Personalized image processing method and system based on AI camera

Abstract

The invention provides a personalized image processing method and system based on an AI camera, and belongs to the technical field of cameras and image processing, and the method comprises the steps: carrying out the feature extraction of input original image data, carrying out the multi-dimensional analysis of texture density and color distribution, and obtaining a quantitative index value reflecting the complexity of image content; according to the quantitative index value, performing classification judgment by adopting a preset threshold interval, and if the quantitative index value exceeds a high-complexity interval, marking the image as a high-complexity category; if the detail loss risk is judged to be significant, dynamically adjusting the sharpening intensity and the denoising proportion in the image processing parameters to obtain an adjusted parameter configuration combination; and applying the adjusted parameter configuration combination to an image processing flow, and performing optimization processing on the original image data to obtain intermediate image data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of cameras and image processing technology, and particularly relates to an AI camera-based personalized image processing method and system. BACKGROUND

[0002] In the field of modern intelligent image processing, especially in the application of real-time processing of artificial intelligence cameras, it is crucial to ensure the balance between image quality and processing efficiency.

[0003] This field is not only the key to improving user experience, but also the core driving force for technological innovation of intelligent devices.

[0004] As the demand for image processing becomes increasingly complex, how to maintain high-quality image output in dynamic environments has become a key direction that needs to be broken through in the industry.

[0005] However, many current solutions often face the problem of insufficient adaptability when dealing with complex scenes.

[0006] Especially when the characteristics of image content change greatly, traditional processing mechanisms have difficulty in flexibly adjusting according to specific conditions, leading to loss of image details or decline in processing efficiency in some environments.

[0007] This limitation is not simply a technical parameter problem, but a lack of deep perception of the essential characteristics of images and dynamic response capabilities, especially when facing diversified shooting scenes, existing methods often cannot achieve precise matching.

[0008] In-depth, the core challenge in this field mainly focuses on how to accurately capture the complexity of image content and realize real-time optimization of processing strategies based on it.

[0009] The complexity of image content is often measured by specific quantitative indicators, and the fluctuation of this indicator will directly affect the selection of processing parameters.

[0010] If this fluctuation cannot be perceived in time, it may lead to a mismatch between parameter settings and actual needs, and thus cause problems such as image quality degradation or waste of system resources.

[0011] Further, the lack of this perception ability also makes the system lack self-adjustment mechanism in abnormal situations, making it difficult to deal with sudden processing deviations.

[0012] Therefore, how to dynamically analyze the complexity of image content in real-time processing, build a mechanism that can automatically adjust processing parameters, and ensure optimal image quality in various scenarios, has become a key problem in current research. SUMMARY

[0013] This invention provides a personalized image processing method based on an AI camera, mainly including:

[0014] By extracting features from the input raw image data and performing multi-dimensional analysis on texture density and color distribution, a quantitative index value reflecting the complexity of the image content is obtained.

[0015] Based on the quantification index value, a preset threshold range is used for classification and judgment. If the quantification index value exceeds the high complexity range, the image is marked as a high complexity category.

[0016] For images of the aforementioned highly complex category, detailed change information of local regions is extracted from their feature data. By analyzing the comparison between the local and the overall image, it is determined whether there is a risk of loss of detail.

[0017] If the risk of loss of detail is determined to be significant, the sharpening intensity and denoising ratio in the image processing parameters are dynamically adjusted to obtain the adjusted parameter configuration combination.

[0018] By applying the adjusted parameter configuration combination to the image processing flow, the original image data is optimized to obtain intermediate image data;

[0019] For the intermediate image data, the processing time and resource consumption are analyzed in real time. If the processing time exceeds the preset range, non-critical areas are compressed while retaining the high-quality output of critical areas to obtain the final processed image.

[0020] By comparing and analyzing the final processed image with the original image data, any abnormal deviations are detected. If a deviation exists, the complexity quantification index is recalculated and the processing parameter configuration is updated.

[0021] This invention provides a personalized image processing system based on an AI camera, mainly comprising:

[0022] The feature extraction module is used to extract features from the input raw image data, perform multi-dimensional analysis on texture density and color distribution, and obtain quantitative index values ​​that reflect the complexity of the image content.

[0023] The classification and judgment module is used to classify and judge the image according to the quantification index value using a preset threshold range. If the quantification index value exceeds the high complexity range, the image is marked as a high complexity category.

[0024] The detail analysis module is used to extract detailed change information of local areas from the feature data of the image of the high complexity category, and to determine whether there is a risk of loss of detail by analyzing the comparison between the local and the whole.

[0025] The parameter adjustment module is used to dynamically adjust the sharpening intensity and denoising ratio in the image processing parameters if the risk of loss of detail is determined to be significant, so as to obtain the adjusted parameter configuration combination.

