Auxiliary correction method and system for color vision disorder

Abstract

The invention provides an auxiliary correction method and system for color vision disorder, and the method comprises the following steps: S1, carrying out the color and texture decoupling of an input image through a perception consistent state space module, and extracting a global color distribution feature and a perception invariant texture feature; generating a personalized enhanced image in combination with dynamic color offset adjustment and CVD perception quality evaluation; s2, constructing a dynamic heterogeneous marshalling architecture, dynamically combining a high-precision model and a lightweight module, and realizing a self-adaptive network architecture with multi-level complexity and performance balance through frequency domain sensing decoupling, heterogeneous multi-dimensional static kernel and implicit neural frequency domain acceleration technologies; and S3, acquiring image data in real time by adopting an embedded hardware platform, optimizing a processing flow through a dynamic complexity threshold constraint model, outputting an enhanced image in combination with AR / VR equipment and a standardized interface, and guiding continuous optimization of a personalized color enhancement algorithm through effect analysis evaluation feedback.

Description

Technical Field

[0001] The present invention proposes an auxiliary correction method and system for color vision impairment, and relates to the field of image enhancement. Background Art

[0002] In recent years, experts in various fields have been conducting research to enhance the color perception ability of CVD patients, including gene therapy, CVD correction glasses, and image processing. Gene therapy has made some progress, but its effectiveness still needs further verification. In some developed countries, CVD correction glasses are commonly used to help CVD patients. However, the correction effect of these glasses is unstable and easily affected by light intensity. Furthermore, the high cost of CVD correction glasses limits their widespread use.

[0003] Current image processing-based recoloring methods generally struggle to effectively balance color contrast enhancement with naturalness, and lack detailed modeling of the diversity of CVD diseases and the differences in cone responses among individual patients. Existing AI-assisted correction technologies have improved generalization capabilities through big data training, but they still lack sufficient attention to disease diversity, ignore individual patient differences, and lack mechanisms for evaluating algorithm performance. Generative AI, with its powerful data generation and processing capabilities, has opened up new avenues for CVD-assisted correction. Summary of the Invention

[0004] In view of this, in order to fill the gaps and shortcomings of the existing technology, the present invention proposes an auxiliary correction method and system for color vision impairment, which is expected to provide CVD patients with more accurate and personalized auxiliary correction solutions to help them better cope with visual challenges.

[0005] The present invention proposes an auxiliary correction method and system for color vision impairment, including the following contents:

[0006] A method and system for assisting correction of color vision impairment, comprising the following:

[0007] Step S1: Generate CVD perception-friendly images by decoupling color and texture, extract global color distribution and perception-invariant texture features from the input image, and combine dynamic color offset adjustment with quality evaluation scores to generate personalized enhanced images;

[0008] Step S2: Adopting a design approach that prioritizes both high-precision models and lightweight models, model modules of varying complexity are dynamically combined through a dynamic heterogeneous grouping architecture, further constructing an adaptive network architecture with multi-level complexity and performance trade-offs.

[0009] Step S3: Use an embedded hardware platform to acquire, process, and output enhanced images in real time. At the same time, combine standardized interfaces with clinical trials to verify the system effect, and optimize algorithm performance through quantitative and qualitative analysis.

[0010] Furthermore, step S1 includes the following contents:

[0011] Step S11: decoupling the color and texture of the input image, separating the global color distribution features and the perceptually invariant texture features through the perceptually consistent state space module;

[0012] Step S12: Based on the complexity of the image X perceived by patients with different degrees of CVD, the color offset δ is used to encode the color deviation information and dynamically guide the enhancement network. At the same time, the CVD perception image automatic evaluation model score S is used to dynamically adjust the color enhancement naturalness and contrast, and finally generate a high-quality CVD perception-friendly image X en ; Further, personalized color enhancement framework M K (·) Include the following:

[0013] X en =M K (X,δ,S;θ k )

[0014] where θ K Indicates M K (·) parameters; Based on the constructed large-scale CVD perception image dataset, a high-precision personalized color enhancement model is designed to enhance image color.

[0015] Furthermore, step S1 also includes the following contents:

[0016] Step S13: Introduce a small sample learning mechanism to construct a small sample adaptive color shift estimation module S(·); by adopting a metric learning method, divide the data into a support set and a query set, and use prototype learning to calculate the color shift prototypes of different CVD types c in the support set; for the input original color shift, extract its feature representation through the embedding network E(·), and calculate the similarity measure d(δ,P c ); further, the softmax activation function is used to calculate its category probability, and the offset is adjusted based on the prototype of the best matching category to generate a personalized color offset component. The above process includes the following formula:

[0017]

[0018] δ * =∑ c P(y=c|δ)P c

[0019] Where P(y=c|δ) is the probability distribution of the input belonging to CVD type c; the personalized color offset component δ * ;P c express:

[0020] Step S14: Designing a texture perception-invariant adaptive color mapping model for CVD patients to generate a CVD perception-friendly image X from the input image X en First, a perceptually consistent state space module Φ(·) is introduced into the network to extract the global features of the input image X and decouple the image from color and texture to generate the global color distribution feature F C and perceptually invariant texture features F T ; Secondly, the input color offset δ is used to generate a personalized color offset component δ through a small sample adaptive color deviation estimation module S(·) * , which is used to guide the adaptive correction of the color shift adjustment state space module G(·) C , and generate the color conversion matrix G C ; Then, the quality perception guided color dynamic adjustment module Q(·) is introduced into the network, and the quality evaluation score S is used to guide G C Adaptively adjust color contrast and naturalness to generate a CVD color-friendly matrix G S Finally, F T With G S The CVD perception-friendly image is generated by the state space reconstruction module R(·). The process of step S14 includes the following formula:

[0021] {F c , F T}=Φ(X)

[0022] δ * =S(δ)

[0023] G c =G(F c , δ * )

[0024] X en =R(F T , Q(G c , S))

[0025] Finally, we get the CVD perception-friendly image X en .

[0026] Furthermore, step S1 also includes the following contents:

[0027] Step S15: Design a loss function including color distribution loss, perception loss, and quality evaluation consistency loss to meet the requirements of CVD perception image color enhancement; the loss function includes the following: per represents the perceptual loss function, L color represents the color distribution loss function, L quality represents the quality evaluation consistency loss function, Y q represents the image quality evaluation score of Y, S represents the enhanced image quality score output by the automatic evaluation model of CVD perception images, λ1, λ2, and λ3 represent the weight coefficients of each constraint item, L is the total loss function, and step S15 includes the following formula:

[0028] L color =D KL (X en ||Y)

[0029] L quality =min|SY q |

[0030] L per =∑ l ||VGG l (X en )-VGG l (Y)||

[0031] L=λ1L color +λ2L per +λ3L quality

[0032] Among them, D KL (·) represents the divergence used to measure the information loss between two distributions, VGG l (·) represents the feature map extracted by pre-trained VGG at the lth layer of the deep network.

[0033] Furthermore, step S2 includes the following contents:

[0034] Step S21: establishing a color enhancement optimization model based on a dynamic complexity threshold constraint, wherein the complexity threshold is adjusted based on different application scenarios, and a research approach is adopted that emphasizes both improved color conversion and lightweight neural networks, with both competing approaches.

[0035] Step S22: In the color conversion strategy, different CVD types are first classified, their negative impact on the perceived quality of the image is evaluated, and the positive gains of the negative impact are repaired, which are listed as Q0, Q1, ..., Q K , Q N-1; Further, for different types of CVD patients, color conversion models M0, M1, ..., M K , M N-1 , whose expected complexity is r0, r1, ..., r N-1 Finally, according to the optimal stopping theory, the above rQ combinations are sorted and the sorting results are re-labeled from 0 to N-1; for any image to be tested and the r T The optimal stopping theory first determines the maximum model supported by the complexity constraint and the maximum K value of the maximum model; further, the optimal stopping theory obtains the optimal color conversion model corresponding to different CVD types based on K+1 CVD types from 0 to K.

[0036] Furthermore, step S2 also includes the following contents:

[0037] Step S23: Based on the CVD user perception optimization requirements and lightweight strategy requirements, construct the color offset component δ * The dual guidance mechanism with the CVD perception image evaluation model score S optimizes the lightweight color enhancement model's processing of the CVD perception image X, thereby ensuring the enhanced image X en It is more in line with the visual perception needs of CVD users, ensuring that the output image meets the requirements of color compensation and naturalness preservation at the same time. By introducing heterogeneous multi-dimensional static kernel and implicit neural frequency domain acceleration computing technology, the computational complexity is further reduced, so that the model can be efficiently deployed in the lightweight framework M0(·) of mobile devices. Step S23 includes the following formula:

[0038] X en =M0(X,δ * ,S;θ0)

[0039] where θ0 represents the parameter of M0(·);

[0040] Step S24: Decoupling module Φ through frequency domain perception t (·) Optimize the perceptually consistent state space module Φ t (·) , through implicit neural frequency domain analysis, global features are efficiently extracted, and color and texture are decoupled to generate global color distribution features F C and perceptually invariant texture features F T .

