An interactive processing method for infrared images

Abstract

The present invention relates to the field of image processing technology, and discloses an interactive processing method for infrared images, comprising the following steps: S1. Acquire an infrared image to be processed; S2. Preprocess the infrared image to extract noise characteristics, edge characteristics and scene characteristics; S3. Receive feedback information input by a user through a user interaction interface, including the definition of enhancement strength, denoising strength and area of ​​interest; S4. Construct an optimization model based on a variational method, and the optimization model guides image processing by describing denoising, detail enhancement and image smoothing optimization objectives; S5. Introduce partial differential equation constraints. The denoising and detail retention capabilities of infrared images are improved through energy functional optimization models and partial differential equation constraints; parameters are dynamically adjusted in combination with a user interaction mechanism to meet personalized needs; an explicit time step iteration method is adopted to improve computational efficiency while taking into account both accuracy and stability, and adapt to complex scenes.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to an interactive processing method for infrared images. Background Art

[0002] With the development of computer vision technology and sensor technology, infrared image processing technology has been widely used in many fields. Infrared images can provide key information that is difficult to obtain under visible light conditions by detecting the thermal radiation information of the target object. They are of irreplaceable importance in scenes such as night vision monitoring, industrial inspection, and medical imaging. For example, in low-light or no-light environments, infrared images can be used to identify targets; in medicine, infrared images can detect abnormal body surface temperature distribution; in industrial inspection, infrared images can be used to locate thermal anomalies in mechanical equipment.

[0003] The core of infrared image processing is to enhance the effective information in the image while suppressing the interference information. Since infrared images are affected by the imaging equipment and environmental conditions, they usually have problems of low contrast and high noise. Common infrared image processing methods mainly include hardware-based enhancement technology and software-based image algorithm optimization. The hardware method improves the image quality from the signal source by improving the optical elements or sensor design of the camera; the software method improves the image usability through algorithms such as image denoising, edge enhancement, and detail enhancement.

[0004] In software image processing technology, filter-based image denoising technology can effectively reduce noise. For example, methods such as Gaussian filtering and bilateral filtering can maintain edge information while smoothing the image; image enhancement technology based on gradient enhancement or multi-scale decomposition can highlight the key details of infrared images. In addition, traditional total variation regularization methods are also widely used in denoising and detail enhancement, and images are optimized by constructing regularized models. However, traditional methods often rely on fixed algorithm processes and parameter settings, and cannot dynamically adapt to different environmental changes and image characteristics. Especially in complex application scenarios, the processing effect is easily limited. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an interactive processing method for infrared images, which solves the problems in the prior art infrared image processing methods that are unable to effectively balance denoising and detail enhancement, lack of adaptive capabilities, and difficulty in meeting user personalized needs.

[0006] To achieve the above object, the present invention is implemented by the following technical scheme: an interactive processing method for infrared images, comprising the following steps:

[0007] S1. Obtaining an infrared image to be processed;

[0008] S2. Preprocessing the infrared image to extract noise characteristics, edge characteristics and scene characteristics;

[0009] S3. receiving feedback information input by the user through the user interaction interface, including the definition of enhancement strength, denoising strength and region of interest;

[0010] S4. constructing an optimization model based on a variational method, wherein the optimization model guides image processing by describing optimization objectives of denoising, detail enhancement, and image smoothing;

[0011] S5. Introducing partial differential equation constraints, the partial differential equation is used to control the dynamic balance between denoising and edge preservation during image processing;

[0012] S6. solving the optimization model by a numerical iteration method and optimizing the infrared image until a preset convergence condition is met;

[0013] S7. Output the processed infrared image, and dynamically update the optimization parameters based on user feedback, repeating steps S4 to S6 until a final result that meets user needs is generated.

[0014] Preferably, the optimization model is represented in the form of an energy functional, which includes a gradient term describing image edge detail enhancement, a smoothing term describing noise suppression, a fidelity term for maintaining original image information, and an interactive control term combining user feedback information.

[0015] Preferably, the feedback information input by the user in step S3 is used to adjust the weight parameters of each item in the optimization model, and the weight parameters are dynamically changed according to user needs to balance the processing effects of detail enhancement, denoising and global fidelity.

[0016] Preferably, the partial differential equation constraints include:

[0017] De-noising equations that characterize the heat diffusion process are used to smooth noise and avoid over-processing the edges;

[0018] The edge-preserving equation that characterizes anisotropic diffusion is used to avoid amplification of high-frequency noise during detail enhancement.