[0026] The optimization processing module is used to optimize the original image data by applying the adjusted parameter configuration combination to the image processing flow to obtain intermediate image data;

[0027] The resource monitoring module is used to analyze the processing time and resource consumption of the intermediate image data in real time. If the processing time exceeds the preset range, the non-critical areas are compressed while the high-quality output of the critical areas is retained to obtain the final processed image.

[0028] The deviation detection module is used to detect whether there is any abnormal deviation by comparing and analyzing the final processed image with the original image data. If there is a deviation, the complexity quantification index is recalculated and the processing parameter configuration is updated.

[0029] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0030] This invention discloses a personalized image processing method and system based on an AI camera. It intelligently processes complex images by extracting features and performing multi-dimensional analysis on the input image to derive quantitative index values ​​to assess the image's complexity. For highly complex images, this invention further analyzes local detail changes, assesses the risk of detail loss, and dynamically adjusts processing parameters. During image processing, this invention monitors time consumption and resource usage in real time, compressing non-critical areas when necessary to improve efficiency. Finally, through comparative analysis, processing deviations are detected and corrected to ensure output quality. The technical advantage of this invention lies in its ability to intelligently identify and optimize the processing of complex images, improving processing efficiency while maintaining image quality, making it particularly suitable for image processing scenarios requiring high-quality output. Attached Figure Description

[0031] Figure 1 This is a flowchart of the AI ​​camera-based personalized image processing method of the present invention.

[0032] Figure 2 This is a schematic diagram of the AI ​​camera-based personalized image processing method of the present invention.

[0033] Figure 3 This is another schematic diagram of the AI ​​camera-based personalized image processing method of the present invention.

[0034] Figure 4 This is a schematic diagram of the structure of the AI ​​camera-based personalized image processing system of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] like Figure 1 , Figure 2 and Figure 3 As shown, the personalized image processing method based on an AI camera in this embodiment may specifically include:

[0037] Step S101: By extracting features from the input raw image data and performing multi-dimensional analysis on texture density and color distribution, a quantitative index value reflecting the complexity of the image content is obtained.

[0038] The original image data is preprocessed, and denoising techniques are used to smooth the image, resulting in a first processed image. Based on this first processed image, texture density features are extracted, and a pre-defined texture analysis method is used to calculate texture density distribution parameters, yielding texture feature values. Color distribution analysis is performed on the first processed image, using color histogram technology to extract color distribution features and determine color distribution parameters. Multi-dimensional analysis is then conducted on the texture feature values ​​and color distribution parameters, fusing multiple feature dimensions. If the texture feature values ​​exceed a pre-defined threshold range, the texture density distribution is weighted and adjusted to determine the initial level of content complexity. Based on the initial level of content complexity, and combined with the feature weights in the analysis dimensions, a support vector machine algorithm is used to classify the multi-dimensional features, obtaining classification results. By quantizing and mapping the classification results, and combining them with a pre-defined index system, a quantitative index value representing content complexity is calculated, determining the final complexity assessment result.

[0039] Specifically, for the quantitative analysis of image content complexity, the first step is to extract features from the input raw image data. The Gray-Level Co-occurrence Matrix (GLCM) algorithm can be used to analyze texture density. The specific method involves converting the image to grayscale. Assuming an image resolution of 512x512 pixels, the distance parameter is set to 1, and the angles are 0 degrees, 45 degrees, 90 degrees, and 135 degrees when calculating the GLCM. Four feature values ​​are extracted: contrast, correlation, energy, and uniformity. For example, the calculated contrast value is 15.3, the correlation value is 0.85, the energy value is 0.12, and the uniformity value is 0.9. These values ​​reflect the roughness and regularity of the texture; higher values ​​indicate a more complex texture. Next, for multi-dimensional analysis of color distribution, HSV color space statistical histograms can be used. Assuming the image is divided into 8x8 grid regions, the mean and variance of hue (H), saturation (S), and lightness (V) distributions for each region are calculated. For example, in a certain region, the mean H is 30, the variance of S is 12.5, and the mean V is 180, indicating the uniformity and variation of color distribution. These statistical values ​​can quantify color complexity. Subsequently, the feature values ​​of texture density and color distribution are normalized. Assuming the normalized score for texture features is 0.75 and the normalized score for color features is 0.65, a weighted fusion method (with weights of 0.6 and 0.4 respectively) is used to calculate the final complexity index value: 0.75 x 0.6 + 0.65 x 0.4 = 0.71. This value serves as a quantitative result of the image content complexity; a higher value indicates greater content complexity. To form a logical chain, this index value can be associated with an image classification task. If the complexity exceeds 0.7, a more refined deep learning model will be automatically triggered for secondary analysis to ensure classification accuracy. The above process is all implemented automatically by the algorithm. Texture and color feature extraction depends on preset parameters, fusion calculation is based on fixed weights, and index output and subsequent task triggering are all completed automatically by the system, reflecting a complete thought chain from feature extraction to complexity quantification.