[0041] Furthermore, step S2 also includes the following contents:

[0042] Step S25: Based on the need to reduce model complexity, dynamic color shift is introduced to adjust the heterogeneous multi-dimensional static kernel G t (·) and quality-aware color mapping heterogeneous multidimensional static kernel Q t(·) respectively optimize the color deviation adjustment state space module G(·) and the quality perception guided color dynamic adjustment module Q(·). The heterogeneous multi-dimensional static core achieves parameter sharing by storing key color conversion mapping, while taking into account parameter efficiency and dynamic adjustment capability; G t (·) Using the color shift component δ * The color cast correction kernel is guided to dynamically adjust the global color map, while Q t (·) The role includes using the quality evaluation score S to guide the quality perception static kernel to adaptively adjust the color contrast to ensure that the enhanced image meets the CVD visual perception optimization requirements;

[0043] Step S26: Using implicit neural frequency domain acceleration computing technology to refine the reconstruction color map and perceptual invariant texture while avoiding the introduction of a large number of neural network parameters; thus, constructing the implicit neural frequency domain reconstruction module R t (·) Optimize the state-space reconstruction module R(·), use the implicit neural frequency domain accelerator to replace the traditional spatial domain reconstruction network, and efficiently reconstruct the color-friendly and texture-invariant CVD perception-friendly image X en ; Further, the lightweight neural network strategy includes the following:

[0044] {F c , F T}=φ t (X)

[0045] δ * =S(δ)

[0046] X en =R t (F T , Q t (G t (F c , δ * ), S))

[0047] Considering that the lightweight personalized color enhancement model M0(·) has a low computational complexity r0, the network is expected to achieve real-time processing on mobile devices and strike a balance between color enhancement and naturalness preservation to obtain a moderate restoration gain Q0;

[0048] Step S27: Adopting a design approach that gives equal importance to high-precision models and lightweight models, model modules of different complexities are dynamically combined through a dynamic heterogeneous grouping architecture, thereby constructing an adaptive network architecture with multi-level complexity and performance trade-offs. This design can not only flexibly respond to performance requirements in different application scenarios, but also achieve an optimal balance between model accuracy and computational efficiency.

[0049] Furthermore, step S3 includes the following contents:

[0050] Step S31: Based on the high-performance color enhancement algorithm proposed above, an embedded deployment method with an AI computer as the core is adopted. The camera collects image data in real time and transmits it to the AI ​​computer board to realize personalized color enhancement through the proposed enhancement algorithm;

[0051] Step S32: The enhanced image can be presented in real time through AR / VR glasses, LCD displays, or projectors, using the OpenXR standardized interface to provide higher-precision color adjustment and a stable and natural visual experience;

[0052] Step S33: Continuously optimize the personalized color enhancement algorithm through effect analysis and evaluation feedback; and continuously improve the system to meet user needs.

[0053] According to a second aspect of the present invention, an auxiliary correction system for color vision impairment includes an electronic device, wherein the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and is characterized in that when the processor executes the computer program, it implements an auxiliary correction method for color vision impairment as described in any one of the present inventions.

[0054] According to a third aspect of the present invention, an auxiliary correction system for color vision impairment includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements an auxiliary correction method for color vision impairment as described in any one of the present inventions.

[0055] Among them, CVD is the abbreviation of color vision deficiency, also known as color vision disorder, which refers to the condition in which the ability to see and distinguish colors is lower than that of ordinary people.

[0056] Among them, Softmax is an activation function that can normalize a numerical vector into a probability distribution vector, and the sum of each probabilities is 1.

[0057] Among them, the AI ​​computers include the Jetson Nan AI computer, which has the performance and energy efficiency to run modern AI workloads, run multiple neural networks in parallel, and process data from multiple high-definition sensors at the same time.

[0058] The present invention has the following advantages:

[0059] The present invention can accurately meet the visual needs of CVD patients, thereby improving their color perception quality in daily life.

[0060] The present invention can accurately match user needs, effectively balance color contrast enhancement and naturalness maintenance, and significantly improve patients' color perception ability.

[0061] The present invention can flexibly adapt to the real-time processing requirements of mobile terminals or embedded devices, taking into account both high efficiency and practicality.

[0062] The present invention can ensure the stable performance of the system in complex lighting environments, provide users with a natural and smooth visual experience, and has wide applicability and scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is an overall workflow diagram in an embodiment of the present invention.

[0064] Figure 2 1 is a flow chart of clinical trials in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings.

[0066] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0067] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form, and it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0068] like Figure 1 As shown in the overall workflow diagram of the embodiment of the present invention, the present invention proposes an auxiliary correction method and system for color vision impairment, including the following contents:

[0069] Step S1: Generate CVD perception-friendly images by decoupling color and texture, extract global color distribution and perception-invariant texture features from the input image, and combine dynamic color offset adjustment with quality evaluation scores to generate personalized enhanced images;

[0070] Step S2: Adopting a design approach that prioritizes both high-precision models and lightweight models, model modules of varying complexity are dynamically combined through a dynamic heterogeneous grouping architecture, further constructing an adaptive network architecture with multi-level complexity and performance trade-offs.

