Intelligent noise removal and image optimization method for camera module

Abstract

The invention discloses an intelligent noise removal and image optimization method for a camera module. The intelligent noise removal and image optimization method comprises the following steps of S1, collecting an input image and performing preprocessing; s2, performing global search by adopting a differential evolution algorithm to determine a deep noise reduction network hyper-parameter; s3, performing noise suppression processing on the preprocessed image by using a deep noise reduction network; s4, inputting the image subjected to noise suppression into an image optimization enhancement module, and searching an optimal image enhancement parameter through a fire bird optimization algorithm in a multi-target manner; s5, performing image enhancement processing on the noise suppression image according to the optimal image enhancement parameter; s6, performing quality evaluation on the enhanced image by adopting a comprehensive joint loss function; and S7, according to a quality evaluation result, adjusting parameters of a fire bird optimization algorithm, and outputting a final image. According to the method, the deep neural network and the swarm intelligent optimization algorithm are fused, adaptive removal of the image noise of the camera module and intelligent optimization of the image quality are realized, and the definition and detail retention of the real-time monitoring image are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to an intelligent noise removal and image optimization method for a camera module. Background Art

[0002] With the rapid development of digital image processing technology and artificial intelligence technology, camera modules have been widely used in many fields such as smartphones, surveillance systems, and vehicle-mounted systems. However, due to the acquisition environment, hardware limitations, and limitations of imaging algorithms, camera modules are prone to generate noise in low-light, complex background, or high-speed motion scenarios, resulting in problems such as reduced image quality, loss of details, and color distortion. Existing image noise removal methods mainly rely on traditional filtering techniques such as mean filtering, median filtering, and Gaussian filtering. These methods have certain effects in simple noise situations, but in the face of complex noise and a mixture of multiple noises, they often have defects such as over-smoothing, loss of details, and blurred edges. In addition, although some techniques based on wavelet transform and adaptive filtering have improved the noise removal effect to a certain extent, their computational complexity is relatively high, and it is difficult to process high-speed image data in real time, which limits their application in real-time image acquisition systems.

[0003] In recent years, with the booming development of deep learning technology, more and more research has started to use deep neural networks for image noise removal. The denoising network constructed using a convolutional neural network (CNN) can automatically learn the noise characteristics in the image and suppress the noise through multi-layer non-linear mapping, thereby retaining the detail information in the image to a certain extent. However, most of the existing deep denoising networks rely on fixed structures and preset hyperparameters, which makes them show certain limitations when facing different types of noise and dynamic environments. Since camera modules often need to adapt to various imaging scenarios in practical applications, the denoising network with fixed parameters cannot achieve an ideal denoising effect in all scenarios. Therefore, an intelligent denoising method that can dynamically adjust network parameters according to the actual noise characteristics of the input image is needed.

[0004] In addition to the noise removal problem, image optimization and enhancement are also an important research direction in current image processing technology. Traditional image enhancement methods mainly focus on adjusting the brightness, contrast, saturation, and sharpening of the image, etc. However, these methods usually adopt fixed enhancement strategies and cannot be adjusted individually according to the specific situation of different images. Existing enhancement methods may cause loss of image details, color distortion, or over-enhancement phenomena in some cases, and cannot take into account both the visual quality and detail retention of the image. Therefore, how to design a method that can adaptively adjust enhancement parameters according to the image content and achieve detail restoration and overall visual effect optimization on the basis of ensuring the image denoising effect has become a technical problem that needs to be solved urgently.

[0005] In recent years, in response to the above problems, some studies have attempted to introduce swarm intelligence optimization algorithms into the field of image processing to achieve dynamic adjustment of network hyperparameters and multi-objective search for image enhancement parameters. The Differential Evolution (DE) algorithm has been widely used in the field of parameter optimization due to its global search ability and simple and efficient calculation method. At the same time, some new swarm intelligence algorithms, such as the Flamingo Optimization Algorithm (FOA), simulate the collaborative search behavior of organisms in nature and show certain advantages in solving multi-objective optimization problems. However, the existing Flamingo Optimization Algorithm mainly focuses on local search and information sharing, fails to fully integrate the global mutation strategy, and lacks an adaptive mechanism in parameter adjustment, resulting in insufficient search efficiency and global convergence in the high-dimensional parameter space. On the other hand, although the Differential Evolution algorithm has a strong global search ability, it is prone to search stagnation when dealing with local refinement and complex constraint problems. Therefore, it is often difficult to meet the requirements of algorithm adaptability and fine adjustment in both image denoising and image enhancement by using only one of these algorithms alone.

[0006] Therefore, how to provide an intelligent noise removal and image optimization method for camera modules is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0007] An object of the present invention is to propose an intelligent noise removal and image optimization method for camera modules. The present invention makes full use of deep vision processing technology, the Differential Evolution algorithm, and an improved Flamingo Optimization Algorithm, and details how to achieve adaptive suppression of image noise and intelligent optimization of image quality through intelligent algorithms. By preprocessing the images collected by the camera, noise modeling, and global optimization of the denoising network structure, the present invention can effectively reduce the noise interference in the images while retaining the detail information such as edges and textures of the images. Further, the present invention adopts an image optimization and enhancement module to perform detail restoration, color correction, contrast adjustment, and sharpening on the denoised images, uses the improved Flamingo Optimization Algorithm to perform multi-objective search for image enhancement parameters, and introduces a Differential Evolution mechanism and an adaptive parameter adjustment strategy, so as to achieve a good balance between global search and local refinement. This method has the advantages of strong adaptability, good real-time performance, strong global search ability, and excellent detail retention effect, and can effectively improve the image quality in complex imaging environments and meet the requirements for image clarity and authenticity in high-demand application scenarios.

[0008] An intelligent noise removal and image optimization method for camera modules according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect the input image and preprocess the input image;

[0010] S2. Use the differential evolution algorithm to globally search for the hyperparameters of the deep denoising network and determine the deep denoising network structure;

[0011] S3. Use the deep denoising network to perform noise suppression on the preprocessed input image, reduce the noise interference in the input image and retain the details of the input image;

[0012] S4. Input the input image after noise suppression processing into the image optimization and enhancement module, and perform multi-objective search on the image enhancement parameters through the improved flamingo optimization algorithm to obtain the optimal image enhancement parameters;

[0013] S5. Perform image enhancement processing on the input image after noise suppression processing according to the optimal image enhancement parameters;

[0014] S6. Use a comprehensive joint loss function to evaluate the quality of the enhanced input image. The joint loss function includes reconstruction error, perceptual error, adversarial error, and parameter regularization error, and measures the overall quality of the enhanced input image;

[0015] S7. According to the feedback quality evaluation results, use the differential evolution algorithm to iteratively adjust the initial parameters of the flamingo optimization algorithm, and finally output the image after noise removal and image optimization processing.