[0040] Step S102: Based on the quantization index value, a preset threshold range is used for classification judgment. If the quantization index value exceeds the high complexity range, the image is marked as a high complexity category.

[0041] By initially processing the quantitative indicators and using a preset interval division method, the indicator data is segmented and organized to obtain preliminary interval distribution results. Based on the preliminary interval distribution results, a classification judgment logic is implemented for the quantitative indicator data within each interval segment. If a segment of indicator data exceeds a preset threshold interval, it is marked as a potentially high-complexity value, thus determining the potential complexity range. A secondary analysis is performed on the potentially high-complexity values, using a support vector machine algorithm to distinguish features in the indicator data and obtain the complexity distribution after feature distinction. Based on the complexity distribution after feature distinction, the marking method is adjusted for the high-complexity value portion of the distribution. If the distribution value meets the high-complexity category standard, its associated image data is marked to obtain the marked image category. The marked image categories are verified by comparing the marking results using a preset category distinction rule. If the marking results are inconsistent with the preset rule, the marking criteria are readjusted to determine the final complexity category assignment. Based on the final complexity category assignment, the data is integrated to obtain the integrated classification data. The completeness of the classification data is judged to obtain the final image complexity classification result. By storing the final image complexity classification results using a structured storage method, the classification data is associated with the original quantification indicators to determine the data index after storage.

[0042] Specifically, in the quantitative analysis of image content complexity, edge features are first extracted from the input image data using the Canny edge detection algorithm. Assuming an image resolution of 1024x1024 pixels, with a low threshold of 50 and a high threshold of 150, the calculated edge pixel ratio is 18.2%. Furthermore, the entropy value of the edge direction distribution is calculated to be 3.5. A higher entropy value indicates a more disordered edge distribution, reflecting the complexity of the image structure. Next, based on the quantitative results of the edge features, the local contrast distribution of the image is calculated. The image is divided into 16x16 sub-blocks, and the standard deviation of brightness for each sub-block is calculated. Assuming a standard deviation of 22.8 for a certain sub-block indicates significant brightness variation in the local area, the average of the standard deviations of all sub-blocks yields a global contrast index of 20.5, which is used to further assess complexity. Subsequently, the edge feature entropy value and global contrast index are standardized. Assuming the standardized edge entropy value is 0.68 and the standardized contrast index is 0.72, the comprehensive complexity index is calculated using a preset weighting formula (weights of 0.55 and 0.45 respectively): 0.68 x 0.55 + 0.72 x 0.45 = 0.698. Finally, the system performs classification based on a preset threshold range. Assuming the high complexity range is greater than 0.65, if the calculated comprehensive complexity index of 0.698 exceeds this range, the image is automatically labeled as high complexity, and the labeling result is associated with subsequent image segmentation tasks. If labeled as high complexity, the system automatically adjusts the segmentation algorithm parameters to optimize processing accuracy. All of the above processes are completed automatically by the algorithm. Feature extraction is based on preset thresholds, index calculation relies on fixed formulas, and classification labeling and task association are driven by system rules, forming a complete logical chain from feature extraction to classification judgment.

[0043] Step S103: For the image of the highly complex category, extract the detail change information of the local area from its feature data, and determine whether there is a risk of loss of detail by analyzing the comparison difference between the local and the whole.

[0044] Image feature data is obtained from highly complex images. Local region information within the feature data is separated using a pre-defined segmentation method, resulting in a set of separated local features. Based on this set, the detail variations in each local region are quantitatively analyzed using statistical tools to measure and determine their distribution range. If the distribution range exceeds a pre-defined threshold, the corresponding local region is labeled, resulting in a labeled local region dataset. The labeled local region dataset is then compared with the overall image characteristics using a difference comparison analysis method to calculate the degree of detail difference between the local region and the overall image, determining if a significant difference exists. If the calculated difference exceeds a pre-defined threshold, detail loss is assessed for the corresponding local region, identifying potential loss areas. Based on these potential loss areas, the image data processing flow is adjusted, and a support vector machine algorithm is used to reconstruct the features of the lost detail areas, yielding reconstructed image detail data. The reconstructed image detail data is then validated through a secondary comparison of the local-to-overall comparison results. If the comparison shows that detail recovery meets pre-defined standards, the final image detail analysis conclusion is determined.

[0045] Specifically, for images of highly complex categories, the system first extracts detailed variation information of local regions from their feature data. Assuming the image resolution is 2048x2048 pixels, the algorithm divides the image into 32x32 sub-regions, calculates the grayscale gradient value for each sub-region, and processes it using the Sobel operator, resulting in a gradient mean of 12.5 for each sub-region. The gradient value reflects the intensity change of local details. Next, the system analyzes the contrast difference between the local and the overall image, comparing the gradient mean of all sub-regions with the global gradient mean of the entire image (assumed to be 10.8), and calculating the difference ratio between the local and the overall image.