[0019] Preferably, the diffusion parameter in the partial differential equation is dynamically adjusted according to the gradient amplitude of the infrared image, wherein the diffusion degree of the region with higher noise is greater, while the diffusion degree of the edge region is less.

[0020] Preferably, the numerical iteration method adopts an explicit time step method to solve the problem, by gradually optimizing the image processing result in discrete time steps until a preset convergence condition is reached.

[0021] Preferably, the spatial derivative calculation in the numerical iteration method is discretized using a finite difference method, the gradient calculation is used to enhance edge information, and the Laplace operator calculation is used to denoise and smooth the image.

[0022] Preferably, the user interaction interface includes:

[0023] Slider controls to adjust image enhancement and denoising strength;

[0024] The region annotation tool is used to define the user's area of ​​interest so that image processing can be prioritized within that area.

[0025] Preferably, the method initializes weight parameters in the optimization model through scenario analysis, comprising:

[0026] Set the weight of the denoising term according to the noise intensity;

[0027] Setting the weight of detail enhancement items according to image edge characteristics;

[0028] Adjust the weight of the global fidelity term based on scene complexity.

[0029] Preferably, in the numerical iteration process of step S6, the iteration speed is optimized according to the change amplitude of the current iteration result by dynamically adjusting the time step, wherein:

[0030] When the change amplitude of the current image processing result is greater than the set threshold, the time step is reduced to improve the convergence stability;

[0031] When the change amplitude of the current image processing result is less than the set threshold, the time step is increased to accelerate the iterative convergence.

[0032] The present invention provides an interactive processing method for infrared images, which has the following beneficial effects:

[0033] 1. The present invention constructs an optimization model based on energy functionals and realizes dynamic adjustment of different areas of the image in combination with partial differential equation constraints. The denoising process is completed by the heat diffusion equation, which can effectively smooth the noise in the low-frequency area; at the same time, the diffusion coefficient design controlled by the gradient amplitude is combined to significantly reduce the diffusion intensity in the detail area, thereby effectively denoising while protecting the edge features. Compared with the traditional fixed parameter method, the present invention can adaptively adjust the processing intensity and significantly improve the image quality in a high-noise environment.

[0034] 2. The present invention innovatively introduces a user interaction mechanism to dynamically adjust the weight parameters and feedback items of the optimization model through user input. Users can choose the enhancement strength, denoising strength, and mark the region of interest. The system can respond to user needs in real time and dynamically optimize the image processing results. Compared with existing automated image processing methods, the present invention gives users greater flexibility, making the processing results more suitable for actual application scenarios and meeting the personalized requirements of different users for image quality.

[0035] 3. By combining the scene characteristics of the image, the optimization model of the present invention can dynamically adjust the task weights. The design of the diffusion coefficient and the edge control function are dynamically adjusted according to the image gradient amplitude, adapting to the image characteristics under different environmental conditions, especially in complex and changeable practical application scenarios, and can always provide efficient and high-quality image processing effects.

[0036] 4. The present invention adopts a numerical iteration method with an explicit time step, and ensures the stability and efficiency of numerical calculation by discretizing the gradient and Laplace operator with finite differences. In addition, the dynamic adjustment mechanism of the present invention can optimize the selection of time step and diffusion coefficient, avoid too many invalid iterations, and thus significantly improve the processing efficiency. In practical applications, this method can achieve high-precision image optimization with low computing resource consumption, meeting real-time requirements.

[0037] 5. By introducing detail enhancement items into the optimization model and combining the anisotropic diffusion equation for edge enhancement processing, the present invention can significantly improve the contrast of image edges and details without amplifying noise. The design of the edge control function ensures that the details of high-gradient areas can be retained first, which is particularly suitable for target recognition and boundary detection tasks in infrared images.

[0038] 6. The image processing method of the present invention is not only applicable to single-band infrared images, but can also be extended to the joint processing of multi-band infrared images. In industrial detection, the present invention can highlight the edge features of the target area; in night vision monitoring, it can effectively enhance key targets under low light conditions; in medical imaging, it can highlight the lesion area in the infrared thermal image. The versatility of the present invention makes it widely adaptable in a variety of practical application scenarios.