[0016] Optionally, the collecting of the input image includes the original image data, the color information of the image, the illumination information of the image, the spatial resolution of the image, and the noise in the image, which is used to provide image data for noise suppression, image optimization, and enhancement processing.

[0017] Optionally, the image preprocessing includes image denoising, image normalization, color space conversion, and size adjustment, which are used to reduce image noise, improve image quality, and unify the image data format.

[0018] Optionally, the S2 specifically includes:

[0019] S21. Construct a candidate solution vector P;

[0020] S22. Initialize the candidate solution set:

[0021] {P i}, i = 1, 2, …, N;

[0022] where N is the total number of candidate solutions, and each candidate solution P i is a set of hyperparameter configurations;

[0023] S23. Calculate the fitness function value F(P i for each candidate solution Pi ):

[0024] F(P i )=α·E recon (P i )+β·E reg (P i )+γ·E struct (P i );

[0025] where α, β, and γ are weight coefficients, and E recon (P i ) represents the reconstruction error, E reg (P i ) represents the regularization error, and E struct (P i ) represents the network structure consistency error;

[0026] S24. For each candidate solution P i Randomly select three different candidate solutions P r1 , P r2 and P r3 , and calculate the mutation vector:

[0027]

[0028] where F0 is the initial scaling factor, F min is the minimum scaling factor, t is the current iteration number, and T is the preset maximum iteration number;

[0029] S25. Perform a crossover operation on the candidate solution P i and the corresponding mutation vector M i to generate a trial vector T i , where the j-th component satisfies:

[0030]

[0031] where CR is the adaptive crossover rate, rand(j) is a value randomly selected from the interval [0,1], T i,j represents the j-th component of the generated trial vector T i , M i,j represents the j-th component of the mutation vector M i , and P i,j represents the j-th component of the original candidate solution P i ;

[0032] S26. Calculate the fitness function value F(T i ) of the trial vector T i ). If F(T i ) < F(P i), then set P i = T i , otherwise retain the original candidate solution;

[0033] S27. Repeat steps S23 to S26 until the preset number of iterations or the fitness convergence condition is reached, and select the candidate solution P with the optimal fitness best = [K * , L * , C * , A * , as the hyperparameter configuration of the deep denoising network, and construct the deep denoising network according to the candidate solution with the optimal fitness. The deep denoising network consists of L * sequentially connected convolutional layers, where each convolutional layer uses a convolutional kernel size of K * , a number of channels C * and activation function parameters A * for configuration.

[0034] Optionally, the candidate solution vector includes the convolutional kernel size, the number of network layers, the number of channels in the convolutional layer, and the activation function parameters, which are used to optimize the structure and performance of the deep denoising network.

[0035] Optionally, the specific steps of S3 are as follows:

[0036] S31. Receive the preprocessed input image, and the input image data format is consistent with the input requirements of the deep denoising network;

[0037] S32. Load the deep denoising network structure constructed with the optimal hyperparameter configuration determined by the differential evolution algorithm. The deep denoising network structure includes multiple sequentially connected convolutional layers, activation layers, and normalization layers;

[0038] S33. Feed the preprocessed input image into the loaded deep denoising network, and process the input image sequentially through the forward propagation of each layer in the deep denoising network;

[0039] S34. During the forward propagation process, through the layer-by-layer action of the convolutional operation and the activation function, the deep denoising network suppresses the noise in the input image while retaining the edge and texture detail information of the input image;

[0040] S35. Obtain the input image result after noise suppression processing at the end of the deep denoising network, and conduct a preliminary evaluation on the input image result to confirm that the noise interference is significantly reduced and the detail information is effectively retained;

[0041] S36. Output the input image after noise suppression processing, and transfer the input image result to the image optimization and enhancement module.

[0042] Optionally, the specific steps of S4 are as follows:

[0043] S41. Receive the input image after noise suppression processing as described in claim 3, denoted as I denoise ;

[0044] S42. Apply the image optimization and enhancement module to perform image enhancement processing. The image optimization and enhancement module includes a parameter optimization sub-module and an image enhancement execution sub-module. The parameter optimization sub-module is used to search for image enhancement parameters through an optimization algorithm, and the image enhancement execution sub-module is used to perform enhancement processing on the image based on the image enhancement parameters;

[0045] S43. Construct an image enhancement parameter candidate solution vector Q;

[0046] S44. Initialize the image enhancement parameter candidate solution set {Q i}, where i = 1, 2,..., M, and M is the number of candidate solutions;

[0047] S45. For each candidate solution Q i calculate the multi-objective fitness function:

[0048] F e (Q i ) = λ·E detail (Q i ) + μ·E color (Q i ) + ν·E contrast (Q i );

[0049] where λ, μ, and ν are the weight coefficients for detail restoration, color restoration, and contrast enhancement respectively, and E detail (Q i ), E color (Q i ), and E contrast (Q i ) are the error evaluation functions of the image enhanced based on the candidate solution Q i in the corresponding dimensions;

[0050] S46. Select the candidate solution with the minimum current fitness function value F e (Q i ) in the population, denoted as Q best ;

[0051] S47. During the process of updating the image enhancement parameters, for the candidate solution Q i , introduce a differential vector term Q r2 - Q r3 , where Q r2 and Q r3 are two different candidate solutions randomly selected from the population, and calculate the updated candidate solution Q′i :

[0052] Q′ i = Q i + η(t)·(Q best - Q i ) + F DE (t)·(Q r2 - Q r3 );

[0053] where η(t) is the step - size control factor, and F DE (t) is the differential scaling factor;

[0054] S48. Define the adaptive update strategies for the step - size control factor η(t) and the differential scaling factor F DE (t) as follows:

[0055]

[0056] F DE (t) = F max ·(1 - D g (t)) + F min ;

[0057] where η max , η min are the initial and minimum step - sizes respectively, F max , F min are the boundary values of the differential scaling factor, T is the maximum number of iterations, and D g (t) is the current population diversity function;

[0058] S49. Recalculate the fitness function F i (Q′ e ) for the updated candidate solution Q′ i , compare the values of F e (Q′ i ) and the original candidate solution F e (Q i ). If F e (Q′ i ) < F e (Q i ), then set Q i = Q′ i , otherwise keep Q i unchanged;

[0059] S410. Repeat steps S45 to S49 until the set maximum number of iterations is reached or the fitness change meets the termination condition;

[0060] S411. Select the candidate solution with the minimum fitness function value from the finally updated candidate solution set, denoted as the optimal image enhancement parameter vector Q opt = [B * , C * , S * , R * , where B * represents the optimized brightness parameter, C * represents the optimized contrast parameter, S * represents the optimized saturation parameter, R * represents the optimized sharpening parameter, and output the optimal image enhancement parameter vector as the optimal image enhancement parameter of the image optimization and enhancement module for image enhancement processing.