[0046] 12.5 / 10.8 = 1.16. When the difference ratio is greater than 1.1, the system marks the sub-region as a region with significant detail. Subsequently, the system further calculates the proportion of regions with significant detail, assuming a proportion of 25.6%, and combines this with the image's texture density index (calculated using a Gabor filter, assuming a density value of 0.82) to comprehensively assess the risk of detail loss. A risk threshold is set at a proportion greater than 20% and a density value greater than 0.75. If these conditions are met, the system automatically generates a risk warning label, marking the image as having a risk of detail loss. Finally, the system associates the risk warning label with subsequent image compression tasks. If a risk exists, the system automatically adjusts the parameters of the compression algorithm, reducing the compression ratio to protect detail information. All of the above processes are driven by the system algorithm. Feature extraction is based on preset rules, difference analysis relies on numerical calculations, risk judgment is achieved through threshold comparison, and task association is automatically completed by system logic, forming a complete chain from detail extraction to risk assessment.

[0047] Step S104: If the risk of loss of detail is determined to be significant, the sharpening intensity and denoising ratio in the image processing parameters are dynamically adjusted to obtain the adjusted parameter configuration combination.

[0048] If the risk of image detail loss is deemed significant through evaluation, a preliminary analysis of the sharpening intensity and denoising ratio in image processing is conducted to identify shortcomings in the current parameter configuration and determine initial parameter adjustment directions. Based on these initial adjustment directions, the sharpening intensity and denoising ratio are dynamically adjusted using a preset threshold range to obtain the adjusted parameter configuration combination. The adjusted parameter configuration combination is then validated to detect the recovery of image details. If the detection results show that detail recovery does not meet the preset standard, the parameter configuration is adjusted a second time to determine a new configuration combination. Based on the new configuration combination, the image processing workflow is optimized, and optimized image data is obtained to determine whether the risk of detail loss has been mitigated. If the risk of detail loss is still deemed significant, a support vector machine algorithm is used to extract features from image details, identifying potential areas of missing detail and determining key processing targets. Local enhancement processing is then performed on these key processing targets to supplement data in areas of missing image details, obtaining enhanced image data and determining whether the detail recovery meets the preset standard. Based on the enhanced image data, a final verification of the overall image processing effect is conducted. If the verification results show that the risk assessment has been met, the final parameter configuration and processing strategy are determined.

[0049] Specifically, to significantly assess the risk of detail loss in image processing, the system first automatically detects the detail distribution characteristics of the image using an algorithm. Assuming the input image resolution is 4096x4096 pixels, the system uses the Laplacian operator to perform edge detection, calculating the edge intensity value for each 64x64 pixel region, obtaining an average intensity value of 8.3, reflecting the detail clarity of local areas. Next, the system compares this value with a preset detail risk threshold of 7.5. If it exceeds the threshold, it is determined that there is a significant risk of detail loss, and the parameter adjustment process is automatically triggered. Subsequently, the system dynamically adjusts the sharpening intensity parameter. The initial sharpening intensity is 1.0. Based on the ratio of the edge intensity value to the threshold (8.3 / 7.5 = 1.11), the system increases the sharpening intensity to 1.2 according to preset rules to enhance detail representation. Simultaneously, the system analyzes the noise level of the image, calculates the noise distribution using a wavelet transform algorithm, assumes a noise mean of 0.65, and, combined with a preset denoising ratio of 0.5, adjusts the denoising ratio to 0.7 based on the ratio 0.65 / 0.5 = 1.3, to balance the effects of detail enhancement and noise suppression. Finally, the adjusted parameter configuration combination (sharpening intensity 1.2, denoising ratio 0.7) is automatically applied to the image processing pipeline and associated with subsequent image storage tasks, ensuring that the processed image preferentially uses lossless encoding during storage format conversion to preserve the adjusted details. The entire process is driven by the system algorithm, with parameter adjustments based on numerical analysis and preset logic, forming a complete automated chain from risk detection to parameter optimization.

[0050] Step S105: By applying the adjusted parameter configuration combination to the image processing flow, the original image data is optimized to obtain intermediate image data.

[0051] By loading the adjusted parameter configuration into the image processing workflow, preliminary optimization is performed on the original image data to obtain an intermediate processed image. Based on the characteristics of the intermediate processed image, the sharpness of the image data is evaluated using a preset threshold range to determine if it meets predetermined standards. If the evaluation result shows that the sharpness of the intermediate processed image does not meet the standards, the parameter configuration in the processing workflow is fine-tuned to obtain an updated adjustment combination. By applying the updated adjustment combination to the processing workflow, a second optimization process is performed on the intermediate processed image to obtain fine-tuned image data. Based on the performance of the fine-tuned image data, a support vector machine algorithm is used to analyze the local features of the image data to determine whether there are areas of missing details. If the analysis result shows that there are areas of missing details, local data augmentation processing is performed on these areas to obtain the final optimized image. By detecting the overall features of the final optimized image, it is determined whether the processing result meets the predetermined target, and the final processing workflow configuration is determined.