[0071] Step S3: Use an embedded hardware platform to acquire, process, and output enhanced images in real time. At the same time, combine standardized interfaces with clinical trials to verify the system effect, and optimize algorithm performance through quantitative and qualitative analysis.

[0072] Furthermore, step S1 includes the following contents:

[0073] Step S11: decoupling the color and texture of the input image, separating the global color distribution features and the perceptually invariant texture features through the perceptually consistent state space module;

[0074] Step S12: Based on the complexity of the image X perceived by patients with different degrees of CVD, the color offset δ is used to encode the color deviation information and dynamically guide the enhancement network. At the same time, the CVD perception image automatic evaluation model score S is used to dynamically adjust the color enhancement naturalness and contrast, and finally generate a high-quality CVD perception-friendly image X en ; Further, personalized color enhancement framework M K (·) Include the following:

[0075] X en =M K (X,δ,S,θ k )

[0076] where θ K Indicates M K (·) parameters; Based on the constructed large-scale CVD perception image dataset, a high-precision personalized color enhancement model is designed to enhance image color.

[0077] Furthermore, step S1 also includes the following contents:

[0078] Step S13: Introduce a small sample learning mechanism to construct a small sample adaptive color shift estimation module S(·); by adopting a metric learning method, divide the data into a support set and a query set, and use prototype learning to calculate the color shift prototypes of different CVD types c in the support set; for the input original color shift, extract its feature representation through the embedding network E(·), and calculate the similarity measure d(δ,P c ); further, softmax is used to calculate its category probability, and the offset is adjusted based on the prototype of the best matching category to generate a personalized color offset component. The above process includes the following formula:

[0079]

[0080] δ * =∑ c P(y=c|δ)P c

[0081] Where P(y=c|δ) is the probability distribution of the input belonging to CVD type c; the personalized color offset component δ * ;P c express:

[0082] Step S14: Designing a texture perception-invariant adaptive color mapping model for CVD patients to generate a CVD perception-friendly image X from the input image X en First, a perceptually consistent state space module Φ(·) is introduced into the network to extract the global features of the input image X and decouple the image from color and texture to generate the global color distribution feature F C and perceptually invariant texture features F T ; Secondly, the input color offset δ is used to generate a personalized color offset component δ through a small sample adaptive color deviation estimation module S(·) * , which is used to guide the adaptive correction of the color shift adjustment state space module G(·) C , and generate the color conversion matrix G C ; Then, the quality perception guided color dynamic adjustment module Q(·) is introduced into the network, and the quality evaluation score S is used to guide G C Adaptively adjust color contrast and naturalness to generate a CVD color-friendly matrix G S Finally, F T With G S The CVD perception-friendly image is generated by the state space reconstruction module R(·). The process of step S14 includes the following formula:

[0083] {F c , F T}=Φ(X)

[0084] δ * =S(δ)

[0085] G c =G(F c , δ * )

[0086] X en =R(F T , Q(G c , S))

[0087] Furthermore, step S1 also includes the following contents:

[0088] Step S15: Design a loss function including color distribution loss, perception loss, and quality evaluation consistency loss to meet the requirements of CVD perception image color enhancement; the loss function includes the following: per represents the perceptual loss function, L color represents the color distribution loss function, Lquality represents the quality evaluation consistency loss function, Y q represents the image quality evaluation score of Y, S represents the enhanced image quality score output by the automatic evaluation model of CVD perception images, λ1, λ2, and λ3 represent the weight coefficients of each constraint item, L is the total loss function, and step S15 includes the following formula:

[0089] L color =D KL (X en ||Y)

[0090] L quality =min|SY q |

[0091] L lpre =∑ l ||VGG l (X en )-VGG l (Y)||

[0092] L=λ1L color +λ2L per +λ3L quality

[0093] Among them, D KL (·) represents the divergence used to measure the information loss between two distributions, VGG l (·) represents the feature map extracted by pre-trained VGG at the lth layer of the deep network.

[0094] Furthermore, step S2 includes the following contents:

[0095] Step S21: Establish a color enhancement optimization model based on a dynamic complexity threshold constraint. Adjust the complexity threshold based on different application scenarios. Adopt a research approach that prioritizes both improved color conversion and lightweight neural networks, with the best strategy being the final strategy.

[0096] Step S22: In the color conversion strategy, different CVD types are first classified, their negative impact on the perceived quality of the image is evaluated, and the positive gains of the negative impact are repaired, which are listed as Q0, Q1, ..., Q K , Q N-1 ; Further, for different types of CVD patients, color conversion models M0, M1, ..., M K , M N-1 , whose expected complexity is r0, r1, ..., r N-1Finally, according to the optimal stopping theory, the above rQ combinations are sorted and the sorting results are re-labeled from 0 to N-1; for any image to be tested and the r T The optimal stopping theory algorithm first determines the maximum model that can be supported by the complexity constraint, that is, the maximum value K that can fully utilize the constraint. The optimal stopping theory further obtains the optimal color conversion model corresponding to different CVD types based on K+1 CVD types from 0 to K.