[0061] Optionally, the image enhancement parameter candidate solution vector includes a brightness parameter, a contrast parameter, a saturation parameter, and a sharpening parameter, which are used to optimize the visual effect of the image and enhance details.

[0062] Optionally, the S5 specifically includes:

[0063] S51. Receive the optimal image enhancement parameter vector Q opt = [B * , C * , S * , R * obtained in claim 4, and the input image I denoise after noise suppression processing output in claim 3;

[0064] S52. Adjust the brightness of the image after noise suppression processing, and adjust the overall brightness level of the image according to the optimized brightness parameter B * to improve the visual visibility of the image in low-light or underexposed scenarios;

[0065] S53. Enhance the contrast of the image according to the optimized contrast parameter C * to improve the layering and structural boundary clarity of the image by adjusting the brightness difference between pixels;

[0066] S54. Correct the color saturation of the image according to the optimized saturation parameter S * to avoid problems such as gray and pale colors caused by noise reduction or environmental factors;

[0067] S55. Enhance the edges of the image according to the optimized sharpening parameter R * to enhance high-frequency information such as edge textures and contours in the image;

[0068] S56. Perform fusion processing on the image after brightness, contrast, saturation, and sharpening processing, and finally output the enhanced input image.

[0069] Optionally, the S6 specifically includes:

[0070] S61. Receive the input image enhanced by the image optimization and enhancement module, and denote it as I enhance ;

[0071] S62. Construct a comprehensive joint loss function, defined as:

[0072] L total = α·L recon + β·L perceptual + γ·L adv + δ·L reg ;

[0073] Among them, α, β, γ, and δ are the weight coefficients of the reconstruction error L recon , the perceptual error L perceptual , the adversarial error L adv and the parameter regularization error L reg respectively;

[0074] S63. Calculate the reconstruction error L recon for the input image processed by the image optimization and enhancement module:

[0075]

[0076] Among them, I target is the reference image, N is the total number of pixels in the image, and w(i) is the weight factor adaptively calculated based on the image edge information;

[0077] S64. Calculate the perceptual error L perceptual for the input image processed by the image optimization and enhancement module:

[0078]

[0079] Among them, Φ j () represents the feature map of the j-th layer in the pre-trained feature extraction network, M is the number of selected feature layers, is the difference metric based on the local texture structure of the image, and κ is the weight coefficient of this difference metric;

[0080] S65. Calculate the adversarial error L adv and the regularization error L reg for the input image processed by the image optimization and enhancement module;

[0081] S66. Weighted sum the various errors calculated in steps S63 to S65 according to the weights given in S62 to calculate the comprehensive joint loss L total and output the joint loss value as an evaluation index for measuring the overall quality of the enhanced image.

[0082] The beneficial effects of the present invention are as follows:

[0083] By deeply integrating a deep neural network and a swarm intelligence optimization algorithm, the present invention realizes the adaptive suppression of image noise and the intelligent optimization of image quality in a camera module, fundamentally improving the imaging effect and visual performance. The differential evolution algorithm is used to globally search the hyperparameters of the deep denoising network, enabling the network to adaptively adjust the convolution kernel size, the number of network layers, the number of channels, and the activation function parameters, thereby optimizing the denoising effect in various noise environments. At the same time, the present invention further uses an improved flamingo optimization algorithm to perform multi-objective search on the image enhancement parameters. By introducing a differential evolution mechanism and an adaptive parameter update strategy, an effective balance between global exploration and local refinement is achieved, ensuring that the image achieves the best performance in aspects such as detail restoration, color correction, contrast enhancement, and sharpening processing.

[0084] Through the technical solution of the present invention, not only can the noise interference in the image be effectively reduced, but also valuable details such as edges and textures in the original image can be retained, thereby avoiding common problems such as over-smoothing and detail loss in traditional methods. At the same time, the comprehensive joint loss function adopted by the present invention comprehensively evaluates the reconstruction error, the perceptual error, the adversarial error, and the regularization error, and provides real-time feedback on the image enhancement effect, providing a scientific basis for the dynamic adjustment of image enhancement parameters. This intelligent image processing method has high adaptability and robustness in complex imaging environments, can achieve real-time processing, and meet the requirements of different application scenarios for image clarity and authenticity.

[0085] Generally speaking, the beneficial effects of the present invention are reflected in that it can significantly reduce noise interference while maintaining the details and realism of the image, and adaptively adjust parameters during the image enhancement process to achieve the overall optimal balance of color, contrast, and brightness. This method not only improves the image quality and processing efficiency, but also has strong application and promotion value, providing reliable technical support for the high-performance applications of camera modules in fields such as smartphones, monitoring systems, and vehicle-mounted systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0087] Figure 1Flowchart of an intelligent noise removal and image optimization method for a camera module proposed by the present invention;

[0088] Figure 2 Schematic diagram of globally searching for hyperparameters of a deep denoising network using a differential evolution algorithm in an intelligent noise removal and image optimization method for a camera module proposed by the present invention. Detailed implementation manners

[0089] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0090] Reference Figure 1 and Figure 2 , an intelligent noise removal and image optimization method for a camera module, includes the following steps:

[0091] S1. Collect the input image and preprocess the input image;

[0092] S2. Use a differential evolution algorithm to globally search for the hyperparameters of the deep denoising network and determine the deep denoising network structure;

[0093] S3. Use the deep denoising network to perform noise suppression processing on the preprocessed input image, reduce the noise interference in the input image and retain the details of the input image;

[0094] S4. Input the input image after noise suppression processing into the image optimization and enhancement module, and perform multi-objective search for the image enhancement parameters through an improved flamingo optimization algorithm to obtain the optimal image enhancement parameters;

[0095] S5. Perform image enhancement processing on the input image after noise suppression processing according to the optimal image enhancement parameters;

[0096] S6. Use a comprehensive joint loss function to evaluate the quality of the enhanced input image. The joint loss function includes reconstruction error, perceptual error, adversarial error, and parameter regularization error, and measures the overall quality of the enhanced input image;

[0097] S7. According to the feedback quality evaluation result, iteratively adjust the initialization parameters of the flamingo optimization algorithm through the differential evolution algorithm, and finally output the image after noise removal and image optimization processing.