[0052] Specifically, in the image processing workflow, the system first acquires the raw image data. Assuming the input image resolution is 3840x2160 pixels, the system uses an automated image analysis module to extract features from the image using the difference of Gaussians algorithm, calculating the texture density of each 32x32 pixel region, obtaining an average texture density value of 6.8, used to assess the image's complexity. Next, the system automatically loads adjusted parameter configurations, such as setting the contrast enhancement coefficient to 1.15 and the brightness adjustment value to 0.9, into the processing engine. A convolutional neural network model processes the image pixels layer by layer, calculating the weighted average of the surrounding 8x8 region for each pixel to ensure uniformity after parameter application, generating intermediate image data. Simultaneously, the system performs quality checks on the intermediate image data, using a peak signal-to-noise ratio (PSNR) algorithm to calculate the difference between the processed and unprocessed images. Assuming the obtained value is 38.5 dB, it is compared with a preset threshold of 35.0 dB to confirm that the processing meets the standard. If the difference value is lower than the threshold, the system automatically triggers a secondary optimization process, adjusting the saturation gain in the parameter configuration to 1.1 and recalculating. To form a complete logical chain, the system also associates intermediate image data with subsequent color correction tasks. Through an automated tone mapping algorithm, it analyzes the color gamut distribution of the image to ensure that the intermediate image data after parameter optimization remains consistent in subsequent processing. The entire process is driven by the system algorithm and requires no manual intervention.

[0053] Step S106: For the intermediate image data, analyze its processing time and resource consumption in real time. If the processing time exceeds the preset range, compress the non-critical areas and retain the high-quality output of the critical areas to obtain the final processed image.

[0054] By monitoring the processing time of intermediate images in real time and obtaining their operational status under the current resource allocation, it is determined whether there are delays exceeding the preset range. If the monitoring results show that the processing time exceeds the preset range, compression operations are performed on non-critical areas to adjust the data processing strategy and obtain an optimized region division scheme. According to the region division scheme, a priority protection strategy is implemented for the data in critical areas to preserve their high-quality characteristics, resulting in initially adjusted image content. A preset threshold range is used to perform resource allocation detection on the initially adjusted image content to determine if there are still abnormal fluctuations in processing time. If the detection results show that the processing time is still abnormal, a second adjustment is performed on the compression operation of non-critical areas to obtain further optimized image data. For the further optimized image data, the high-quality characteristics of critical areas are verified using a support vector machine algorithm to determine whether they meet the predetermined standards. Based on the verification results, the overall data processing effect of the final image is recorded, resulting in a complete processing flow log.

[0055] Specifically, in the image processing workflow, the system performs real-time analysis of intermediate image data. First, the built-in performance monitoring module records the processing time. Assuming a processing session takes 2.3 seconds, while the preset time range is within 1.5 seconds, resource usage is monitored, revealing a memory usage rate of 78%, exceeding the preset threshold of 60%. Based on this data, the system automatically initiates an optimization strategy, using a region segmentation algorithm to divide the image into critical and non-critical regions. Assuming the critical region accounts for 40%, mainly concentrated in the image center and high-detail areas, while the non-critical region accounts for 60%, the system then applies a dynamic compression algorithm to the non-critical regions, reducing their resolution from the original 1920x1080 to 960x540, a compression ratio of 50%, to reduce computational load, while the critical regions maintain their original resolution. Next, the system recalculates the processing time using a resource allocation model, obtaining an optimized processing time of 1.4 seconds, within the preset range, and the memory usage rate also drops to 55%, below the threshold. The entire process forms a closed-loop logic. If subsequent detections indicate that the processing time or resource consumption exceeds the limit again, the system will automatically adjust the compression ratio or call a backup processing thread to ensure a balance between output quality and efficiency, ultimately generating the processed image data. To further improve the logic chain, the system associates the processing results with subsequent storage optimization tasks. Through an automated data tiering algorithm, critical area data is preferentially stored in a high-speed cache, while non-critical area data is stored in a low-speed storage area to improve the overall system response speed.

[0056] Step S107: By comparing and analyzing the final processed image with the original image data, it is possible to detect whether there is any abnormal deviation. If there is a deviation, the complexity quantification index is recalculated and the processing parameter configuration is updated.

[0057] By comparing the final image with the original data, the differences between the two are identified to determine if significant deviations exist. If the comparison analysis shows significant deviations, further detection is performed on the data differences, using a preset threshold range to determine if the deviation exceeds the normal range, thus obtaining the deviation detection results. Based on the deviation detection results, the complexity is evaluated, and corresponding quantitative indicators are obtained to determine their impact on subsequent processing. Through analysis of the quantitative indicators, processing parameters are adjusted to obtain an optimized parameter configuration scheme. If the adjusted parameter configuration scheme still does not eliminate significant deviations, a support vector machine algorithm is used for secondary verification of image quality to determine the specific distribution characteristics of the data differences. Based on the distribution characteristics of the data differences, local correction operations are performed on the final image to obtain the adjusted image content. By comparing the adjusted image content with the original data again, it is determined whether data differences still exist, thus obtaining the final comparison result.