[0097] Furthermore, step S2 also includes the following contents:

[0098] Step S23: Based on the CVD user perception optimization requirements and lightweight strategy requirements, construct the color offset component δ * The dual guidance mechanism with the CVD perception image evaluation model score S optimizes the lightweight color enhancement model's processing of the CVD perception image X, thereby ensuring the enhanced image X en It better meets the visual perception needs of CVD users and ensures that the output image meets the requirements of color compensation and naturalness preservation. By introducing heterogeneous multi-dimensional static kernels and implicit neural frequency domain acceleration computing technology, the computational complexity is further reduced, enabling the model to be efficiently deployed on the lightweight framework M0(·) of mobile devices. Step S23 includes the following formula:

[0099] X en =M0(X,δ * ,S,θ0)

[0100] where θ0 represents the parameter of M0(·);

[0101] Step S24: Decoupling module Φ through frequency domain perception t (·) Optimize the perceptually consistent state space module Φ t (·) , through implicit neural frequency domain analysis, global features are efficiently extracted, and color and texture are decoupled to generate global color distribution features F C and perceptually invariant texture features F T .

[0102] Furthermore, step S2 also includes the following contents:

[0103] Step S25: Based on the need to reduce model complexity, dynamic color shift is introduced to adjust the heterogeneous multi-dimensional static kernel G t (·) and quality-aware color mapping heterogeneous multidimensional static kernel Q t (·) respectively optimize the color deviation adjustment state space module G(·) and the quality perception guided color dynamic adjustment module Q(·). The heterogeneous multi-dimensional static core achieves parameter sharing by storing key color conversion mapping, while taking into account parameter efficiency and dynamic adjustment capability; G t(·) Using the color shift component δ * The color cast correction kernel is guided to dynamically adjust the global color map, while Q t (·) Using the quality evaluation score S to guide the quality-aware static kernel to adaptively adjust the color contrast, ensuring that the enhanced image meets the CVD visual perception optimization requirements;

[0104] Step S26: Using implicit neural frequency domain acceleration computing technology to refine the reconstruction color map and perceptual invariant texture, while avoiding the introduction of a large number of neural network parameters; thus, constructing the implicit neural frequency domain reconstruction module R t (·) Optimize the state-space reconstruction module R(·), use the implicit neural frequency domain accelerator to replace the traditional spatial domain reconstruction network, and efficiently reconstruct the color-friendly and texture-invariant CVD perception-friendly image X en ; Further, the lightweight neural network strategy includes the following:

[0105] {F c , F T}=Φ t (X)

[0106] δ * =S(δ)

[0107] X en =R t (F T , Q t (G t (F c , δ * ), S))

[0108] Considering that the lightweight personalized color enhancement model M0(·) has a low computational complexity r0, the network is expected to achieve real-time processing on mobile devices and strike a balance between color enhancement and naturalness preservation to obtain a moderate restoration gain Q0;

[0109] Step S27: Adopting a design approach that gives equal importance to high-precision models and lightweight models, model modules of different complexities are dynamically combined through a dynamic heterogeneous grouping architecture, thereby constructing an adaptive network architecture with multi-level complexity and performance trade-offs. This design can not only flexibly respond to performance requirements in different application scenarios, but also achieve an optimal balance between model accuracy and computational efficiency.

[0110] Furthermore, step S3 includes the following contents:

[0111] Step S31: Based on the high-performance color enhancement algorithm proposed above, an embedded deployment method with Jetson Nano as the core is adopted. The camera collects image data in real time and transmits it to the Jetson Nano board to realize personalized color enhancement through the proposed enhancement algorithm.

[0112] Step S32: The enhanced image can be presented in real time through AR / VR glasses, LCD displays, or projectors, using the OpenXR standardized interface to provide higher-precision color adjustment and a stable and natural visual experience;

[0113] Step S33: Continuously optimize the personalized color enhancement algorithm through effect analysis and evaluation feedback; and continuously improve the system to meet user needs.

[0114] According to a second aspect of the present invention, an auxiliary correction system for color vision impairment includes an electronic device, wherein the electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and is characterized in that when the processor executes the computer program, it implements an auxiliary correction method for color vision impairment as described in any one of the present inventions.

[0115] According to a third aspect of the present invention, an auxiliary correction system for color vision impairment includes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements an auxiliary correction method for color vision impairment as described in any one of the present inventions.

[0116] Among them, CVD is the abbreviation of color vision deficiency, also known as color vision disorder, which refers to the condition in which the ability to see and distinguish colors is lower than that of ordinary people.

[0117] Among them, Softmax is an activation function that can normalize a numerical vector into a probability distribution vector, and the sum of each probabilities is 1.