[0039] 7. The present invention can automatically extract the noise and gradient characteristics of the image and realize dynamic adjustment of parameters by combining user interactive input, which significantly reduces the dependence on manual debugging and empirical parameters. The system can automatically initialize the model parameters according to the characteristics of the input image, and adaptively adjust the weights and step sizes during the iteration process, which greatly reduces the complexity and labor cost of the image processing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1Schematic diagram of the overall steps of the method of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] Example:

[0043] Please refer to the attached Figure 1 The present invention proposes an infrared image interactive processing method based on the calculus of variations and partial differential equation (PDE) constraints. Through a multi-task optimization framework and a user interaction mechanism, a dynamic balance between detail enhancement and noise suppression of infrared images is achieved, while ensuring that the results meet the needs of different users. The method includes the following steps:

[0044] Step 1: Input infrared image and preprocess it:

[0045] Step 1 is the basic link for infrared image processing in the present invention, which aims to obtain the infrared image to be processed and perform sufficient preprocessing operations to extract noise characteristics, edge characteristics and scene characteristics, and provide necessary input for the construction of subsequent optimization models and parameter initialization. Through a scientific and reasonable preprocessing process, the efficiency and accuracy of the optimization model can be improved, and data support can be provided for subsequent user interaction and dynamic adjustment.

[0046] The preprocessing method in the present invention not only targets the basic processing requirements of traditional infrared images, but also combines the characteristics of infrared images, including low signal-to-noise ratio, high dynamic range and complex background, to design a targeted processing flow to ensure that the extracted characteristics have physical meaning and are suitable for subsequent processing steps.

[0047] In this embodiment, the acquisition of infrared images and the extraction of noise characteristics include the following:

[0048] Infrared images can be acquired directly through infrared sensors or by reading existing infrared image data through storage devices. The input image is defined as .

[0049] As an option, if the sensor for collecting images supports multi-spectral or multi-band information, the present invention can also process multi-band infrared images simultaneously to further optimize the image feature extraction effect.

[0050] After the image is input, the noise characteristics are first extracted. Specifically, the image noise characteristics are extracted by analyzing the intensity and type of noise distribution in the infrared image. It should be noted that the noise of infrared images mainly includes random noise and thermal noise. Random noise usually appears as a Gaussian distribution with a mean of zero, while thermal noise usually appears as a low-frequency distribution.

[0051] In one possible implementation, the image noise intensity is estimated using a local statistical analysis method, for example, by calculating the local variance or standard deviation of the image to quantify the noise level. Suppose a local window of the image is , the noise intensity estimation expression is:

[0052]

[0053] in, represents the mean value of the pixels in the window, is the window size.

[0054] It can be understood that the above noise intensity estimation results will be directly used to initialize the diffusion parameters of the heat diffusion equation As an option, when the image noise intensity is high, the diffusion parameter can be increased to improve the denoising effect.

[0055] In this embodiment, edge feature extraction of infrared images includes the following contents:

[0056] The core of edge feature extraction is to calculate the gradient amplitude of the image , which is used to characterize the local rate of change of gray values ​​in an image, thereby identifying the edge area in the image.

[0057] Specifically, the expression of the gradient amplitude is:

[0058]

[0059] in, and Respectively represent the image along and The gradient in the direction. It can be discretized and calculated using the finite difference method:

[0060]

[0061] As an option, if the edge characteristics are more complex, a multi-scale edge extraction method can be used to smooth the image by setting Gaussian filters of different scales and then calculating its gradient. The expression of multi-scale edge extraction is:

[0062]

[0063] in, The standard deviation is The Gaussian kernel of Represents a convolution operation.

[0064] It should be noted that the result of edge feature extraction is directly used to initialize the control function of the edge preservation equation For example, if the gradient amplitude in the edge region is high, the suppression effect of the edge preservation function will be significantly enhanced, thereby preventing the edge details from being over-smoothed.

[0065] In this embodiment, the scene characteristic analysis of the infrared image includes the following contents:

[0066] The goal of scene characteristic analysis is to evaluate the global characteristics of the image, such as brightness distribution and contrast, to provide a global reference for initialization of the optimization model.

[0067] Specifically, scene characteristic analysis characterizes brightness and contrast by calculating the global mean and standard deviation of the image. The calculation formulas for the global mean and standard deviation are:

[0068] in, Represents the total number of pixels in the image.

[0069] As an option, when the global contrast of the image is low, the weight parameter of the gradient enhancement term in the optimization model can be appropriately increased. , to enhance the edges and details of the image.

[0070] As an extension, the image preprocessing process of the present invention can also be combined with an adaptive histogram equalization method to further improve the contrast of the infrared image. For example, the brightness distribution can be adjusted by dividing the local window and independently calculating the histogram of each window, thereby improving the visibility of details.

[0071] It is understandable that the preprocessing process of infrared images is not limited to the above implementation methods, and can also be adjusted according to the needs of specific application scenarios. For example, in medical imaging, noise suppression can give priority to a stronger filtering method, while in industrial detection, edge feature extraction can be combined with morphological operations to further optimize the results. The technical content disclosed in the above specific implementation methods provides sufficient support for those skilled in the art to reproduce the present invention.