[0098] Through the deep integration of a deep neural network and a swarm intelligence optimization algorithm, the present invention realizes the adaptive removal of image noise and the intelligent optimization of image quality in a camera module, significantly improving the clarity, detail retention, and color restoration effect of the image. The differential evolution algorithm is used to globally search for the hyperparameters of the noise reduction network, enabling the network structure to be adaptively adjusted to achieve precise noise reduction for different types of noise, avoiding the problems of over-smoothing and detail loss in traditional methods. At the same time, the improved flamingo optimization algorithm is used to perform multi-objective search on the image enhancement parameters. By introducing a differential evolution mechanism and an adaptive parameter adjustment strategy, the image achieves an optimal configuration in terms of brightness, contrast, saturation, and sharpening processing, comprehensively improving the visual effect. After the quality of the enhanced image is evaluated by a comprehensive combined loss function, the parameters are further optimized through a feedback mechanism to achieve full-process adaptive adjustment. Experimental data show that the method of the present invention significantly improves the image clarity, color restoration degree, and detail retention rate in complex environments such as low light and high ISO, greatly shortening the processing time, meeting the application requirements of real-time monitoring, etc., and having high robustness and application promotion value.

[0099] In this embodiment, the collected input image includes the original image data, the color information of the image, the illumination information of the image, the spatial resolution of the image, and the noise in the image, which is used to provide image data for noise suppression, image optimization, and enhancement processing.

[0100] In this embodiment, the image preprocessing includes image denoising, image normalization, color space conversion, and size adjustment, which are used to reduce image noise, improve image quality, and unify the image data format.

[0101] In this embodiment, S2 specifically includes:

[0102] S21. Construct a candidate solution vector P;

[0103] S22. Initialize the candidate solution set:

[0104] {P i},i = 1, 2, …, N;

[0105] where N is the total number of candidate solutions, and each candidate solution P i is a set of hyperparameter configurations;

[0106] S23. Calculate the fitness function value F(P i ) for each candidate solution P i :

[0107] F(P i ) = α·E recon (P i ) + β·E reg (P i ) + γ·Estruct (P i );

[0108] where α, β, and γ are weight coefficients, and E recon (P i ) represents the reconstruction error, and E reg (P i ) represents the regularization error, and E struct (P i ) represents the network structure consistency error;

[0109] S24. For each candidate solution P i Randomly select three different candidate solutions P r1 , P r2 , and P r3 , and calculate the mutation vector:

[0110]

[0111] where F0 is the initial scaling factor, F min is the minimum scaling factor, t is the current iteration number, and T is the preset maximum iteration number;

[0112] S25. Perform a crossover operation on the candidate solution P i and the corresponding mutation vector M i to generate a trial vector T i , where the j-th component satisfies:

[0113]

[0114] where CR is the adaptive crossover rate, rand(j) is a randomly selected value from the interval [0, 1], T i,j represents the j-th component of the generated trial vector T i , M i,j represents the j-th component of the mutation vector M i , and P i,j represents the j-th component of the original candidate solution P i ;

[0115] S26. Calculate the fitness function value F(T i ) of the trial vector T i . If F(T i ) < F(P i ), then let P i = T i , otherwise retain the original candidate solution;

[0116] S27. Repeat steps S23 to S26 until the preset iteration number or fitness convergence condition is reached, and select the candidate solution P best = [K* , L * , C * , A * , as the hyperparameter configuration of the deep denoising network, and construct the deep denoising network according to the fitness-optimal candidate solution. The deep denoising network is composed of L * sequentially connected convolutional layers, where each convolutional layer adopts a convolutional kernel size of K * , the number of channels C * and the activation function parameter A * configuration.

[0117] In the present invention, by using the differential evolution algorithm to globally search for the hyperparameters of the deep denoising network, the network structure design is significantly optimized. By means of constructing candidate solution vectors, initializing the candidate solution set, and calculating the fitness function values for each candidate solution, the system can efficiently search for the optimal hyperparameter configuration in the high-dimensional parameter space. By randomly selecting candidate solutions and calculating the mutation vectors, and then combining with the crossover operation to generate trial vectors, and continuously iterating and updating, finally select the candidate solution with the optimal fitness, so as to construct a deep denoising network composed of multiple sequentially connected convolutional layers. This network can not only effectively suppress the noise interference in the input image in terms of the denoising effect, but also fully retain the image details and avoid the over-smoothing phenomenon that appears in traditional methods. Under the multi-objective comprehensive evaluation such as reconstruction error, regularization error, and network structure consistency error, the optimized network has higher image denoising accuracy and robustness. At the same time, the technical solution of the present invention greatly shortens the network tuning time, improves the real-time processing ability of the system, provides strong technical support for the high-quality output of the camera module image in complex scenarios, and has significant practicality and promotion value.

[0118] In this embodiment, the candidate solution vector includes the convolutional kernel size, the number of network layers, the number of channels in the convolutional layer, and the activation function parameter, which are used to optimize the structure and performance of the deep denoising network.

[0119] In this embodiment, the S3 specifically includes:

[0120] S31. Receive the preprocessed input image, and the input image data format is consistent with the input requirements of the deep denoising network;

[0121] S32. Load the deep denoising network structure constructed by the optimal hyperparameter configuration determined by the differential evolution algorithm. The deep denoising network structure includes multiple sequentially connected convolutional layers, activation layers, and normalization layers;

[0122] S33. Send the preprocessed input image into the loaded deep denoising network, and sequentially process the input image through the forward propagation of each layer in the deep denoising network;

[0123] S34. During the forward propagation process, through the layer-by-layer action of convolution operations and activation functions, the deep denoising network suppresses the noise points in the input image while retaining the edge and texture detail information of the input image;

[0124] S35. Obtain the input image result after noise suppression processing at the end of the deep denoising network, and conduct a preliminary evaluation on the input image result to confirm that the noise interference is significantly reduced and the detail information is effectively retained;

[0125] S36. Output the input image after noise suppression processing, and transfer the input image result to the image optimization and enhancement module.