[0118] Among them, the AI ​​computers include the Jetson Nan AI computer, which has the performance and energy efficiency to run modern AI workloads, run multiple neural networks in parallel, and process data from multiple high-definition sensors at the same time.

[0119] In one embodiment of the present invention, 100 subjects were recruited. The experimental group included 50 patients with CVD, including 30 with red-green CVD (mild = 8, moderate = 18, severe = 4) and 20 with yellow-blue CVD (mild = 4, moderate = 8, severe = 8). The control group consisted of 50 subjects with normal color vision, aged between 18 and 60 years old, with a balanced male-female ratio. During the experiment, factors such as the environment (indoor vs. outdoor) and lighting (natural vs. artificial) remained consistent. Different subjective experimental tasks were designed for four color discrimination tasks faced by CVD patients (comparative, indicative, symbolic, and aesthetic). For the comparative task, the subject was required to compare the colors in the image to find similarities or differences and use a scale to evaluate their discrimination ability and comfort. For the indicative task, the subject was required to identify the colors in the image based on the color name or reference and record their feedback. For the symbolic task, the subject was required to identify the meaning represented by the colors in the image and evaluate its acceptability and naturalness. For aesthetic tasks, participants are required to evaluate the aesthetic effects of colors in images and record their subjective perceptions. The collected subjective data will be organized, categorized, and statistically analyzed to compare differences between tasks and explore correlations with other factors. This project aims to develop a behavioral method based on human visual characteristics and attention mechanisms to capture the characteristics of participants' perception of image color.

[0120] In one embodiment of the present invention, behavioral data is collected by completing a color vision search task, which involves obtaining color information from an image. The collected data is statistically analyzed for accuracy and reaction time, and the behavioral performance of the observers is compared and a correlation analysis is performed. During the test, problems such as sample bias, environmental control, task design, and data analysis may be encountered. In order to optimize the test, it is proposed to adopt a diversified sample source, strengthen environmental control and standardization, optimize task design, and select appropriate data analysis methods. Finally, based on the patient's feedback and actual situation, the order and duration of the test tasks are adjusted to ensure that the test process does not cause excessive stress and discomfort to the patient.

[0121] like Figure 2As shown in the clinical trial flow chart of an embodiment of the present invention, a combination of quantitative and qualitative methods is used in one embodiment of the present invention to ensure the reliability and comprehensiveness of the collected data. Quantitative analysis can be performed using standardized tests such as the Ishihara test and the Farnsworth-Munsell 100Hue test to evaluate the performance improvement of color enhancement technology on specific tasks for CVD patients. Subjective feedback is based on four color discrimination tasks designed in the clinical trial, and further analyzes the intuitive perceptions and personal preferences of CVD patients through interviews, questionnaires, and user logs. A key goal of color enhancement in CVD-perceived images is to compensate for the contrast loss in patient perception. Subjective evaluation data is obtained by evaluating the enhanced images based on naturalness, contrast, and preference, and then analyzed and calculated to quantify user satisfaction with the color enhancement effect. User evaluation and feedback are indispensable in the iteration process of the CVD assistive device (i.e., the demonstration system constructed above), and can directly influence the design and optimization of the assistive device, making color enhancement technology more suitable for the actual needs of CVD patients.

[0122] In one embodiment of the present invention, based on the automatic evaluation system of CVD perception images, the automatic evaluation model score S is first calculated. In order to comprehensively evaluate the effect of color enhancement technology, an effect analysis system S' is designed. Specifically, first, a correlation analysis is performed by combining subjective evaluation data with S to verify the characterization ability of the automatic evaluation model score on the visual experience of CVD patients. If the correlation is strong, and both S and the score show a high evaluation, it indicates that the evaluation system can effectively quantify and reflect the improvement effect of the enhancement algorithm on the visual perception of CVD patients, and the enhancement algorithm itself has a significant effect in practical applications; if the correlation is weak, it indicates that the existing evaluation system may not fully capture the subjective perception differences of the enhancement effect, and the evaluation criteria need to be optimized:

[0123] S'=αS+βQ sub

[0124] The meanings of the various parameter symbols are as follows: S represents the automatic evaluation model score; S' represents the effect analysis system; Q sub represents subjective evaluation data, and α and β represent the parameters that need to be trained.

[0125] The optimized evaluation system is used to further guide the design and optimization of color enhancement networks, so that they can more accurately meet the visual needs of CVD patients, thereby improving the quality of color perception in their daily lives.

