Image processing method and system based on AI identification
By combining deep convolutional neural networks and semantic segmentation algorithms, the problem of insufficient adaptability of traditional image processing methods has been solved, and efficient and accurate multi-task image processing has been achieved. It is suitable for security monitoring, autonomous driving, medical imaging, industrial inspection and other fields.
Patent Information
- Application Number
- CN202510648878.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-20
- Publication Date
- 2025-09-12
AI Technical Summary
Traditional image processing methods have poor adaptability to image data, making it difficult to achieve universal processing of image data from different sources and imaging conditions. They also lack a complete solution for multi-task integration, which increases system complexity and computational costs.
The image processing method based on deep convolutional neural network and semantic segmentation algorithm is adopted, including preprocessing, feature extraction, classification, segmentation and labeling. The deep learning model is used to automatically extract deep features of the image, and the semantic segmentation algorithm is combined to accurately divide the semantic areas. The modules work together to achieve pipeline processing.
It improves the accuracy and adaptability of image processing, reduces system complexity, reduces manual intervention, meets real-time processing needs, and provides efficient and accurate solutions for applications in multiple fields.
Smart Images
Figure CN120635529A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to an image processing method and system based on AI recognition. Background Art
[0002] The rapid development of digital imaging technology has led to increasingly widespread and complex applications of images in numerous fields, including security surveillance, autonomous driving, medical imaging, and industrial inspection. However, traditional image processing methods have many shortcomings when dealing with these complex scenarios, making it difficult to meet the growing demand for high-quality image processing.
[0003] Traditional image processing methods rely primarily on manually designed feature extraction algorithms and fixed image processing pipelines. For example, in image classification tasks, it is often necessary to manually extract image features such as color, texture, and shape, and then use these features for classification. However, manually extracted features often have limitations and are difficult to fully and accurately represent the semantic content of the image. For object detection tasks, traditional methods typically use sliding windows or region-based candidate box generation methods, combined with manual features and classifiers for object detection. These methods are prone to missed detections and false detections when dealing with complex backgrounds, large variations in object scale, and diverse object shapes. In image segmentation tasks, traditional segmentation algorithms mainly perform segmentation based on low-level features such as grayscale, color, and texture. This makes it difficult to accurately divide different semantic regions in the image, especially in cases where boundaries are blurred or there is occlusion, and the segmentation results are often less than ideal.
[0004] Currently, traditional image processing methods have poor adaptability to image data. Images from different sources and under different imaging conditions have different characteristics, such as lighting conditions, noise levels, and resolution. Traditional methods often require parameter adjustment and optimization for specific datasets or application scenarios, making it difficult to achieve universal processing of diverse image data. For example, images captured in outdoor scenes and indoor scenes have significantly different lighting conditions and noise characteristics. Traditional image processing methods may require the design of different processing flows and parameter settings to achieve optimal processing results, which increases system complexity and maintenance costs.
[0005] With the rise of deep learning technology, image processing methods based on artificial intelligence have gradually become a research hotspot. These methods can automatically learn deep feature representations in images by constructing deep neural network models, and have stronger feature extraction capabilities and adaptability. However, existing AI-based image processing methods mostly focus on a single task, such as simple image classification, target detection or image segmentation, and lack a complete solution that can organically combine multiple image processing tasks. In practical applications, users often need to perform multiple processing operations on images. For example, in security monitoring systems, it is necessary not only to detect and recognize targets in monitoring images, but also to segment and label target areas in order to more intuitively display the monitoring results. Existing methods usually need to call different models and algorithms separately to complete these tasks, which increases the complexity and computational cost of the system. To this end, we propose an image processing method and system based on AI recognition. Summary of the Invention
[0006] The purpose of the present invention is to provide an image processing method and system based on AI recognition to solve the problems raised in the above background technology.
[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0008] An image processing method based on AI recognition includes the following steps:
[0009] S1: Preprocess the input original image;
[0010] S2: Use the pre-trained deep convolutional neural network model to extract features from the pre-processed image and generate feature vectors;
[0011] S3: Input the extracted feature vector into the classifier, use the AI recognition algorithm to classify and identify the target in the image, and obtain preliminary recognition results;
[0012] S4: Based on the recognition results, the image is segmented and labeled using a semantic segmentation algorithm;
[0013] S5: Perform post-processing operations on the segmented and labeled image, and finally output the processed image.