[0126] The present invention uses a deep denoising network to perform noise suppression processing on the preprocessed input image. By loading the optimal hyperparameter configuration determined by the differential evolution algorithm to construct the network, it ensures that the input image data format is consistent with the network requirements, thereby realizing efficient forward propagation processing. During the entire processing process, each convolutional layer, activation layer, and normalization layer perform operations on the image in sequence according to the pre-determined optimal configuration, enabling the network to effectively suppress the noise points in the image during the layer-by-layer transfer process, while fully retaining the edge and texture detail information, avoiding the over-smoothing and detail loss phenomena that occur in traditional noise reduction methods. The image processed by the deep denoising network shows a significant reduction in noise interference after preliminary evaluation, and the image details and structural features are better retained, further providing high-quality input for the subsequent image optimization and enhancement module. This technical solution not only improves the accuracy and robustness of image noise suppression, but also greatly enhances the real-time performance of image processing, providing reliable technical support for high-quality imaging of camera modules in complex lighting environments, and having significant practical value and promotion prospects.

[0127] In this embodiment, the specific steps of S4 are as follows:

[0128] S41. Receive the input image after noise suppression processing described in claim 3, denoted as I denoise ;

[0129] S42. Apply the image optimization and enhancement module to perform image enhancement processing. The image optimization and enhancement module includes a parameter optimization sub-module and an image enhancement execution sub-module. The parameter optimization sub-module is used to search for image enhancement parameters through an optimization algorithm, and the image enhancement execution sub-module is used to perform enhancement processing on the image based on the image enhancement parameters;

[0130] S43. Construct an image enhancement parameter candidate solution vector Q;

[0131] S44. Initialize the image enhancement parameter candidate solution set {Q i}, where i = 1, 2,..., M, and M is the number of candidate solutions;

[0132] S45. For each candidate solution Q i Calculate the multi-objective fitness function:

[0133] F e (Q i ) = λ·E detail (Q i ) + μ·E color (Q i ) + ν·E contrast (Q i );

[0134] where λ, μ, and ν are the weight coefficients for detail restoration, color restoration, and contrast enhancement respectively, and E detail (Q i ), E color (Q i ), and E contrast (Q i ) are the error evaluation functions of the enhanced image based on the candidate solution Q i in the corresponding dimensions;

[0135] S46. Select the candidate solution with the minimum value of the current fitness function F e (Q i ) in the population, denoted as Q best ;

[0136] S47. During the process of updating the image enhancement parameters, for the candidate solution Q i , introduce the difference vector term Q r2 - Q r3 , where Q r2 and Q r3 are two different candidate solutions randomly selected from the population, and calculate the updated candidate solution Q′ i :

[0137] Q′ i = Q i + η(t)·(Q best - Q i ) + F DE (t)·(Q r2 - Q r3 );

[0138] where η(t) is the step size control factor and F DE (t) is the difference scaling factor;

[0139] S48. Define the adaptive update strategies for the step size control factor η(t) and the difference scaling factor F DE (t) as follows:

[0140]

[0141] F DE f(t) = F max ·(1 - D g (t)) + F min ;

[0142] where η max and η min are the initial and minimum step sizes respectively, F max and F min are the boundary values of the differential scaling factor, T is the maximum number of iterations, and D g (t) is the current population diversity function;

[0143] S49. Recalculate the fitness function F i (Q′ e ) for the updated candidate solution Q′ i . Compare the value of F e (Q′ i ) with that of the original candidate solution F e (Q i ). If F e (Q′ i ) < F e (Q′ i ), then let Q i = Q′ i , otherwise keep Q i unchanged;

[0144] S410. Repeat steps S45 to S49 until the set maximum number of iterations is reached or the fitness change meets the termination condition;

[0145] S411. Select the candidate solution with the minimum fitness function value from the finally updated candidate solution set, denoted as the optimal image enhancement parameter vector Q opt = [B * , C * , S * , R * , where B * represents the optimized brightness parameter, C * represents the optimized contrast parameter, S * represents the optimized saturation parameter, and R * represents the optimized sharpening parameter, and output the optimal image enhancement parameter vector as the optimal image enhancement parameter of the image optimization and enhancement module for image enhancement processing.

[0146] The present invention combines an improved flamingo optimization algorithm with a differential evolution mechanism to perform multi-objective search on image enhancement parameters, significantly enhancing the adaptive regulation ability and optimization accuracy of the image optimization and enhancement module. In this method, by constructing a candidate solution vector of image enhancement parameters and dynamically adjusting the step size and differential scaling factor using an adaptive update strategy, the parameter search process can both globally explore and locally refine, effectively avoiding the defect of being easily trapped in local optima in traditional algorithms. After multiple rounds of iterative updates, the candidate solution with the smallest fitness function value is finally selected as the optimal image enhancement parameter, thereby achieving the best effects in aspects such as brightness, contrast, saturation, and sharpening processing. Experimental data shows that the method of the present invention can make the image colors more vivid and the edge details clearer in low-light and high-noise environments, while significantly shortening the image processing time, meeting the requirements of real-time monitoring and high-precision imaging. While improving the image quality, this technical solution significantly enhances the robustness and stability of the system, providing strong technical support for the application of camera modules in intelligent monitoring, vehicle-mounted systems, and other high-performance image processing fields, and having broad prospects for popularization and application.

[0147] In this embodiment, the candidate solution vector of image enhancement parameters includes a brightness parameter, a contrast parameter, a saturation parameter, and a sharpening parameter, which are used to optimize the visual effect and detail enhancement of the image.

[0148] In this embodiment, S5 specifically includes:

[0149] S51. Receive the optimal image enhancement parameter vector Q opt = [B * , C * , S * , R * obtained in claim 4, and the input image I after noise suppression processing output in claim 3 denoise ;

[0150] S52. Adjust the brightness of the image after noise suppression processing, and adjust the overall brightness level of the image according to the optimized brightness parameter B * to improve the visual visibility of the image in low-light or underexposed scenes;

[0151] S53. Enhance the contrast of the image according to the optimized contrast parameter C * to improve the layering and structural boundary clarity of the image by adjusting the brightness difference between pixels;

[0152] S54. Correct the color saturation of the image according to the optimized saturation parameter S * to avoid problems such as grayish or pale colors caused by noise reduction or environmental factors;

[0153] S55. According to the optimized sharpening parameter R * perform edge enhancement processing on the image to enhance high-frequency information such as edge textures and contours in the image;

[0154] S56. Perform fusion processing on the image that has undergone brightness, contrast, saturation, and sharpening processing, and finally output the enhanced input image.

