Convolutional neural network driven image noise reduction method and system

Abstract

The invention relates to the technical field of image noise reduction, in particular to an image noise reduction method and system driven by a convolutional neural network, and the method comprises the following steps: calculating the edge density and texture richness of a sample image based on a training sample image set, evaluating the complexity of the image, classifying and screening the sample image, and training a noise reduction model to obtain a noise reduction model; and generating an image noise recognition model. According to the method, the training data set is optimized, targeted data input is provided for training of the noise reduction model, the image noise recognition and classification accuracy is improved, the reconstruction error is calculated, the noise recognition threshold value is dynamically adjusted, accurate recognition and marking of the noise area are achieved, and the noise recognition accuracy is improved in combination with noise type classification. Personalized processing according to noise characteristics is achieved, adaptability and flexibility of the model are enhanced, visual feature extraction and scene analysis of the processed image are combined, and it is ensured that a high-quality noise reduction result is obtained under various scenes and illumination conditions.

Description

Technical Field

[0001] The present invention relates to the technical field of image denoising, and in particular to a convolutional neural network driven image denoising method and system. Background Art

[0002] The field of image noise reduction technology involves the use of various methods to improve the quality of images captured by cameras and various image acquisition devices, aiming to remove random noise in images caused by optical, electronic, and sensor defects. It covers everything from basic image processing technology to complex algorithm design, including the use of hardware and software filtering technology, statistical methods, and artificial intelligence algorithms to identify and eliminate noise, and combines a variety of machine learning and deep learning technologies to improve noise reduction efficiency and the overall clarity of the image.

[0003] Among them, image denoising methods focus on reducing or eliminating image noise by processing image data using a variety of algorithms or programs, covering noise identification and image data processing, including the use of statistical methods, filtering techniques, and machine learning methods to learn data features, predict and repair noise in images, specifically involving local pixel comparison, noise pattern recognition, and pixel correction strategies to achieve the reduction or elimination of image noise and reduce visual noise without affecting the main content of the image.

[0004] Traditional image denoising methods are difficult to achieve ideal denoising effects when faced with highly complex or dynamically changing noise environments. They rely on preset parameters and static processing methods, lack in-depth analysis and targeted processing of the complexity of image content, resulting in poor results when processing diverse image content. In rapidly changing scenes, such as dynamic video or multiple light sources, common filtering and statistical methods are difficult to adapt to the changing noise characteristics, are usually not flexible enough in automatically adjusting processing parameters, and lack an effective real-time optimization mechanism, limiting their breadth and effectiveness in practical applications. In real-time processing systems that require rapid response, they cannot meet the requirements of efficient processing and real-time response, resulting in unstable processing quality and affecting the image display effect. Summary of the invention

[0005] In order to solve the technical problem of unstable processing quality in the prior art, an embodiment of the present invention provides a convolutional neural network driven image denoising method and system. The technical solution is as follows:

[0006] On the one hand, a convolutional neural network driven image denoising method is provided, the method comprising:

[0007] S1: Based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained to generate an image noise recognition model;

[0008] S2: using the image noise recognition model, inputting the target processed image, calculating the error between the image and the reconstructed image, comparing the error size and identifying the noise area in the image, and adjusting the noise recognition threshold according to the noise characteristics of the target image to generate noise feature record data;

[0009] S3: Based on the noise feature recording data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type to generate a noise type classification result;

[0010] S4: According to the noise type classification result, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating a noise reduction parameter optimization configuration;

[0011] S5: Using the noise reduction parameter optimization configuration, by analyzing the brightness and color distribution data of the input image, identifying the scene type and lighting conditions of the image, dynamically adjusting the threshold setting for noise elimination and processing the image, to obtain the input image noise reduction result.

[0012] As a further solution of the present invention, the image noise recognition model specifically includes edge density analysis results, texture richness evaluation results, and sample image screening results; the noise feature record data includes reconstruction error calculation results, noise area identification results, and noise identification thresholds; the noise type classification results specifically refer to data feature comparison results, noise type identification records, and noise reduction processing method matching lists; the noise reduction parameter optimization configuration includes noise processing effect information, processing parameter adjustment records, and model parameter verification results; the input image denoising results specifically include image feature extraction records, scene type identification information, and noise elimination threshold configuration.