[0072] Step 2: Construct energy functional model:

[0073] Definition of energy functional:

[0074] The energy functional model describes the optimization objectives of image denoising, detail enhancement and smoothing, and its expression is:

[0075]

[0076] in: : Energy functional is the objective function of infrared image optimization. The optimization goal is to minimize the energy.

[0077] : The optimized image (i.e. the processed infrared image) is the variable optimized by the energy functional;

[0078] : The domain of the image, which represents the two-dimensional pixel space of the image;

[0079] : Weight parameters, corresponding to the contribution weights of gradient enhancement term, smoothing denoising term and fidelity term respectively;

[0080] : Gradient enhancement item, used to describe the detailed information of the edge of the image;

[0081] : The magnitude of the image gradient, which represents the rate of change of the pixel value;

[0082] : Exponential parameter, controlling the nonlinearity of gradient enhancement, usually takes the value ;

[0083] : Smoothing denoising term, used to describe the high-frequency noise of the image;

[0084] : The Laplacian operator of the image, representing the second-order change of pixel values;

[0085] : Fidelity item, used to ensure the optimized image and the input image Global consistency;

[0086] : Input original infrared image;

[0087] : User feedback item, used to dynamically adjust the optimization model based on user input;

[0088] : pixel position coordinates in the image;

[0089] : Time step, used for dynamic evolution during numerical iteration;

[0090] : User feedback control parameter, used to adjust the weight of the optimization objective.

[0091] It should be noted that:

[0092] Item 1 It is a gradient enhancement item, which is used to enhance the edge details of the image;

[0093] gradient Indicates the rate of change of image pixel values, which can be expressed as:

[0094]

[0095] in:

[0096] : Gradient amplitude, which indicates the magnitude of the change in pixel value in the image;

[0097] : current image;

[0098] : The gradient of the image pixel value along the horizontal (x) direction;

[0099] : The gradient of the image pixel value along the vertical (y) direction.

[0100] index Used to adjust the nonlinearity of gradient enhancement;

[0101] As an option, The value of can be adjusted according to the user's feedback on detail enhancement needs. The higher the detail enhancement needs, The larger the value of .

[0102] Item 2: is a smoothing term used to suppress high-frequency noise in the image;

[0103] in, The Laplacian operator representing the image is expressed as:

[0104]

[0105] in, : The Laplacian operator of the image, representing the second-order change of pixel values;

[0106] : The second-order partial derivative of the image pixel value along the horizontal (x) direction, indicating the change in the rate of change in the horizontal direction;

[0107] : The second-order partial derivative of the image pixel value along the vertical (y) direction, which represents the change in the rate of change in the vertical direction.

[0108] It should be noted that this item eliminates high-frequency noise in the image by smoothing the changes in pixel values;

[0109] For example, The value of can be dynamically adjusted according to the estimated result of the noise intensity in the image. The higher the noise intensity, The larger the value of

[0110] Item 3: It is a fidelity term used to limit the global consistency between the output image and the input image;

[0111] Specifically, this term is used to prevent the optimization process from over-modifying image details;

[0112] The value of is usually initialized according to the scene complexity of the image. The higher the complexity, The smaller the value of .

[0113] It needs to be further explained that , and are task weight parameters, which can be initialized by preset values ​​or adjusted in real time based on user feedback.

[0114] User feedback items It is introduced into the energy functional and is specifically used to dynamically control the enhancement and denoising strength of the image.

[0115] User interaction controls:

[0116] The user inputs the enhancement strength, denoising strength and the annotation of the area of ​​interest through the interactive interface.

[0117] The user feedback control is defined as:

[0118]

[0119] in, A control parameter entered by the user to describe the strength of enhancement or denoising;

[0120] The weight distribution function of the region of interest marked by the user indicates the processing priority of the user-specified region.

[0121] Alternatively, the user can use a slider to adjust the The value of , thereby dynamically changing the intensity of enhancement or denoising; at the same time, the user draws the area of ​​interest through the annotation tool to generate the weight function .

[0122] In another implementation, user feedback can be combined with scene characteristics of the input image to automatically generate a default initial value, such as giving priority to increasing the denoising strength in scenes with high noise.

[0123] In this embodiment, the dynamic adjustment mechanism of the optimization model includes the following contents:

[0124] The weight parameters in the optimization model will be dynamically adjusted during the iteration process based on user feedback and image characteristics to ensure that the final result meets user needs.