[0126] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions and effects do not exceed the scope of the technical solution of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. An auxiliary correction method for color vision impairment, characterized in that: The following steps are involved: Step S1: Generate CVD perception-friendly images by decoupling color and texture, extract global color distribution and perception-invariant texture features from the input image, and combine dynamic color offset adjustment with quality evaluation scores to generate personalized enhanced images; Step S2: Adopting a design approach that prioritizes both high-precision models and lightweight models, model modules of varying complexity are dynamically combined through a dynamic heterogeneous grouping architecture, further constructing an adaptive network architecture with multi-level complexity and performance trade-offs. Step S3: Use an embedded hardware platform to acquire, process, and output enhanced images in real time. At the same time, combine standardized interfaces with clinical trials to verify the system effect, and optimize algorithm performance through quantitative and qualitative analysis.

2. The auxiliary correction method for color vision impairment according to claim 1, characterized in that: Step S1 includes the following contents: Step S11: decoupling the color and texture of the input image, separating the global color distribution features and the perceptually invariant texture features through the perceptually consistent state space module; Step S12: Based on the complexity of the image X perceived by patients with different degrees of CVD, the color offset δ is used to encode the color deviation information and dynamically guide the enhancement network. At the same time, the CVD perception image automatic evaluation model score S is used to dynamically adjust the color enhancement naturalness and contrast, and finally generate a high-quality CVD perception-friendly image X en ; Further, personalized color enhancement framework M K (·) Include the following: X en =M K (X,δ,S;θ k ) where θ K Indicates M K (·) parameters; Based on the constructed large-scale CVD perception image dataset, a high-precision personalized color enhancement model is designed to enhance image color.

3. The auxiliary correction method for color vision impairment according to claim 2, characterized in that: Step S1 also includes the following: Step S13: Introduce a small sample learning mechanism to construct a small sample adaptive color shift estimation module S(·); by adopting a metric learning method, divide the data into a support set and a query set, and use prototype learning to calculate the color shift prototypes of different CVD types c in the support set; for the input original color shift, extract its feature representation through the embedding network E(·), and calculate the similarity measure d(δ,P c ); further, the softmax activation function is used to calculate its category probability, and the offset is adjusted based on the prototype of the best matching category to generate a personalized color offset component. The above process includes the following formula: d * =∑ c P(y=c|δ)P c Where P(y=c|δ) is the probability distribution of the input belonging to CVD type c; the personalized color offset component δ * ;P c Represents the color shift prototype; Step S14: Designing a texture perception-invariant adaptive color mapping model for CVD patients to generate a CVD perception-friendly image X from the input image X en First, a perceptually consistent state space module Φ(·) is introduced into the network to extract the global features of the input image X and decouple the image from color and texture to generate the global color distribution feature F C and perceptually invariant texture features F T ; Secondly, the input color offset δ is used to generate a personalized color offset component δ through a small sample adaptive color deviation estimation module S(·) * , which is used to guide the adaptive correction of the color shift adjustment state space module G(·) C , and generate the color conversion matrix G C ; Then, the quality perception guided color dynamic adjustment module Q(·) is introduced into the network, and the quality evaluation score S is used to guide G C Adaptively adjust color contrast and naturalness to generate a CVD color-friendly matrix G S Finally, F T With G S The CVD perception-friendly image is generated by the state space reconstruction module R(·). The process of step S14 includes the following formula: {F c ,F T }=Φ(X) d * =S(δ) G c =G(F c ,d * ) X en =R(F T ,Q(G c ,S)) Finally, we get the CVD perception-friendly image X en .

4. The auxiliary correction method for color vision impairment according to claim 3, characterized in that: Step S1 also includes the following: Step S15: Design a loss function including color distribution loss, perception loss, and quality evaluation consistency loss to meet the requirements of CVD perception image color enhancement; the loss function includes the following: per represents the perceptual loss function, L color represents the color distribution loss function, L quality represents the quality evaluation consistency loss function, Y q represents the image quality evaluation score of Y, S represents the enhanced image quality score output by the automatic evaluation model of CVD perception images, λ1, λ2, and λ3 represent the weight coefficients of each constraint item, L is the total loss function, and step S15 includes the following formula: L color =D KL (X en ||Y) L quality =min|SY q | L per =∑l||VGG l (X en )-VGG l (Y)|| L=λ1L color +λ2L per +λ3L quality Among them, D KL (·) represents the divergence used to measure the information loss between two distributions, VGG l (·) represents the feature map extracted by pre-trained VGG at the lth layer of the deep network.

5. The auxiliary correction method for color vision impairment according to claim 1, characterized in that: Step S2 includes the following contents: Step S21: establishing a color enhancement optimization model based on a dynamic complexity threshold constraint, wherein the complexity threshold is adjusted based on different application scenarios, and a research approach is adopted that emphasizes both improved color conversion and lightweight neural networks, with both competing approaches. Step S22: In the color conversion strategy, different CVD types are first classified, their negative impact on the perceived quality of the image is evaluated, and the positive gains of the negative impact are repaired, which are listed as Q0, Q1, ..., Q K , Q N-1 ; Further, for different types of CVD patients, color conversion models M0, M1, ..., M K , M N-1 , whose expected complexity is r0, r1, ..., r N-1 Finally, according to the optimal stopping theory, the above rQ combinations are sorted and the sorting results are re-labeled from 0 to N-1; for any image to be tested and the r T The optimal stopping theory first determines the maximum model supported by the complexity constraint and the maximum K value of the maximum model; further, the optimal stopping theory obtains the optimal color conversion model corresponding to different CVD types based on K+1 CVD types from 0 to K.