[0058] Specifically, in the image processing workflow, the system first compares and analyzes the final processed image with the original image data. Using a pixel-level difference detection algorithm, it calculates the deviation values ​​of the two images in brightness, contrast, and color distribution. Assuming a brightness deviation of 3.5% and a contrast deviation of 2.8%, while the preset normal deviation range is brightness not exceeding 2.0% and contrast not exceeding 1.5%, the system determines that there is an abnormal deviation. Next, the system automatically triggers a recalculation of the complexity quantification index. Using an entropy-based image complexity analysis algorithm, the system finds the complexity index of the original image to be 7.2 (out of 10), while the complexity index of the final processed image decreases to 5.8, indicating that some detailed information may have been lost during processing. Subsequently, the system updates the processing parameter configuration based on the new complexity index. Through adaptive parameter adjustment of the model, the image enhancement coefficient is increased from the default 1.2 to 1.5, while the kernel size of the smoothing filter is reduced from 5x5 to 3x3 to retain more detailed features. The entire process forms a closed-loop logic. If subsequent comparative analysis still reveals discrepancies, the system will further link it to an image content classification task, using a deep learning model to perform semantic analysis on the image content, automatically identifying the main object regions and prioritizing the optimization of their parameters to ensure processing accuracy. All of the above steps are completed automatically by the system, driven by algorithms and models, ensuring rigorous logic and seamless transitions.

[0059] like Figure 4 As shown, this invention provides a personalized image processing system based on an AI camera, mainly comprising:

[0060] The feature extraction module is used to extract features from the input raw image data, perform multi-dimensional analysis on texture density and color distribution, and obtain quantitative index values ​​that reflect the complexity of the image content.

[0061] The classification and judgment module is used to classify and judge the image according to the quantification index value using a preset threshold range. If the quantification index value exceeds the high complexity range, the image is marked as a high complexity category.

[0062] The detail analysis module is used to extract detailed change information of local areas from the feature data of the image of the high complexity category, and to determine whether there is a risk of loss of detail by analyzing the comparison between the local and the whole.

[0063] The parameter adjustment module is used to dynamically adjust the sharpening intensity and denoising ratio in the image processing parameters if the risk of loss of detail is determined to be significant, so as to obtain the adjusted parameter configuration combination.

[0064] The optimization processing module is used to optimize the original image data by applying the adjusted parameter configuration combination to the image processing flow to obtain intermediate image data;

[0065] The resource monitoring module is used to analyze the processing time and resource consumption of the intermediate image data in real time. If the processing time exceeds the preset range, the non-critical areas are compressed while the high-quality output of the critical areas is retained to obtain the final processed image.

[0066] The deviation detection module is used to detect whether there is any abnormal deviation by comparing and analyzing the final processed image with the original image data. If there is a deviation, the complexity quantification index is recalculated and the processing parameter configuration is updated.

[0067] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A personalized image processing method based on an AI camera, characterized in that, The method includes: By extracting features from the input raw image data and performing multi-dimensional analysis on texture density and color distribution, a quantitative index value reflecting the complexity of the image content is obtained. Based on the quantification index value, a preset threshold range is used for classification and judgment. If the quantification index value exceeds the high complexity range, the image is marked as a high complexity category. For images of the aforementioned highly complex category, detailed change information of local regions is extracted from their feature data. By analyzing the comparison between the local and the overall image, it is determined whether there is a risk of loss of detail. If the risk of loss of detail is determined to be significant, the sharpening intensity and denoising ratio in the image processing parameters are dynamically adjusted to obtain the adjusted parameter configuration combination. By applying the adjusted parameter configuration combination to the image processing flow, the original image data is optimized to obtain intermediate image data; For the intermediate image data, the processing time and resource consumption are analyzed in real time. If the processing time exceeds the preset range, non-critical areas are compressed while retaining the high-quality output of critical areas to obtain the final processed image. By comparing and analyzing the final processed image with the original image data, any abnormal deviations are detected. If a deviation exists, the complexity quantification index is recalculated and the processing parameter configuration is updated.