[0013] As a further solution of the present invention, based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained, and the steps of generating the image noise recognition model are specifically as follows:

[0014] S101: Based on the training sample image set, analyze and record the edge density and texture richness of multiple sample images to generate sample image feature data;

[0015] S102: Calculating the complexity index of the image based on the sample image feature data, and classifying the image to generate classified image data;

[0016] S103: Based on the classified image data, sample images are screened according to a preset complexity step, a training data set is optimized, and a noise reduction model is trained to generate an image noise recognition model.

[0017] As a further solution of the present invention, the specific formula for calculating the complexity index of the image is:

[0018]

[0019] Among them, CI is the complexity index of the target sample image, i is the sequence number of the feature data, n is the total number of features, and w i is the weight of the i-th feature, x i is the measured value of the i-th feature, μ i is the mean of the i-th eigenvalue.

[0020] As a further solution of the present invention, the steps of using the image noise recognition model, inputting the target processed image, calculating the error between the image and the reconstructed image, comparing the error size and identifying the noise area in the image, and adjusting the noise recognition threshold according to the noise characteristics of the target image to generate noise feature record data are specifically as follows:

[0021] S201: Based on the image noise recognition model, input a target processing image, calculate the error between the image and the model reconstructed image, and generate reconstruction error data;

[0022] S202: Based on the reconstructed error data, by comparing errors, identifying and marking the noise area in the image, and generating a noise area mark;

[0023] S203: According to the noise region mark and the noise characteristics of the target image, the noise recognition threshold is adjusted to optimize the recognition accuracy of the noise region, and noise feature recording data is generated.

[0024] As a further solution of the present invention, based on the noise feature recording data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type. The steps of generating the noise type classification result are specifically as follows:

[0025] S301: Based on the noise feature recording data, analyzing the recognized image noise, extracting data features of the noise, including physical and visual characteristics of the noise, and generating noise feature information;

[0026] S302: using the noise feature information, comparing the target noise with data features of multiple known noise types, calculating similarities, and generating similarity information;

[0027] S303: Identify multiple noise types, including Gaussian noise and salt and pepper noise, based on the similarity information, and match noise reduction processing methods and parameter configurations according to the types to generate a noise type classification result.

[0028] As a further solution of the present invention, the specific formula for calculating the similarity is:

[0029]

[0030] Among them, S pq represents the similarity value between the target noise p and the noise type q, m is the dimension of the feature data, and f pk is the value of the target noise p at the kth feature, f qk is the value of the kth feature of the known noise type q, p is the index of the target noise, q is the index of the known noise type, and k is the feature index.

[0031] As a further solution of the present invention, according to the noise type classification result, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating the noise reduction parameter optimization configuration steps are specifically as follows:

[0032] S401: Based on the noise type classification result, evaluating the noise reduction effects of multiple noise types, and generating a processing effect analysis result;

[0033] S402: According to the processing effect analysis result, identifying the recognition model parameters that need to be adjusted, including data processing depth, data processing speed, and data processing sensitivity, and generating an adjustment demand analysis result;

[0034] S403: According to the adjustment demand analysis result, the parameters of the recognition model are adjusted in consideration of processing efficiency and noise reduction effect, and an optimized configuration of noise reduction parameters is generated.

[0035] As a further solution of the present invention, the noise reduction parameter optimization configuration is used to analyze the brightness and color distribution data of the input image, identify the scene type and lighting conditions of the image, dynamically adjust the threshold setting of noise elimination and process the image, and obtain the noise reduction result of the input image in the following specific steps:

[0036] S501: Analyze input image data using the noise reduction parameter optimization configuration, extract image brightness and color distribution data, and generate visual feature data;

[0037] S502: Based on the visual feature data, identifying the scene type and lighting conditions in the image according to the respective features of brightness and color, and generating a scene and lighting recognition result;

[0038] S503: According to the scene and lighting recognition result, the noise elimination threshold setting is adjusted according to the scene and lighting conditions, and the input image is subjected to noise reduction processing to generate an input image noise reduction result.

