intelligent character images, and the corresponding set of scene elements is matched in a preset historical scene library; the set of scene elements includes natural landscape entities and architectural structure entities. Based on the virtual intelligent character image, the matched set of scene elements is stylized and rendered to construct a virtual interactive scene that is consistent with the rendering style of the virtual intelligent character image.

5. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 4, characterized in that: In step S3, placing the virtual intelligent character image in a virtual interactive scene specifically includes: The virtual intelligent character is deployed at a distant coordinate point in the virtual interactive scene, far from the virtual viewpoint; and the initial orientation vector of the virtual intelligent character is adjusted so that it faces away from the virtual viewpoint. Based on the entity attributes in the scene element set, semi-transparent occlusion entities are deployed on the visual path between the virtual viewpoint and the distant coordinate point; the semi-transparent occlusion entities are configured as partition models with non-uniform transparency fields or hollow texture mapping to perform local occlusion processing on the virtual intelligent character image. By using dynamic light sources in a virtual interactive scene to project light and shadow onto a semi-transparent obscuring entity, a virtual intelligent character image is projected onto the semi-transparent obscuring entity in real time.

6. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 5, characterized in that: In step S4, based on the current virtual interaction scenario, generating and outputting guiding voice specifically includes: Extract environmental style feature words and rendering realism feature words from the current virtual interaction scene; By inputting environmental style features, rendering realism features, and user interaction speech into a large language model, guiding speech containing environmental metaphors is generated.

7. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 6, characterized in that: In step S5, semantic analysis is performed on the feedback voice to determine the matching degree between the currently displayed virtual intelligent character and the user's expectations. Specifically, this includes: Extract the user's feedback voice in response to the guided voice input, perform natural language processing on the feedback voice, and input it into a large language model to obtain the user's satisfaction evaluation value for the current virtual interaction scenario; Based on the shared rendering relationship between virtual interactive scenes and virtual intelligent character images, the satisfaction evaluation value of virtual interactive scenes is mapped to the matching degree between the current virtual intelligent character image and the user's expectations.

8. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 7, characterized in that: In step S5, if the matching degree does not meet the preset conditions, the virtual intelligent character image is modified according to the feedback voice, specifically including: Semantic elements representing users' evaluation of virtual interaction scenarios are extracted from feedback voice, and these semantic elements are correlated with environmental style feature words and rendering realism feature words in the current virtual interaction scenario to determine the rendering dimension labels to be corrected. The rendering dimension labels include at least the realism-abstract dimension, the cool-warm color dimension, and the hard-soft texture dimension. Calculate the offset vector of semantic elements under each rendering dimension label, and use the offset vector to adjust the virtual intelligent character image while keeping the current outline topology unchanged.

9. The cognitive reconstruction method based on the imagery of intelligent figures according to claim 8, characterized in that: In step S5, improving the display and recognition of the virtual intelligent character in the virtual interaction scene during subsequent voice interaction specifically includes: The system dynamically adjusts the spatial distance between the virtual viewpoint and the virtual intelligent character, gradually reducing the Euclidean distance between the virtual viewpoint and the distant coordinate point to make the user visually closer to the virtual intelligent character. Based on changes in the position of the virtual viewpoint, the system gradually adjusts the transparency parameter of the semi-transparent occluding entity or removes the semi-transparent occluding entity to remove partial occlusion of the virtual intelligent character. When the distance between the virtual viewpoint and the virtual intelligent character is less than a preset interaction distance threshold, the system modifies the initial orientation vector of the virtual intelligent character so that the virtual intelligent character faces the virtual viewpoint and outputs a response voice with its face towards the user.

Images