2. The personalized image processing method based on an AI camera according to claim 1, characterized in that, The process involves feature extraction from the input raw image data, multi-dimensional analysis of texture density and color distribution, to obtain quantitative index values ​​reflecting the complexity of the image content, including: The first processed image is obtained by preprocessing the original image data and smoothing the image using denoising techniques. Based on the first processed image, texture density features are extracted, and texture density distribution parameters are calculated using a preset texture analysis method to obtain texture feature values. By performing color distribution analysis on the first processed image and using color histogram technology, color distribution features are extracted and color distribution parameters are determined. Multi-dimensional analysis is performed on texture feature values ​​and color distribution parameters, and multiple feature dimensions are integrated. If the texture feature value exceeds the preset threshold range, the texture density distribution is weighted and adjusted to determine the preliminary level of content complexity. Based on the initial level of content complexity and the feature weights in the analysis dimensions, the support vector machine algorithm is used to classify the multidimensional features and obtain the classification results. By quantifying and mapping the classification results and combining them with a pre-defined index system, a quantitative index value representing the complexity of the content is calculated, and the final complexity assessment result is determined.

3. The personalized image processing method based on an AI camera according to claim 1, characterized in that, The step of classifying the image based on the quantized index value using a preset threshold range, and marking the image as a high-complexity category if the quantized index value exceeds the high-complexity range, includes: By performing preliminary processing on the quantitative indicators and using a preset interval division method, the indicator data is segmented and organized to obtain preliminary interval distribution results. Based on the preliminary interval distribution results, a classification judgment logic is implemented for the quantitative indicator data within each interval segment. If a certain segment of indicator data exceeds the preset threshold range, it is marked as a potentially high-complexity value, and the potential complexity range is determined. By performing secondary analysis on potentially highly complex values, the support vector machine algorithm is used to distinguish the features of the index data and obtain the complexity distribution after feature distinction. Based on the complexity distribution after feature differentiation, the labeling method is adjusted for the high complexity value part of the distribution. If the distribution value meets the high complexity category standard, the associated image data is labeled to obtain the labeled image category. The labeled image categories are verified by comparing the labeling results with the preset category distinction rules. If the labeling results are inconsistent with the preset rules, the labeling criteria are readjusted to determine the final complex category assignment. Based on the final complex category assignment, the data is integrated according to the assignment results, the integrated classification data is obtained, the completeness of the classification data is judged, and the final image complexity classification result is obtained. By storing the final image complexity classification results using a structured storage method, the classification data is associated with the original quantification indicators to determine the data index after storage.

4. The personalized image processing method based on an AI camera according to claim 1, characterized in that, For images of the highly complex category, extracting detailed change information of local regions from their feature data, and determining whether there is a risk of detail loss by analyzing the comparison between the local and the overall image, including: By obtaining image feature data from images of highly complex categories, separating local region information in the feature data, and using a preset segmentation method to divide the local region information, a set of separated local features is obtained. Based on the separated local feature set, the detailed change information of each local region is quantitatively analyzed, and statistical tools are used to measure the detailed change information to determine the distribution range of the detailed change. If the distribution range of the detailed changes exceeds the preset threshold range, the corresponding local area information is marked to obtain the marked local area dataset. By comparing and analyzing the characteristics of the labeled local region dataset with the overall image, the difference comparison analysis method is used to calculate the degree of difference in details between the local region and the overall image, and to determine whether there are significant differences. If the calculated difference exceeds the preset difference threshold, then the corresponding local area information is judged for loss of detail to identify potential areas of loss of detail. Based on the potential areas of lost detail, the image data processing flow is adjusted, and the support vector machine algorithm is used to reconstruct the features of the areas of lost detail to obtain the reconstructed image detail data. By verifying the reconstructed image detail data, a second comparison is performed on the results of the local-to-overall comparison. If the comparison results show that the detail recovery meets the preset standards, the final image detail analysis conclusion is determined.

5. The personalized image processing method based on an AI camera according to claim 1, characterized in that, If the risk of detail loss is determined to be significant, the sharpening intensity and denoising ratio in the image processing parameters are dynamically adjusted to obtain an adjusted parameter configuration combination, including: If the risk of losing image details is determined to be significant through evaluation, a preliminary analysis is conducted on the sharpening intensity and denoising ratio in image processing to identify the shortcomings of the current parameter configuration and obtain a preliminary direction for parameter adjustment. Based on the initial parameter adjustment direction, the sharpening intensity and noise reduction ratio are dynamically adjusted using a preset threshold range to obtain the adjusted parameter configuration combination; The adjusted parameter configuration combination is verified, and the recovery of image details is detected. If the detection result shows that the detail recovery does not meet the preset standard, the parameter configuration is adjusted a second time to determine a new configuration combination. Based on the new configuration combination, the image processing workflow is optimized and adjusted to obtain optimized image data and determine whether the risk of detail loss has been mitigated. If the risk of losing details is still judged to be significant, the support vector machine algorithm is used to extract features from the image details, thereby obtaining potential areas of missing details and determining key processing targets. By performing local enhancement processing on key processing targets, data is supplemented for areas with missing image details, enhanced image data is obtained, and it is determined whether the detail restoration meets the preset standards. Based on the enhanced image data, a final verification of the overall image processing effect is conducted. If the verification results show that the risk assessment has been met, the final parameter configuration and processing strategy are determined.