[0039] On the other hand, a convolutional neural network driven image denoising system is provided, the system is applied to a convolutional neural network driven image denoising method, the system comprising:

[0040] The sample classification module is based on the training sample image set. It evaluates the complexity of the image according to the edge density and texture richness of the sample image, classifies and screens the sample images, trains the denoising model, and generates an image noise recognition model.

[0041] The noise recognition module inputs the target processing image according to the image noise recognition model, calculates the error between the image and the reconstruction, recognizes and marks the noise area, adjusts the recognition threshold according to the noise characteristics of the image, and generates noise feature record data;

[0042] The classification processing module uses the noise feature recording data to analyze the identified image noise, and classifies the target noise, including Gaussian noise and salt and pepper noise, by comparing the data features of known noise types, and matches the noise reduction processing method to generate a noise type classification result;

[0043] The parameter optimization module generates a noise reduction parameter optimization configuration by evaluating the processing effects of various noise types according to the noise type classification results;

[0044] The scene adjustment module uses the noise reduction parameter optimization configuration, analyzes the brightness and color distribution of the input image, identifies the scene type and lighting conditions, dynamically adjusts the noise elimination threshold and processes the image to obtain the input image noise reduction result.

[0045] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0046] By optimizing the training data set, targeted data input is provided for the training of the denoising model, the accuracy of image noise recognition and classification is improved, the calculation of reconstruction error and dynamic adjustment of noise recognition threshold are carried out, accurate recognition and marking of noise areas are achieved, and the classification of noise types is combined to achieve personalized processing according to noise characteristics, enhance the adaptability and flexibility of the model, and combine the visual feature extraction and scene analysis of the processed image to ensure high-quality denoising results in a variety of scenes and lighting conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0048] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0049] Figure 2 This is a detailed flow chart of S1 of the present invention;

[0050] Figure 3 This is a detailed flow chart of S2 of the present invention;

[0051] Figure 4 This is a detailed flow chart of S3 of the present invention;

[0052] Figure 5 This is a detailed flow chart of S4 of the present invention;

[0053] Figure 6 This is a detailed flow chart of S5 of the present invention;

[0054] Figure 7 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0056] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0057] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0058] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0059] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0060] The embodiment of the present invention provides a convolutional neural network driven image denoising method, such as Figure 1 The process flow of the image denoising method driven by a convolutional neural network is shown in FIG. 1 . The process flow of the method may include the following steps:

[0061] S1: Based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained to generate an image noise recognition model;

[0062] S2: Using the image noise recognition model, input the target processed image, calculate the error between the image and the reconstructed image, compare the error size and identify the noise area in the image, and adjust the noise recognition threshold according to the noise characteristics of the target image to generate noise feature record data;

[0063] S3: Based on the noise feature recording data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type to generate the noise type classification result;

[0064] S4: According to the noise type classification results, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating the noise reduction parameter optimization configuration;

[0065] S5: Use the noise reduction parameter optimization configuration to analyze the brightness and color distribution data of the input image, identify the scene type and lighting conditions of the image, dynamically adjust the threshold setting of noise removal and process the image to obtain the noise reduction result of the input image.

[0066] The image noise recognition model specifically includes the edge density analysis results, texture richness assessment results, and sample image screening results. The noise feature record data includes the reconstruction error calculation results, the noise area identification results, and the noise identification threshold. The noise type classification results specifically refer to the data feature comparison results, the noise type identification records, and the noise reduction processing method matching list. The noise reduction parameter optimization configuration includes the noise processing effect information, the processing parameter adjustment records, and the model parameter verification results. The input image denoising results specifically include the image feature extraction records, the scene type identification information, and the noise elimination threshold configuration.