[0125] Specifically:

[0126] When users have a high demand for detail enhancement, the system will give priority to increasing The value of to increase the weight of the gradient enhancement term;

[0127] When the user has higher requirements for denoising strength, the system will give priority to increasing To improve the contribution of the smoothing term;

[0128] When the user marks the area of ​​interest, The weight value of will be significantly increased to give priority to enhancing or denoising this area.

[0129] As an extension, the optimization model of the present invention can also be optimized in combination with multi-band information of infrared images. In some embodiments, a joint optimization model can be constructed based on the input multi-band infrared image to further improve the image processing effect by fusing the characteristics of multiple bands.

[0130] It should be noted that the initialization and dynamic adjustment mechanism of the weight parameters in the optimization model can be customized according to the specific application scenario. For example, in industrial detection, The initial value of can be set higher to ensure that key edges are highlighted. In night vision monitoring, The initial value of can be higher to prioritize noise removal.

[0131] It can be understood that the optimization model construction method of the present invention has high flexibility and scalability, and its specific implementation method can be adapted according to different user needs and application scenarios, and the implementation method given in the above embodiment does not constitute a limitation on the scope of protection of the present invention.

[0132] Step 3: Introduce partial differential equation constraints:

[0133] The core of this step is to construct a dynamically controlled image processing mechanism by introducing partial differential equation (PDE) constraints, thereby achieving denoising and smoothing of infrared images while retaining edge characteristics in detail areas. Through the introduction of PDE, combined with image gradient amplitude and user interaction feedback, differentiated processing of different areas of the image can be achieved. This step directly serves the optimization solution of the energy functional model and provides the necessary physical and mathematical constraints for subsequent iterations.

[0134] It can be understood that the introduction of PDE can not only significantly improve the robustness of image processing, but also ensure that the key details of the infrared image are not weakened by the careful design of diffusion parameters and edge enhancement control functions. It should be noted that the PDE used in this embodiment includes the heat diffusion equation and the anisotropic diffusion equation, which play an important role in denoising and smoothing, and detail enhancement, respectively.

[0135] Heat diffusion constraints:

[0136] The heat diffusion equation is used to describe the denoising and smoothing process of infrared images. By simulating the heat diffusion phenomenon, the low-frequency areas with large noise in the image can be effectively smoothed. The heat diffusion equation is expressed as:

[0137]

[0138] in, is the time evolution term of the image, which indicates the rate of change of the image during the iteration process;

[0139] represents the gradient operator;

[0140] is the diffusion coefficient, which indicates the diffusion intensity of different image regions;

[0141] : The gradient of an image, used to describe the direction of change of pixel values.

[0142] Alternatively, the diffusion coefficient The value of is dynamically adjusted according to the gradient amplitude of the image. Specifically, the diffusion coefficient is expressed as:

[0143]

[0144] in, : Diffusion coefficient, used to control the denoising strength of the image;

[0145] : pixel position coordinates in the image;

[0146] : represents the gradient amplitude of the image, which is used to reflect the rate of change of pixel values ​​in the image;

[0147] is a control parameter used to adjust the sensitivity of diffusion intensity to gradient amplitude.

[0148] : Nonlinear adjustment item, used to reduce the diffusion intensity in high gradient areas, thereby protecting the edge of the image.

[0149] It should be noted that the gradient amplitude The calculation formula is:

[0150]

[0151] Through the above design, in the edge area with large gradient amplitude, the diffusion coefficient The value of is small, which effectively avoids excessive smoothing of the edge; in the low-frequency area with small gradient amplitude, the diffusion coefficient The larger the value of , the better the noise smoothing ability.

[0152] In a possible implementation, the diffusion coefficient can also be dynamically adjusted in combination with user interaction input. For example, when the user chooses to enhance the denoising effect, the diffusion coefficient can be increased. value, thereby accelerating the diffusion process.

[0153] Edge Preserving Constraints:

[0154] The edge preservation equation is used to enhance image details by adaptively adjusting the gradient amplitude to achieve priority processing in detail areas. Specifically, the edge preservation equation is expressed as:

[0155]

[0156] in, It is an edge control function, which is used to adjust the processing intensity of different gradient areas of the image;

[0157] : current image;

[0158] : time step, used for numerical iteration;

[0159] Exemplary edge control function The expression is:

[0160]

[0161] in, Represents the gradient magnitude of the image; It is the edge sensitivity parameter, which is used to control the degree of enhancement of edge areas.

[0162] It should be noted that the edge control function The design can significantly weaken the diffusion intensity in the edge area (where the gradient amplitude is large), thereby protecting the edge details of the image; in the non-edge area (where the gradient amplitude is small), a higher intensity diffusion is allowed to suppress noise.