[0155] The present invention performs a series of enhancement processes on the image after noise suppression processing using the optimal image enhancement parameter vector, so that the finally output image achieves the best effect in all aspects. By receiving the optimal image enhancement parameters and the noise-reduced image, the present invention first adjusts the brightness of the image according to the optimized brightness parameter, thereby significantly improving the overall brightness and visual visibility of the image in low-light or underexposed scenarios; then, enhances the contrast of the image according to the optimized contrast parameter, effectively strengthening the edges and details of the objects in the image and making the image levels more distinct; at the same time, corrects the image color according to the optimized saturation parameter, improving the problem of gray and pale colors caused by noise suppression or environmental influence, and making the image colors more vivid and real; in addition, enhances the image edges using the optimized sharpening parameter to highlight high-frequency information and further improve the image detail and contour clarity. The image that has undergone brightness, contrast, saturation, and sharpening processing undergoes fusion processing, and the finally output enhanced image not only has significantly reduced noise interference, but also achieves an all-round improvement in visual effects. Experimental data show that for this image enhancement processing method, in complex lighting and high-noise environments, both the image clarity and the detail retention rate are significantly improved, the system runs stably, and the real-time performance is excellent, providing reliable technical support and broad application prospects for camera modules in fields such as intelligent monitoring and vehicle-mounted systems.

[0156] In this embodiment, the specific steps of S6 are as follows:

[0157] S61. Receive the input image enhanced by the image optimization and enhancement module, and denote it as I enhance ;

[0158] S62. Construct a comprehensive joint loss function, defined as:

[0159] L total =α·L recon +β·L perceptual +γ·L adv +δ·L reg ;

[0160] where α, β, γ, and δ are the weight coefficients of the reconstruction error L recon , the perceptual error L perceptual , the adversarial error L adv , and the parameter regularization error L reg respectively;

[0161] S63. Calculate the reconstruction error L for the input image processed by the image optimization and enhancement module recon :

[0162]

[0163] where I target is the reference image, N is the total number of pixels in the image, and w(i) is the weight factor adaptively calculated based on the image edge information;

[0164] S64. Calculate the perceptual error L for the input image processed by the image optimization and enhancement module perceptual :

[0165]

[0166] where Φ j () represents the feature map of the j-th layer in the pre-trained feature extraction network, M is the number of selected feature layers, is the difference metric based on the local texture structure of the image, and κ is the weight coefficient of this difference metric;

[0167] S65. Calculate the adversarial error L adv and the regularization error L reg ;

[0168] S66. Weighted sum the various errors calculated in steps S63 to S65 according to the weights given in S62 to calculate the comprehensive joint loss L total , and output this joint loss value as an evaluation index for measuring the overall quality of the enhanced image.

[0169] The present invention uses a comprehensive combined loss function to quantitatively evaluate the overall quality of the enhanced image. By weighted summation of the reconstruction error, perceptual error, adversarial error, and parameter regularization error, a global quality evaluation index is formed, so as to accurately reflect the detail restoration and visual consistency of the image at all levels. In this method, by calculating the reconstruction error of the enhanced image and introducing an adaptively calculated weight factor, the importance of image edge and texture information can be more effectively highlighted, and the fine measurement of the image reconstruction quality can be realized. At the same time, based on the outputs of multiple feature layers in the pre-trained feature extraction network and the local texture structure difference measurement, the perceptual error provides effective feedback for the retention of high-level semantic information of the image; the introduction of the adversarial error and the parameter regularization error ensures the improvement of the image realism and the smoothness of the model parameters, thus avoiding the overfitting problem. Through the comprehensive combined loss function, the present invention can not only monitor the change of the image quality in real time, but also provide an accurate adjustment basis for subsequent optimization of the enhancement parameters, greatly improving the stability and robustness of the image enhancement effect. Experimental results show that this method can significantly reduce noise interference in complex environments, improve the detail retention and color restoration ability of the image, and provide reliable technical support for the camera module in intelligent monitoring and high-precision imaging applications.

[0170] Example 1:

[0171] To verify the feasibility of the present invention in implementation, the present invention is applied to the intelligent security monitoring system of a certain city. Traditional camera modules often have problems such as serious image noise, blurred details, and color distortion in nighttime and complex lighting environments, resulting in unclear monitoring pictures and affecting the real-time judgment of security personnel on the on-site situation. To solve this problem, the intelligent noise removal and image optimization method for camera modules proposed by the present invention has been successfully applied. This method realizes image denoising through a deep neural network, and uses the differential evolution algorithm and the improved flamingo optimization algorithm to adaptively optimize the network hyperparameters and image enhancement parameters, so as to significantly improve the image quality in different environments such as different lighting, low light, and high ISO.

[0172] In this embodiment, an experiment was conducted at a monitoring point in the central business district of a certain city. The monitoring camera was installed on the top of a building, about 30 meters above the ground, mainly responsible for monitoring the flow of people and vehicles at night. In the images collected by traditional cameras in low-light environments, there are many noise points, the detailed information is blurred, and the image quality can only reach about 50% clarity, often with problems such as color deviation to gray and blurred edges. For this reason, the intelligent noise removal and image optimization method of the present invention was introduced into this monitoring system. The system first preprocesses the collected images and models the noise, and then uses a deep denoising network based on global search of the differential evolution algorithm to adaptively suppress the noise points in the images. After denoising, an improved flamingo optimization algorithm is used to perform multi-objective search on the image enhancement parameters to achieve the joint optimization of brightness, contrast, saturation, and sharpening processing.

[0173] During the actual operation process, the system continuously monitors the data from 20:00 every night to 4:00 the next morning. After preliminary statistics, the average image clarity of the images processed by the traditional camera module during the above period is 52%, the color restoration degree is 48%, and the detail retention rate is about 45%. For the images processed by the method of the present invention, the average clarity is increased to 89%, the color restoration degree reaches 85%, and the detail retention rate is increased to 82%. The outlines of vehicles, pedestrians, and the surrounding environment in the monitoring video are clearer, the fine noise points are effectively removed, and the overall visual effect of the picture is significantly improved.