[0067] See also Figure 2Based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained, and the steps of generating the image noise recognition model are as follows:

[0068] S101: Based on the training sample image set, analyze and record the edge density and texture richness of multiple sample images to generate sample image feature data;

[0069] Image processing algorithms are used to perform edge detection and texture analysis on each sample image. Specifically, the Canny edge detection algorithm is used to determine the edge density in the image. By calculating the gradient strength and direction of each pixel in the image, the edge and non-edge areas are effectively distinguished. The gray-level co-occurrence matrix method is used to analyze the statistical distribution characteristics of the image texture, including contrast, correlation, uniformity and entropy. The target features can describe the richness and regularity of the image texture. Through these two techniques, the edge density and texture richness of the image are calculated and recorded to generate specific sample image feature data. The data set includes the edge density value and texture statistical parameters of each image, providing accurate basic data for complexity evaluation and image classification.

[0070] S102: Calculating the complexity index of the image based on the sample image feature data, and classifying the image to generate classified image data;

[0071] The specific formula for calculating the complexity index of an image is:

[0072]

[0073] Among them, CI is the complexity index of the target sample image, i is the sequence number of the feature data, n is the total number of features, and w i is the weight of the i-th feature, x i is the measured value of the i-th feature, μ i is the mean of the i-th eigenvalue.

[0074] formula:

[0075]

[0076] Detailed explanation of the formula and the process of formula calculation and derivation:

[0077] The formula is used to calculate the complexity index of the image, and the result is used to classify the image and further denoise it;

[0078] Parameter meaning and setting value:

[0079] n: total number of features, assuming it is 2, n = 2, reflecting the number of feature dimensions considered;

[0080] wi : The weight of the i-th feature, assuming w1 = 0.6, w2 = 0.4;

[0081] x i : The measured value of the i-th feature, assuming x1=150, x2=250;

[0082] μ i : The mean of the i-th eigenvalue, assuming μ1=100, μ2=200;

[0083] Substitute the parameters into the formula for calculation:

[0084]

[0085] The result 35.36 shows that the complexity index of the image is high, reflecting that the image has high complexity. The result is used to screen sample images and optimize the training process of the recognition model.

[0086] S103: According to the classified image data, sample images are screened according to a preset complexity step, a training data set is optimized, and a noise reduction model is trained to generate an image noise recognition model;

[0087] According to the set complexity threshold, suitable samples are selected for denoising model training. Convolutional neural network is used for model training. The neural network extracts deep features of the image through multi-layer convolution and pooling operations. The network weights are optimized through back propagation and gradient descent methods during the training process to ensure that the model has good recognition and processing capabilities for different types of noise. After the training is completed, the obtained image noise recognition model can quickly and accurately calibrate the noise area according to the input image data and provide processing strategies. The generated model shows high recognition rate and denoising ability in actual denoising applications, optimizes the entire training data set, and ensures the robustness and effectiveness of the model in the face of actual noise environments.

[0088] See also Figure 3 , using the image noise recognition model, input the target processed image, calculate the error between the image and the reconstructed image, compare the error size and identify the noise area in the image, and adjust the noise recognition threshold according to the noise characteristics of the target image. The specific steps of generating noise feature record data are as follows:

[0089] S201: Based on the image noise recognition model, input the target processing image, calculate the error between the image and the model reconstructed image, and generate reconstruction error data;

[0090] Image reconstruction is performed through convolutional neural networks implemented in deep learning frameworks such as TensorFlow. The process involves multiple convolutional layers and pooling layers to extract image features and reconstruct the image. The mean square error between the original image and the reconstructed image is calculated as the evaluation criterion. MSE measures the similarity between images by the average of the squared differences. A lower MSE value indicates a high similarity, and a high error indicates a high noise area. By quantifying the differences between images, reconstruction error data is generated. The data clearly indicates the error level of each pixel, providing accurate data support for subsequent noise area identification.

[0091] S202: Based on the reconstructed error data, by comparing the errors, identifying and marking the noise area in the image, and generating a noise area mark;

[0092] The threshold processing method in image processing technology is used, such as Otsu's automatic threshold method to determine the dividing line between noise and non-noise, automatically calculate the histogram analysis of the image grayscale, and select the optimal threshold in a way that maximizes the inter-class variance. The area above this threshold in the reconstructed error data is marked as a noise area. This technology automatically adjusts the threshold according to the error size, sensitively marks the potential noise area in the image, and generates noise area marking data. The target data records in detail the location and range of the image area identified as noise, providing an accurate basis for the formulation of further noise processing strategies.