6. The personalized image processing method based on an AI camera according to claim 1, characterized in that, The process of optimizing the original image data by applying the adjusted parameter configuration combination to the image processing flow to obtain intermediate image data includes: By loading the adjusted parameter configuration into the image processing flow, preliminary optimization operations are performed on the original image data to obtain the intermediate processed image; Based on the characteristics of the intermediate processed image, the clarity of the image data is evaluated using a preset threshold range to determine whether it meets the predetermined standard. If the evaluation results show that the clarity of the intermediate processed image does not meet the standard, the parameter configuration in the processing flow is fine-tuned to obtain the updated adjustment combination; By applying the updated adjustment combination to the processing flow, the intermediate processed images are subjected to secondary optimization to obtain fine-tuned image data; Based on the performance of the fine-tuned image data, the support vector machine algorithm is used to analyze the local features of the image data to determine whether there are areas with missing details. If the analysis results show that there are areas with missing details, local data augmentation processing is performed on those areas to obtain the final optimized image; By detecting the overall features of the final optimized image, it is determined whether the processing result has achieved the predetermined goal, and the final processing flow configuration is determined.

7. The personalized image processing method based on an AI camera according to claim 1, characterized in that, The process involves real-time analysis of the processing time and resource consumption of the intermediate image data. If the processing time exceeds a preset range, non-critical areas are compressed while retaining high-quality output of critical areas to obtain the final processed image. This includes: By monitoring the processing time of intermediate images in real time, we can obtain their running status under the current resource allocation and determine whether there is a delay exceeding the preset range. If the monitoring results show that the processing time exceeds the preset range, then a compression operation is performed on non-critical areas, the data processing strategy is adjusted, and an optimized area division scheme is obtained. Based on the regional division scheme, a priority protection strategy is implemented for the data in key areas to preserve their high-quality characteristics, resulting in the initially adjusted image content. The resource allocation detection of the initially adjusted image content is performed using a preset threshold range to determine whether there are still abnormal fluctuations in processing time. If the detection results show that the processing time is still abnormal, the compression operation of non-critical areas will be adjusted again to obtain further optimized image data. For the further optimized image data, the support vector machine algorithm is used to verify the high-quality characteristics of key regions and determine whether they meet the predetermined standards. Based on the verification results, the overall data processing effect of the final image is recorded, resulting in a complete processing log.

8. The personalized image processing method based on an AI camera according to claim 1, characterized in that, The process involves comparing the final processed image with the original image data to detect any abnormal deviations. If a deviation exists, the complexity quantification index is recalculated, and the processing parameter configuration is updated. This includes: By performing comparative analysis between the final image and the original data, the data differences between the two are obtained, and it is determined whether there is a significant deviation. If the comparison analysis results show significant deviations, further detection will be carried out on the data differences. A preset threshold range will be used to determine whether the deviation exceeds the normal range, and the deviation detection results will be obtained. Based on the deviation detection results, the complexity is assessed, corresponding quantitative indicators are obtained, and the degree of impact on subsequent processing is determined. By analyzing quantitative indicators, processing parameters are adjusted to obtain an optimized parameter configuration scheme. If the adjusted parameter configuration scheme still fails to eliminate significant deviations, the support vector machine algorithm is used to perform a secondary verification of image quality to determine the specific distribution characteristics of data differences. Based on the distribution characteristics of the data differences, a local correction operation is performed on the final image to obtain the adjusted image content; By comparing the adjusted image content with the original data again, it is determined whether there are still data differences, and the final comparison result is obtained.

9. A personalized image processing system based on an AI camera, characterized in that, The system includes: The feature extraction module is used to extract features from the input raw image data, perform multi-dimensional analysis on texture density and color distribution, and obtain quantitative index values ​​that reflect the complexity of the image content. The classification and judgment module is used to classify and judge the image according to the quantification index value using a preset threshold range. If the quantification index value exceeds the high complexity range, the image is marked as a high complexity category. The detail analysis module is used to extract detailed change information of local regions from the feature data of the image of the high complexity category, and to determine whether there is a risk of loss of detail by analyzing the comparison between the local and the whole. The parameter adjustment module is used to dynamically adjust the sharpening intensity and denoising ratio in the image processing parameters if the risk of loss of detail is determined to be significant, so as to obtain the adjusted parameter configuration combination. The optimization processing module is used to optimize the original image data by applying the adjusted parameter configuration combination to the image processing flow to obtain intermediate image data; The resource monitoring module is used to analyze the processing time and resource consumption of the intermediate image data in real time. If the processing time exceeds the preset range, the non-critical areas are compressed while the high-quality output of the critical areas is retained to obtain the final processed image. The deviation detection module is used to detect whether there is any abnormal deviation by comparing and analyzing the final processed image with the original image data. If there is a deviation, the complexity quantification index is recalculated and the processing parameter configuration is updated.

Images