[0174] In addition, the method of the present invention also has obvious advantages in processing time. The optimized deep denoising network and image optimization and enhancement module, under the condition of GPU acceleration, shorten the processing time of a single frame of image from 80 milliseconds of the traditional method to 45 milliseconds, meeting the requirements of real-time monitoring. During the experiment of the system, a total of about 500,000 frames of image data were processed. Among them, the images processed by the method of the present invention met the requirements of real-time video monitoring, and there were no obvious delays or resource bottleneck problems during the processing. The experimental data show that in different lighting environments, the method of the present invention can stably output high-quality images, and the image optimization effect has good robustness.

[0175] In order to further verify the beneficial effects of the present invention, we compared and analyzed the image quality and processing performance of the images processed by the traditional image processing method and the method of the present invention. The detailed data are shown in Table 1. Through statistics, it is shown that the method of the present invention is superior to the traditional method in multiple indicators such as image denoising, detail restoration, and color correction, and has obvious advantages in terms of real-time performance and stability. The experimental location is a monitoring point in the central business district of a certain city, and the test time is from October 2024 to January 2025. Continuous monitoring data is collected every night, and each indicator is statistically analyzed.

[0176] Table 1 Comparison data table of image quality and processing performance of camera modules

[0177]

[0178] The above table details the significant differences between the traditional method and the method of the present invention in terms of the image quality and processing performance of the camera module. From the comparison data, it can be seen that in terms of image clarity, the traditional method only reaches 52% on average, while the method of the present invention can increase the image clarity to 89%, with an increase of 37 percentage points. This indicates that after adopting the technology of the present invention, the noise in the image is effectively suppressed, and the image edge and texture details are better retained. The color restoration also shows a similar improvement effect. The color restoration of the traditional method is 48%, while the method of the present invention can reach 85%. This improvement makes the colors of the image more vivid and natural, avoiding the phenomenon of color distortion. The detail retention rate is increased from 45% of the traditional method to 82%, fully demonstrating the obvious advantage of the present invention in detail restoration.

[0179] In addition, in terms of the single-frame processing time, the traditional method requires 80 milliseconds on average, while the method of the present invention only requires 45 milliseconds, and the processing time is shortened by 35 milliseconds. This is of great significance in application scenarios such as real-time video monitoring, ensuring the real-time performance and response speed during the image processing process. The real-time score of the system is also increased from 5 points of the traditional method to 9 points of the method of the present invention, further verifying the stability and high efficiency of the overall system under high-load conditions. In terms of the continuous operation time, the traditional method can only ensure the system to run for about 6 hours, while the method of the present invention realizes 12 hours of continuous and stable operation.

[0180] In summary, the data in the table fully illustrate that the method of the present invention has made significant improvements in image quality improvement, processing efficiency, and system stability. These data not only verify the actual effects of the technology of the present invention in reducing image noise, enhancing image details, and optimizing visual effects, but also show that the technical solution has obvious advantages in real-time performance and long-term operation stability, providing strong technical support for the application of the camera module in intelligent monitoring, vehicle-mounted systems, and other high-performance image processing fields.

[0181] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An intelligent noise removal and image optimization method for a camera module, characterized in that, It includes the following steps: S1. Collect the input image and preprocess the input image; S2. Use the differential evolution algorithm to globally search for the hyperparameters of the deep denoising network and determine the deep denoising network structure; S3. Use the deep denoising network to perform noise suppression processing on the preprocessed input image, reduce the noise interference in the input image and retain the details of the input image; S4. Input the input image after noise suppression processing into the image optimization and enhancement module, and perform multi-objective search on the image enhancement parameters through the improved flamingo optimization algorithm to obtain the optimal image enhancement parameters; S5. Perform image enhancement processing on the input image after noise suppression processing according to the optimal image enhancement parameters; S6. Use a comprehensive joint loss function to evaluate the quality of the enhanced input image. The joint loss function includes reconstruction error, perceptual error, adversarial error, and parameter regularization error, and measures the overall quality of the enhanced input image; S7. According to the feedback quality evaluation results, iteratively adjust the initial parameters of the flamingo optimization algorithm through the differential evolution algorithm, and finally output the image after noise removal and image optimization processing.

2. The intelligent noise removal and image optimization method for a camera module according to claim 1, characterized in that, The collection of the input image includes the original image data, the color information of the image, the illumination information of the image, the spatial resolution of the image, and the noise in the image, which is used to provide image data for noise suppression, image optimization, and enhancement processing.

3. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The image preprocessing includes image denoising, image normalization, color space conversion, and size adjustment, which are used to reduce image noise, improve image quality, and unify the image data format.

4. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The specific steps of S2 include: S21. Construct the candidate solution vector P; S22. Initialize the candidate solution set: {P i}, where \(i = 1, 2, \ldots, N\); where N is the total number of candidate solutions, and each candidate solution P i is a set of hyperparameter configurations; S23. For each candidate solution P i calculate the fitness function value F(P i ): F(P i ) = α·E recon (P i ) + β·E reg (P i ) + γ·E struct (P i ); where α, β, and γ are weight coefficients, E recon (P i ) represents the reconstruction error, E reg (P i ) represents the regularization error, E struct (P i ) represents the network structure consistency error; S24. For each candidate solution P i Randomly select three different candidate solutions P r1 , P r2 and P r3 , and calculate the mutation vector: Among them, F0 is the initial scaling factor, F min is the minimum scaling factor, t is the current iteration number, and T is the preset maximum iteration number; S25. Perform crossover operation on the candidate solution P i and the corresponding mutation vector M i to generate a trial vector T i , where the j-th component satisfies: Among them, CR is the adaptive crossover rate, rand(j) is a value randomly selected from the interval [0, 1], and T i,j represents the j-th component of the generated trial vector T i , M i,j represents the j-th component of the mutation vector M i , P i,j represents the j-th component of the original candidate solution P i ; S26. Calculate the fitness function value F(T i ) of the test vector T i . If F(T i ) < F(P i ), then let P i = T i , otherwise retain the original candidate solution; S27. Repeat steps S23 to S26 until a preset number of iterations or a fitness convergence condition is reached, and select the candidate solution P with the optimal fitness best = [K * , L * , C * , A * , as the hyperparameter configuration of the deep denoising network, and construct the deep denoising network according to the candidate solution with the optimal fitness. The deep denoising network consists of L * sequentially connected convolutional layers, where each convolutional layer adopts a convolutional kernel size of K * , a number of channels of C * and an activation function parameter of A * .

5. A method for intelligent noise removal and image optimization of a camera module according to claim 3, characterized in that, The candidate solution vector includes the convolution kernel size, the number of network layers, the number of channels in the convolutional layer, and the activation function parameters, which are used to optimize the structure and performance of the deep denoising network.

6. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The specific steps of S3 include: S31. Receive the preprocessed input image, and the input image data format is consistent with the input requirements of the deep denoising network; S32. Load the deep denoising network structure constructed by the optimal hyperparameter configuration determined by the differential evolution algorithm. The deep denoising network structure includes multiple sequentially connected convolutional layers, activation layers, and normalization layers; S33. Send the preprocessed input image into the loaded deep denoising network, and process the input image sequentially through the forward propagation of each layer in the deep denoising network; S34. During the forward propagation process, through the layer-by-layer action of the convolution operation and the activation function, the deep denoising network suppresses the noise in the input image, and at the same time retains the edge and texture detail information of the input image; S35. Obtain the input image result after noise suppression processing at the end of the deep denoising network, and perform a preliminary evaluation on the input image result to confirm that the noise interference is significantly reduced and the detail information is effectively retained; S36. Output the input image after noise suppression processing, and transfer the input image result to the image optimization and enhancement module.

7. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The specific steps of S4 include: S41. Receive the input image after the noise suppression process described in claim 3, denoted as I denoise ; S42. Apply the image optimization and enhancement module to image enhancement processing. The image optimization and enhancement module includes a parameter optimization sub-module and an image enhancement execution sub-module. The parameter optimization sub-module is used to search for image enhancement parameters through an optimization algorithm, and the image enhancement execution sub-module is used to enhance the image based on the image enhancement parameters; S43. Construct an image enhancement parameter candidate solution vector Q; S44. Initialize the candidate solution set {Q of the image enhancement parameters i}, where i = 1, 2, …, M, and M is the number of candidate solutions; S45. For each candidate solution Q i Calculate the multi-objective fitness function: F e (Q i ) = λ·E detail (Q i ) + μ·E color (Q i ) + ν·E contrast (Q i ); where λ, μ, and ν are the weight coefficients for detail restoration, color restoration, and contrast enhancement, respectively, and E detail (Q i ), E color (Q i ), and E contrast (Q i ) are the error evaluation functions of the enhanced image based on the candidate solution Q i in the corresponding dimensions, respectively; S46. Select the current fitness function F in the population e (Q i ) Select the candidate solution with the smallest value, denoted as Q best ; S47. During the process of updating the image enhancement parameters, for the candidate solution Q i , introduce the difference vector term Q r2 - Q r3 , where Q r2 , Q r3 are two different candidate solutions randomly selected from the population, and calculate the updated candidate solution Q′ i : Q′ i = Q i + η(t)·(Q best - Q i ) + F DE (t)·(Q r2 - Q r3 ); where η(t) is the step size control factor and F DE (t) is the differential scaling factor; S48. Define the adaptive update strategy of the step control factor η(t) and the differential scaling factor F DE (t) as follows: F DE F(t) = F max ·(1 - D g (t)) + F min ; Among them, η max and η min are the initial and minimum step sizes respectively, F max and F min are the boundary values of the differential scaling factor, T is the maximum number of iterations, and D g (t) is the current population diversity function; S49. Recalculate the fitness function F for the updated candidate solution Q′ i and compare the value of F(Q′) e e (Q′ i ) with that of the original candidate solution F(Q) e e (Q′ i ). If F(Q′) e e (Q i )<F(Q′) e e (Q′ i )<F(Q) i e (Q i ), then let Q i i = Q′ i i , otherwise keep Q i i unchanged; S410. Repeat steps S45 to S49 until the set maximum number of iterations is reached or the fitness change meets the termination condition; S411. Select the candidate solution with the minimum fitness function value from the finally updated candidate solution set, denoted as the optimal image enhancement parameter vector Q opt =[[B * ,[[C * ,[[S * ,[[R * , where B * represents the optimized brightness parameter, C * represents the optimized contrast parameter, S * represents the optimized saturation parameter, R * represents the optimized sharpness parameter, and output the optimal image enhancement parameter vector as the optimal image enhancement parameter of the image optimization and enhancement module for image enhancement processing.

8. A method for intelligent noise removal and image optimization of a camera module according to claim 6, characterized in that, The image enhancement parameter candidate solution vector includes brightness parameters, contrast parameters, saturation parameters, and sharpening parameters, which are used to optimize the visual effect and detail enhancement of the image.

9. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The specific steps of S5 are as follows: S51. Receive the optimal image enhancement parameter vector Q obtained in claim 4 opt = [B * , C * , S * , R * , and the input image I after noise suppression processing output in claim 3 denoise ; S52. Adjust the brightness of the image after noise suppression processing according to the optimized brightness parameter B * Adjust the overall brightness level of the image to improve the visual visibility of the image in low-light or underexposed scenarios; S53. According to the optimized contrast parameter C * Perform contrast enhancement on the image, and improve the layering and structural boundary clarity of the image by adjusting the brightness difference between pixels; S54. Correct the color saturation of the image according to the optimized saturation parameter S * to avoid problems such as grayish or faded colors caused by noise reduction or environmental factors; S55. According to the optimized sharpening parameter R * Perform enhancement processing on the image edges to enhance high-frequency information such as edge textures and contours in the image; S56. Perform fusion processing on the image after brightness, contrast, saturation, and sharpening processing, and finally output the enhanced input image.

10. A method for intelligent noise removal and image optimization of a camera module according to claim 1, characterized in that, The specific steps of S6 are as follows: S61. Receive the input image enhanced by the image optimization and enhancement module, and denote it as I enhance ; S62. Construct a comprehensive joint loss function, defined as: L total = α·L recon + β·L perceptual + γ·L adv + δ·L reg ; where α, β, γ, and δ are the weight coefficients of the reconstruction error L recon , the perception error L perceptual , the adversarial error L adv , and the parameter regularization error L reg , respectively; S63. For the enhanced image I enhance Calculate the reconstruction error L recon , where the reconstruction error is used to measure the similarity of the image at the pixel level; S64. Calculate the perceptual error L for the enhanced image I enhance where the perceptual error is used to measure the difference of the image in the high-level feature space; perceptual ​ S65. For the enhanced image I enhance Compute the adversarial error L adv and the parameter regularization error L reg , where L adv measures the deviation of the image realism, and L reg is used to constrain the smoothness of the enhanced model parameters; S66. Weighted sum the various errors calculated in steps S63 to S65 according to the weights given in S62 to calculate the comprehensive joint loss L total , and output the joint loss value as an evaluation index for measuring the overall quality of the enhanced image.

Images