[0093] S203: According to the noise region mark and the noise characteristics of the target image, the noise recognition threshold is adjusted to optimize the recognition accuracy of the noise region, and noise feature recording data is generated;

[0094] Adaptive threshold technology is used to fine-tune the noise recognition threshold. The process analyzes the error characteristics of the marked noise area and adjusts the threshold setting to enhance the model's adaptability to different noise environments. An adaptive algorithm is used to automatically optimize the threshold based on the distribution characteristics of the noise area. This adjustment method is automatically performed based on the changes in noise intensity and type within the image, thereby improving the accuracy of noise area identification. The generated noise feature record data includes optimized noise area markings and adjusted threshold parameters, providing a meticulous noise characteristic analysis for the denoising model, ensuring the efficiency and accuracy of the denoising process.

[0095] See also Figure 4 , based on the noise feature record data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type. The specific steps for generating the noise type classification result are as follows:

[0096] S301: Based on the noise feature recording data, analyzing the recognized image noise, extracting the data features of the noise, including the physical characteristics and visual characteristics of the noise, and generating noise feature information;

[0097] Digital image processing methods such as histogram analysis and waveform analysis are used to analyze the physical characteristics of each noise point, including intensity and frequency distribution. Edge detection techniques such as the Sobel operator are used to describe the visual characteristics of noise, including the shape of the noise and its position in the image. The analysis helps to understand the nature and manifestation of noise. The generated noise feature information records the physical and visual parameters of the noise in detail, providing accurate basic data for subsequent noise type comparison and classification.

[0098] S302: using the noise feature information, comparing the target noise with data features of multiple known noise types, calculating similarities, and generating similarity information;

[0099] The specific formula for calculating similarity is:

[0100]

[0101] Among them, S pq represents the similarity value between the target noise p and the noise type q, m is the dimension of the feature data, and f pk is the value of the target noise p at the kth feature, f qk is the value of the kth feature of the known noise type q, p is the index of the target noise, q is the index of the known noise type, and k is the feature index.

[0102] formula:

[0103]

[0104] Detailed explanation of the formula and the process of formula calculation and derivation:

[0105] The formula is used to calculate the similarity between two noise feature vectors, analyze the similarity of two sets of data in the multidimensional feature space, and the results are used to determine the degree of match between the target noise and the known noise types;

[0106] Parameter meaning and setting value:

[0107] S pq Represents the similarity value between the target noise p and the noise type q;

[0108] m is the dimension of feature data, assumed to be 4;

[0109] f pk and f qkare the values ​​of the target noise p and the known noise type q at the kth feature, assuming that the feature values ​​of the target noise p are 12, 22, 34, 51, and the feature values ​​of the known noise type q are 16, 26, 28, 42;

[0110] Substitute the parameters into the formula for calculation:

[0111]

[0112] S pq =0.988;

[0113] The result 0.988 indicates that the similarity between the target noise and the known noise types is 0.988, indicating that in the feature space, the target noise and the known noise types are very similar in multiple dimensions. The calculation process is used to classify the noise type and match the processing method.

[0114] S303: Identify multiple noise types, including Gaussian noise and salt and pepper noise, based on the similarity information, and match the noise reduction processing method and parameter configuration according to the type to generate a noise type classification result;

[0115] A decision tree classification algorithm is used to identify the type of noise. The algorithm divides the noise into categories such as Gaussian noise and salt and pepper noise according to different parameter values ​​of noise characteristics, such as similarity index and physical properties. Each noise type corresponds to a different noise reduction processing method and parameter configuration. For example, Gaussian noise uses a Gaussian filter, and salt and pepper noise uses a median filter, which ensures the pertinence and effectiveness of the noise reduction strategy. The generated noise type classification results guide the specific implementation of subsequent noise reduction processing and ensure the optimal selection and application efficiency of the processing strategy.