[0163] In one possible implementation, the edge sensitivity parameter It can be adjusted dynamically based on user input. For example, when the user has a high demand for detail enhancement, the to further weaken the diffusion in the edge area.

[0164] In order to perform differential processing on the image in combination with user feedback, the present embodiment also generates a feedback term through user input and introduces it into the partial differential equation. Specifically, the expression form of the user feedback term is:

[0165]

[0166] in, Indicates the enhancement or denoising strength of the user input, The weight distribution function of the region of interest marked by the user. It should be noted that the user feedback item directly acts on the evolution process of the image, so that the region of interest can be enhanced or denoised first.

[0167] For example, in some embodiments, when a user marks an area of ​​interest and chooses to increase detail enhancement, the feedback item The value of will increase significantly, thus exerting stronger protection on the gradient amplitude in this area and preventing the details from being weakened by the diffusion process.

[0168] As an extension, the partial differential equation introduced in this embodiment is not only applicable to the processing of single-band infrared images, but also can be applied to the joint processing of multi-band infrared images. In some embodiments, diffusion parameters and edge control functions can be constructed for different regions of the multi-band image, so as to optimize the information characteristics of different bands.

[0169] It can be understood that the introduction of partial differential equations in the present invention provides dynamic adjustment capabilities for the optimization model, and its specific implementation method can be adjusted according to the specific application scenario, and the implementation method given in the above embodiment does not constitute a limitation on the scope of protection of the present invention.

[0170] Step 4: Numerical solution:

[0171] Numerical solution is an important step in realizing infrared image optimization processing of the present invention. In the present invention, the minimization of energy functional and the dynamic evolution of partial differential equation constraints are realized by numerical iteration method to generate high-quality image processing results. Specifically, the continuous mathematical optimization problem is discretized into a computable numerical form by using an iterative method with an explicit time step, thereby realizing the solution of the model. It should be noted that the core of numerical iteration is to ensure convergence stability, while dynamically responding to weight adjustments in the optimization model and user feedback.

[0172] The specific process of numerical solution includes the following:

[0173] The numerical iterative solution of the optimization model is based on the explicit time step method, which discretizes the continuous partial differential equation and constructs the update formula. The update formula of the discretized time step is:

[0174]

[0175] in: Represents the image of the current iteration step;

[0176] Represents the image of the next iteration step;

[0177] is the time step, which is used to control the update amplitude during the iteration process;

[0178] It is the evolution item of the current time step, including denoising, detail enhancement and user feedback.

[0179] As an option, The value of can be adjusted according to the image characteristics and optimization stability. A larger time step can speed up the iteration but may cause numerical oscillation; a smaller time step can improve stability but reduce computational efficiency.

[0180] In this embodiment, the specific method of numerical discretization includes the following:

[0181] For the gradient term in the partial differential equation , the finite difference method is used for spatial discretization, and the expression is:

[0182]

[0183] in: and Represents the space interval in the horizontal and vertical directions respectively;

[0184] Indicates the image The pixel value of a pixel.

[0185] For the Laplace operator , discretized using the central difference method, the expression is:

[0186]

[0187] It should be noted that these discretization methods can effectively balance computational complexity and numerical accuracy, making the solution of partial differential equations highly adaptable.

[0188] In this embodiment, the specific construction of the numerical evolution term includes the following contents:

[0189] Numerical evolution term It includes the following three parts:

[0190] Denoising evolution term:

[0191]

[0192] Among them, the diffusion coefficient Dynamically adjusted according to the gradient amplitude, it is used to smooth low-frequency areas while protecting high-frequency edges.

[0193] Detail Enhancement:

[0194]

[0195] Among them, the edge control function It is used to limit the diffusion intensity of the edge area, thereby enhancing the image details. The specific form is:

[0196]

[0197] in, is the edge enhancement sensitivity parameter

[0198] User feedback items:

[0199]

[0200] User feedback items are controlled by user input parameters and weight function Realize the priority optimization processing of the area of ​​interest.

[0201] Finally, the complete numerical evolution formula is:

[0202]

[0203] In this embodiment, the convergence judgment conditions of the numerical solution include the following:

[0204] In order to ensure the stability and accuracy of the iterative solution, the present invention sets the convergence condition, and the judgment basis is:

[0205]

[0206] in:

[0207] Indicates the change between the current iteration image and the previous iteration image;

[0208] It is a preset convergence threshold used to control the convergence accuracy of the processing results.