6. The auxiliary correction method for color vision impairment according to claim 5, characterized in that: Step S2 also Includes the following: Step S23: Based on the CVD user perception optimization requirements and lightweight strategy requirements, construct the color shift component δ * The dual guidance mechanism with the CVD perception image evaluation model score S optimizes the lightweight color enhancement model's processing of the CVD perception image X, thereby ensuring the enhanced image X en It is more in line with the visual perception needs of CVD users, ensuring that the output image meets the requirements of color compensation and naturalness preservation at the same time. By introducing heterogeneous multi-dimensional static kernel and implicit neural frequency domain acceleration computing technology, the computational complexity is further reduced, so that the model can be efficiently deployed in the lightweight framework M0(·) of mobile devices. Step S23 includes the following formula: X en =M0(X,δ * ,S;θ0) where θ0 represents the parameter of M0(·); Step S24: Decoupling module Φ through frequency domain perception t (·) Optimize the perceptually consistent state space module Φ t (·) , through implicit neural frequency domain analysis, global features are efficiently extracted, and color and texture are decoupled to generate global color distribution features F C and perceptually invariant texture features F T .

7. The auxiliary correction method for color vision impairment according to claim 6, characterized in that: Step S2 also includes the following: Step S25: Based on the need to reduce model complexity, dynamic color shift is introduced to adjust the heterogeneous multi-dimensional static kernel G t (·) and quality-aware color mapping heterogeneous multidimensional static kernel Q t (·) respectively optimize the color deviation adjustment state space module G(·) and the quality perception guided color dynamic adjustment module Q(·). The heterogeneous multi-dimensional static core achieves parameter sharing by storing key color conversion mapping, while taking into account parameter efficiency and dynamic adjustment capability; G t (·) Using the color shift component δ * The color cast correction kernel is guided to dynamically adjust the global color map, while Q t (·) The functions include: using the quality evaluation score S to guide the quality-aware static kernel to adaptively adjust the color contrast to ensure that the enhanced image meets the CVD visual perception optimization requirements; Step S26: Using implicit neural frequency domain acceleration computing technology to refine the reconstruction color map and perceptual invariant texture, while avoiding the introduction of a large number of neural network parameters; thus, constructing the implicit neural frequency domain reconstruction module R t (·) Optimize the state-space reconstruction module R(·), use the implicit neural frequency domain accelerator to replace the traditional spatial domain reconstruction network, and efficiently reconstruct the color-friendly and texture-invariant CVD perception-friendly image X en ; Further, the lightweight neural network strategy includes the following: {F c ,F T }=Φ t (X) d * =S(δ) X en =R t (F T ,Q t (G t (F c ,δ * ),S)) Considering that the lightweight personalized color enhancement model M0(·) has a low computational complexity r0, the network is expected to achieve real-time processing on mobile devices and strike a balance between color enhancement and naturalness preservation to obtain a moderate restoration gain Q0; Step S27: Adopting a design approach that gives equal importance to high-precision models and lightweight models, model modules of different complexities are dynamically combined through a dynamic heterogeneous grouping architecture, thereby constructing an adaptive network architecture with multi-level complexity and performance trade-offs. This design can not only flexibly respond to performance requirements in different application scenarios, but also achieve an optimal balance between model accuracy and computational efficiency.

8. The auxiliary correction method for color vision impairment according to claim 1, characterized in that: Step S3 includes the following contents: Step S31: Based on the high-performance color enhancement algorithm proposed above, an embedded deployment method with an AI computer as the core is adopted. The camera collects image data in real time and transmits it to the AI ​​computer board to realize personalized color enhancement through the proposed enhancement algorithm; Step S32: The enhanced image can be presented in real time through AR glasses, VR glasses, LCD displays, or projectors, using the OpenXR standardized interface to provide higher-precision color adjustment and a stable and natural visual experience; Step S33: Continuously optimize the personalized color enhancement algorithm through effect analysis and evaluation feedback; and continuously improve the system to meet user needs.

9. An auxiliary correction system for color vision impairment, comprising an electronic device, wherein the electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the processor implements the auxiliary correction method for color vision impairment as described in any one of claims 1 to 8.

10. An auxiliary correction system for color vision impairment, comprising a computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the computer program implements the auxiliary correction method for color vision impairment as described in any one of claims 1 to 8.

Images