[0116] See also Figure 5 According to the noise type classification results, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating the noise reduction parameter optimization configuration steps are as follows:

[0117] S401: Based on the noise type classification result, evaluate the noise reduction effects of various noise types and generate processing effect analysis results;

[0118] Quantitative analysis methods such as signal-to-noise ratio and mean square error calculation are used to evaluate the noise reduction processing effect of each noise type. The process involves comparing the image after each noise treatment with the original image, recording the degree of improvement brought about by each processing strategy on noise reduction, and helping to determine the performance of each noise reduction technology under different noise conditions. The generated processing effect analysis results provide a scientific evaluation basis for optimizing the noise reduction model, ensuring the selection of the most effective noise reduction strategy.

[0119] S402: According to the processing effect analysis result, identifying the recognition model parameters that need to be adjusted, including data processing depth, data processing speed, and data processing sensitivity, and generating adjustment demand analysis results;

[0120] Parameter sensitivity analysis is performed through data analysis software to identify the model parameters that have the greatest impact on the noise reduction effect, including analysis of how data processing depth, processing speed and sensitivity affect the accuracy and rate of the noise reduction results. By determining which parameter adjustments can significantly improve performance, adjustment demand analysis results are generated. The target results guide subsequent parameter optimization decisions to ensure that the noise reduction model achieves optimal performance in practical applications.

[0121] S403: According to the adjustment demand analysis result, the parameters of the recognition model are adjusted in consideration of processing efficiency and noise reduction effect, and an optimized configuration of noise reduction parameters is generated;

[0122] Implement parameter optimization strategies to adjust key parameters that affect model performance and improve processing efficiency and noise reduction effects. The steps involve applying automated tools to adjust parameters such as data processing depth, speed, and sensitivity, and verifying their impact on model performance through experiments. The optimized parameter configuration effectively improves the overall noise reduction capability of the system by reducing errors and increasing processing speed. The generated noise reduction parameter optimization configuration ensures that the system can maintain excellent performance standards under various operating environments.

[0123] See also Figure 6 , using the optimized configuration of noise reduction parameters, by analyzing the brightness and color distribution data of the input image, identifying the scene type and lighting conditions of the image, dynamically adjusting the threshold setting for noise removal and processing the image, the steps for obtaining the noise reduction result of the input image are as follows:

[0124] S501: Analyze input image data using noise reduction parameter optimization configuration, extract image brightness and color distribution data, and generate visual feature data;

[0125] The input image data is analyzed by image processing software, and the brightness and color distribution data of each pixel are extracted by applying color space conversion and brightness measurement technology, including conversion from RGB to HSV color space and histogram equalization method, to ensure that the most accurate visual features are extracted from images under various lighting and color conditions. The color and brightness data of each image are systematically recorded and analyzed, and the generated visual feature data contains key image attributes. The target attributes provide basic data for the next step of scene and lighting condition analysis.

[0126] S502: Based on the visual feature data, according to the respective features of brightness and color, identifying the scene type and lighting conditions in the image, and generating scene and lighting recognition results;

[0127] Machine learning techniques such as support vector machines and decision trees are used to analyze target visual feature data, identify scene types and current lighting conditions in images, compare visual features with data models of known scene types and lighting conditions, and perform pattern recognition and classification. Each image is accurately classified into the corresponding scene and lighting environment based on its brightness and color characteristics. The generated scene and lighting recognition results help the system understand the image content and environmental conditions more accurately.

[0128] S503: According to the scene and lighting recognition results, the noise removal threshold setting is adjusted according to the scene and lighting conditions, and the input image is subjected to noise reduction processing to generate an input image noise reduction result;

[0129] Dynamic threshold adjustment technology, such as the adaptive threshold adjustment method, is used to adjust the noise elimination strategy. By comparing the typical noise characteristics and image quality requirements in different scenes, the best noise reduction effect can be achieved under different lighting and environmental conditions. The target adjusted parameters are used in the actual image processing process to perform real-time noise reduction on the input image. The generated input image denoising result shows the optimized image quality, significantly reduces visual noise, and improves the overall visibility and practicality of the image.