[0209] It can be understood that when the image change amount is less than the threshold, it means that the optimization model has reached a stable state and the iteration process can be terminated.

[0210] Dynamic adjustment mechanism:

[0211] In the numerical solution of the present invention, the time step is , diffusion coefficient , edge control function Parameters such as image quality and image quality can be dynamically adjusted based on image characteristics and user feedback. For example:

[0212] When the image noise is large, the diffusion coefficient The value of will increase to speed up the smoothing process of low-frequency areas;

[0213] When users have higher requirements for detail enhancement, edge control parameters The value of will decrease, thereby enhancing the protection of edge details;

[0214] Time step A smaller initial value can be selected according to the scenario requirements to ensure the stability of the numerical solution.

[0215] Step 5: Output processing results:

[0216] First of all, it should be noted that the main function of step 5 is to constrain the processing of infrared images through partial differential equations (PDEs), so that the image can maintain edge and detail characteristics while denoising and smoothing. By dynamically adjusting the diffusion parameters and edge preservation functions, step 5 achieves adaptive optimization processing for different regions. This step provides a theoretical basis for subsequent numerical solutions, and at the same time models and constrains the physical characteristics of the image to ensure that the optimization results meet the actual application requirements.

[0217] It should be noted that the introduction of PDE not only helps to constrain the physical consistency of the image, but also can dynamically adjust parameters according to the characteristics of the input image, so that the processing results are more in line with the needs of specific application scenarios.

[0218] In this embodiment, the construction of partial differential equation constraints includes the following contents:

[0219] The constraint forms of partial differential equations mainly include heat diffusion equation and edge preservation equation, which are responsible for denoising and smoothing as well as protection of details and edges respectively.

[0220] Specifically, the mathematical expression of the heat diffusion equation is:

[0221]

[0222] in, represents the gradient operator of the image, is the diffusion coefficient.

[0223] It should be noted that the diffusion coefficient It is dynamically adjusted, mainly based on the gradient amplitude of the input image Calculate. The expression of the diffusion coefficient is:

[0224]

[0225] in, It is a control parameter used to adjust the response of the diffusion coefficient to the gradient amplitude. It can be understood that the area with a large gradient amplitude usually corresponds to the edge of the image. In order to avoid excessive smoothing of the edge, the diffusion coefficient takes a smaller value in these areas. The area with a small gradient amplitude usually corresponds to a noisy or flat area. The diffusion coefficient takes a larger value in these areas to enhance the smoothing effect of the noise.

[0226] As an option, when applied to infrared images of complex scenes, the To control the diffusion ratio of different areas, you can use Set the value of to a smaller value to enhance the diffusion effect in low gradient areas.

[0227] In the edge preservation process, the present invention uses an anisotropic diffusion model, which is mathematically expressed as:

[0228]

[0229] in, is the edge suppression function, which is used to control the diffusion intensity of the edge area. Specifically, the expression of the edge suppression function is:

[0230]

[0231] in, It is a control parameter used to adjust the sensitivity of the edge area.

[0232] In a possible implementation, the edge suppression function can dynamically adjust the diffusion intensity according to the magnitude of the gradient amplitude. Regions with larger gradient amplitudes correspond to the edge parts of the image, and the edge suppression function takes smaller values ​​in these regions, thereby effectively limiting the impact of the diffusion process on the edges and avoiding the loss of edge details. Regions with smaller gradient amplitudes correspond to non-edge regions, and the edge suppression function takes larger values ​​in these regions to achieve smoothing of non-edge regions.

[0233] It should be noted that The value of has a direct impact on the effect of edge preservation. For example, in application scenarios where edge clarity needs to be enhanced, the value of to increase the sensitivity of the edge suppression function to high gradient areas.

[0234] In this embodiment, the combination and expansion of the partial differential equation includes the following contents:

[0235] The heat diffusion equation and the edge preservation equation can be combined in a unified framework to achieve comprehensive processing of denoising, smoothing and edge preservation. The unified expression is:

[0236]

[0237] in, and represent the diffusion coefficient and edge suppression function respectively.

[0238] In a possible implementation, the partial differential equation can also be combined with user feedback information to dynamically adjust the parameters of the diffusion process. The user feedback term can be expressed as:

[0239]

[0240] in, is the intensity control parameter entered by the user, The weight distribution function of the region of interest marked by the user. By adding user feedback to the partial differential equation, it is possible to prioritize specific areas. For example, in industrial inspection, users can mark cracks or defect areas and adjust The value of is used to give priority to enhancing these areas.

[0241] It can be understood that the parameters in the partial differential equation can not only be automatically calculated according to the image characteristics, but also can be adjusted through user interaction, thereby further enhancing the flexibility and adaptability of the processing.