[0130] See also Figure 7 , a convolutional neural network driven image denoising system, the convolutional neural network driven image denoising system is used to execute the above convolutional neural network driven image denoising method, the system comprises:

[0131] The sample classification module is based on the training sample image set. It evaluates the complexity of the image according to the edge density and texture richness of the sample image, classifies and screens the sample images, trains the denoising model, and generates an image noise recognition model.

[0132] The noise recognition module inputs the target processing image according to the image noise recognition model, calculates the error between the image and the reconstruction, identifies and marks the noise area, and adjusts the recognition threshold according to the noise characteristics of the image to generate noise feature record data;

[0133] The classification processing module uses noise feature recording data to analyze the identified image noise, and classifies the target noise by comparing the data features of known noise types, including Gaussian noise and salt and pepper noise, and matches the noise reduction processing method to generate the noise type classification result;

[0134] The parameter optimization module analyzes and adjusts multiple parameters of the recognition model based on the noise type classification results, by evaluating the processing effects of various noise types, and generates the optimized configuration of noise reduction parameters;

[0135] The scene adjustment module uses the optimized configuration of noise reduction parameters, analyzes the brightness and color distribution of the input image, identifies the scene type and lighting conditions, dynamically adjusts the noise removal threshold and processes the image to obtain the noise reduction result of the input image.

[0136] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0137] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0138] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0139] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0140] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0143] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0144] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0145] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0146] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A convolutional neural network driven image denoising method, characterized in that: The method comprises: S1: Based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained to generate an image noise recognition model; S2: using the image noise recognition model, inputting the target processed image, calculating the error between the image and the reconstructed image, comparing the error size and identifying the noise area in the image, and adjusting the noise recognition threshold according to the noise characteristics of the target image to generate noise feature record data; S3: Based on the noise feature recording data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type to generate a noise type classification result; S4: According to the noise type classification result, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating a noise reduction parameter optimization configuration; S5: Using the noise reduction parameter optimization configuration, by analyzing the brightness and color distribution data of the input image, identifying the scene type and lighting conditions of the image, dynamically adjusting the threshold setting for noise elimination and processing the image, to obtain the input image noise reduction result.

2. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: The image noise recognition model specifically includes edge density analysis results, texture richness assessment results, and sample image screening results. The noise feature record data includes reconstruction error calculation results, noise area identification results, and noise identification thresholds. The noise type classification results specifically refer to data feature comparison results, noise type identification records, and noise reduction processing method matching lists. The noise reduction parameter optimization configuration includes noise processing effect information, processing parameter adjustment records, and model parameter verification results. The input image denoising results specifically include image feature extraction records, scene type identification information, and noise elimination threshold configuration.

3. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: Based on the training sample image set, by calculating the edge density and texture richness of the sample image, evaluating the complexity of the image and classifying and screening the sample image, the denoising model is trained, and the steps of generating the image noise recognition model are as follows: S101: Based on the training sample image set, analyze and record the edge density and texture richness of multiple sample images to generate sample image feature data; S102: Calculating the complexity index of the image based on the sample image feature data, and classifying the image to generate classified image data; S103: Based on the classified image data, sample images are screened according to a preset complexity step, a training data set is optimized, and a noise reduction model is trained to generate an image noise recognition model.

4. The image denoising method driven by a convolutional neural network according to claim 3, characterized in that: The specific formula for calculating the complexity index of the image is: Among them, CI is the complexity index of the target sample image, i is the sequence number of the feature data, n is the total number of features, and w i is the weight of the i-th feature, x i is the measured value of the i-th feature, μ i is the mean of the i-th eigenvalue.

5. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: The steps of using the image noise recognition model, inputting the target processed image, calculating the error between the image and the reconstructed image, comparing the error size and identifying the noise area in the image, and adjusting the noise recognition threshold according to the noise characteristics of the target image to generate noise feature record data are as follows: S201: Based on the image noise recognition model, input a target processing image, calculate the error between the image and the model reconstructed image, and generate reconstruction error data; S202: Based on the reconstructed error data, by comparing errors, identifying and marking the noise area in the image, and generating a noise area mark; S203: According to the noise region mark and the noise characteristics of the target image, the noise recognition threshold is adjusted to optimize the recognition accuracy of the noise region, and noise feature recording data is generated.

6. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: Based on the noise feature recording data, the identified image noise is analyzed, the target noise is classified by comparing the data features of the target noise and the known noise types, and the noise reduction processing method is automatically mapped according to the noise type. The steps of generating the noise type classification result are specifically as follows: S301: Based on the noise feature recording data, analyzing the recognized image noise, extracting data features of the noise, including physical and visual characteristics of the noise, and generating noise feature information; S302: using the noise feature information, comparing the target noise with data features of multiple known noise types, calculating similarities, and generating similarity information; S303: Identify multiple noise types, including Gaussian noise and salt and pepper noise, based on the similarity information, and match noise reduction processing methods and parameter configurations according to the types to generate a noise type classification result.

7. The image denoising method driven by a convolutional neural network according to claim 6, characterized in that: The specific formula for calculating the similarity is: Among them, S pq represents the similarity value between the target noise p and the noise type q, m is the dimension of the feature data, and f pk is the value of the target noise p at the kth feature, f qk is the value of the kth feature of the known noise type q, p is the index of the target noise, q is the index of the known noise type, and k is the feature index.

8. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: According to the noise type classification result, by analyzing the processing effects of various noise types, evaluating the parameter types that need to be adjusted, adjusting the parameters of the recognition model, optimizing the noise reduction effect, and generating the noise reduction parameter optimization configuration steps are specifically as follows: S401: Based on the noise type classification result, evaluating the noise reduction effects of multiple noise types, and generating a processing effect analysis result; S402: According to the processing effect analysis result, identifying the recognition model parameters that need to be adjusted, including data processing depth, data processing speed, and data processing sensitivity, and generating an adjustment demand analysis result; S403: According to the adjustment demand analysis result, the parameters of the recognition model are adjusted in consideration of processing efficiency and noise reduction effect, and an optimized configuration of noise reduction parameters is generated.

9. The image denoising method driven by a convolutional neural network according to claim 1, characterized in that: Using the noise reduction parameter optimization configuration, by analyzing the brightness and color distribution data of the input image, identifying the scene type and lighting conditions of the image, dynamically adjusting the threshold setting of noise removal and processing the image, the steps of obtaining the noise reduction result of the input image are specifically as follows: S501: Analyze input image data using the noise reduction parameter optimization configuration, extract image brightness and color distribution data, and generate visual feature data; S502: Based on the visual feature data, identifying the scene type and lighting conditions in the image according to the respective features of brightness and color, and generating a scene and lighting recognition result; S503: According to the scene and lighting recognition result, the noise elimination threshold setting is adjusted according to the scene and lighting conditions, and the input image is subjected to noise reduction processing to generate an input image noise reduction result.

10. A convolutional neural network driven image denoising system, characterized in that: According to the convolutional neural network driven image denoising method according to any one of claims 1 to 9, the system comprises: The sample classification module is based on the training sample image set. It evaluates the complexity of the image according to the edge density and texture richness of the sample image, classifies and screens the sample images, trains the denoising model, and generates an image noise recognition model. The noise recognition module inputs the target processing image according to the image noise recognition model, calculates the error between the image and the reconstruction, recognizes and marks the noise area, adjusts the recognition threshold according to the noise characteristics of the image, and generates noise feature record data; The classification processing module uses the noise feature recording data to analyze the identified image noise, and classifies the target noise, including Gaussian noise and salt and pepper noise, by comparing the data features of known noise types, and matches the noise reduction processing method to generate a noise type classification result; The parameter optimization module generates a noise reduction parameter optimization configuration by evaluating the processing effects of various noise types according to the noise type classification results; The scene adjustment module uses the noise reduction parameter optimization configuration, analyzes the brightness and color distribution of the input image, identifies the scene type and lighting conditions, dynamically adjusts the noise elimination threshold and processes the image to obtain the input image noise reduction result.

Images