[0242] Extended content:

[0243] As an extension, the partial differential equation constraint of the present invention can also be combined with multi-scale analysis technology to meet the processing requirements of infrared images with different resolutions. For example, by introducing Gaussian filters of different scales to perform multi-scale decomposition of the image, different partial differential equation constraints can be applied to the low-frequency and high-frequency components respectively, thereby achieving a more refined processing effect.

[0244] It should be noted that the specific implementation of partial differential equations can be customized according to different application scenarios. For example, in medical imaging, the weight of the edge preservation function can be enhanced first to highlight the boundary of the lesion area. In night vision monitoring, the diffusion effect of the heat diffusion equation can be enhanced first to remove noise under low light conditions.

[0245] The above specific implementation methods provide sufficient support for those skilled in the art to reproduce the present invention, and also clarify the innovative features and practical application value of the present invention.

[0246] After the infrared image is output, the image noise is effectively suppressed and the edge details are enhanced.

[0247] If you are not satisfied with the results, you can adjust the parameters through the interactive interface. The system will dynamically update the optimization model and repeat the solution process until the user's needs are met.

[0248] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for interactive processing of infrared images, characterized in that: The following steps are involved: S1. Obtaining an infrared image to be processed; S2. Preprocessing the infrared image to extract noise characteristics, edge characteristics and scene characteristics; S3. receiving feedback information input by the user through the user interaction interface, including the definition of enhancement strength, denoising strength and region of interest; S4. constructing an optimization model based on a variational method, wherein the optimization model guides image processing by describing optimization objectives of denoising, detail enhancement, and image smoothing; S5. Introducing a partial differential equation constraint, wherein the partial differential equation is used to control the dynamic balance between denoising and edge preservation during image processing; S6. solving the optimization model by a numerical iteration method and optimizing the infrared image until a preset convergence condition is met; S7. Output the processed infrared image, and dynamically update the optimization parameters based on user feedback, and repeat steps S4 to S6 until a final result that meets user needs is generated; The optimization model is expressed in the form of an energy functional, which includes a detail enhancement term describing image edge detail enhancement, a denoising term describing noise suppression, a global fidelity term for maintaining original image information, and an interactive control term combining user feedback information.

2. The interactive processing method for infrared images according to claim 1, characterized in that: The feedback information input by the user in step S3 is used to adjust the weight parameters of each item in the optimization model. The weight parameters are dynamically changed according to user needs to balance the processing effects of detail enhancement, denoising and global fidelity.

3. The interactive processing method for infrared images according to claim 1, characterized in that: The partial differential equation constraints include: De-noising equations that characterize the heat diffusion process are used to smooth noise and avoid over-processing the edges; The edge-preserving equation that characterizes anisotropic diffusion is used to avoid amplification of high-frequency noise during detail enhancement.

4. The interactive processing method for infrared images according to claim 3, characterized in that: The diffusion parameter in the partial differential equation is dynamically adjusted according to the gradient amplitude of the infrared image, wherein the diffusion degree of the area with higher noise is larger, while the diffusion degree of the edge area is smaller.

5. The interactive processing method for infrared images according to claim 1, characterized in that: The numerical iteration method adopts an explicit time step method to solve the problem, and gradually optimizes the image processing results in discrete time steps until a preset convergence condition is reached.

6. The interactive processing method for infrared images according to claim 5, characterized in that: The spatial derivative calculation in the numerical iteration method is discretized by using the finite difference method, the gradient calculation is used to enhance edge information, and the Laplace operator calculation is used to remove noise and smooth the image.

7. The interactive processing method for infrared images according to claim 1, characterized in that: The user interaction interface comprises: Slider controls to adjust image enhancement and denoising strength; The region annotation tool is used to define the user's area of ​​interest so that image processing can be prioritized within that area.

8. The interactive processing method for infrared images according to claim 1, characterized in that: The method initializes weight parameters in the optimization model through scenario analysis, including: Set the weight of the denoising term according to the noise intensity; Setting the weight of detail enhancement items according to image edge characteristics; Adjust the weight of the global fidelity term based on scene complexity.

9. The interactive processing method for infrared images according to claim 1, characterized in that: In the numerical iteration process of step S6, the iteration speed is optimized according to the change range of the current iteration result by dynamically adjusting the time step, wherein: When the change amplitude of the current image processing result is greater than the set threshold, the time step is reduced to improve the convergence stability; When the change amplitude of the current image processing result is less than the set threshold, the time step is increased to accelerate the iterative convergence.

Images