[0014] Preferably, said S1 comprises the following steps:
[0015] S11: performing denoising processing on the input original image to remove noise in the image;
[0016] S12: performing enhancement processing on the denoised image to improve the contrast and clarity of the image;
[0017] S13: performing normalization processing on the enhanced image, scaling the pixel values to a preset range;
[0018] S14: grayscale processing is performed on the normalized image to convert the color image into a grayscale image.
[0019] Preferably, the step S1 further includes resizing the image to unify the image size for facilitating subsequent processing.
[0020] Preferably, the S2 includes:
[0021] S21: Select a pre-trained deep convolutional neural network model, wherein the deep convolutional neural network model is VGGNet or ResNet;
[0022] S22: Input the preprocessed image into the selected deep convolutional neural network model;
[0023] S23: performing convolution, activation, and pooling operations on the image through the deep convolutional neural network model to extract deep features of the image;
[0024] S24: Generate a feature vector containing deep features of the image.
[0025] Preferably, the S3 includes:
[0026] S31: Select a classifier, which is a Softmax classifier;
[0027] S32: input the feature vector into the selected classifier;
[0028] S33: Use AI recognition algorithm to process the feature vector and calculate the probability distribution of the image belonging to different categories;
[0029] S34: Determine the category of the target in the image based on the probability distribution and obtain a preliminary recognition result.
[0030] Preferably, the S4 includes:
[0031] S41: Select a semantic segmentation algorithm, which may be U-Net or Mask R-CNN.
[0032] S42: inputting the recognition result and the pre-processed image into the selected semantic segmentation algorithm;
[0033] S43: Divide the image into different semantic regions through semantic segmentation algorithm;
[0034] S44: Annotate the segmented image and use different labels or colors to identify different semantic areas.
[0035] Preferably, the S5 includes:
[0036] S51: performing boundary optimization processing on the segmented and annotated image to remove noise points and false targets on the segmentation boundary;
[0037] S52: Filling the gaps in the segmented area to make the segmentation result more complete;
[0038] S53: Perform format conversion and encoding on the optimized image to generate a final processing result, and then output the processed image.
[0039] An image processing system based on AI recognition is applicable to an image processing method based on AI recognition, comprising:
[0040] Image input module, used to receive and read raw image data in various formats;
[0041] Preprocessing module, which performs preprocessing operations on the input image;
[0042] The feature extraction module extracts features from the pre-processed image based on the pre-trained deep convolutional neural network model and generates feature vectors;
[0043] The AI recognition module uses a classifier and AI recognition algorithm to process the extracted feature vectors to achieve classification, recognition, and semantic understanding of the targets in the image, and outputs preliminary recognition results;
[0044] The segmentation and annotation module uses the semantic segmentation algorithm to segment the image and annotate it according to the recognition results;
[0045] The post-processing module performs post-processing operations on the segmented and annotated images, optimizes the segmentation results, improves the image quality, and finally generates the processed image output;
[0046] The result output module displays the processed image and related recognition results and segmentation annotation information to the user in a visual manner.
[0047] Preferably, the pre-processing module includes:
[0048] A denoising unit, used for performing denoising processing on the input original image;
[0049] An enhancement unit, used for enhancing the denoised image;
[0050] A normalization unit, used for normalizing the enhanced image;
[0051] The grayscale unit is used to perform grayscale processing on the normalized image.
[0052] It can be seen without a doubt that the above-mentioned technical solution of this application can definitely solve the technical problem to be solved by this application.
[0053] At the same time, through the above technical solutions, the present invention has at least the following beneficial effects:
[0054] 1. This invention uses a deep convolutional neural network to automatically extract deep image features and combines it with a semantic segmentation algorithm to accurately divide semantic regions, making image classification, target detection, and segmentation results more accurate. Preprocessing optimizes image quality and unifies the format. The modules work together to achieve pipeline processing. The parallel computing capability of the deep learning model accelerates feature extraction and recognition to meet real-time processing requirements.
[0055] 2. This invention can automatically learn the characteristics of different image types, adapt to complex scenarios and changing conditions, and has good versatility for images from different sources, reducing system complexity. Its high degree of automation reduces manual intervention and reliance on professional operators, simplifies subsequent maintenance, and saves time and resources. This effectively addresses the lack of accuracy and adaptability of traditional image processing methods in complex scenarios, providing an efficient, accurate, and reliable solution for the widespread application of image processing technology in security monitoring, autonomous driving, medical imaging, industrial inspection, and other fields, with significant application value and market prospects. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the technical solutions of 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0057] Figure 1 It is a schematic diagram of the process of the present invention;
[0058] Figure 2 This is a schematic diagram of step S1 of the present invention;
[0059] Figure 3 Schematic diagram of step S2 of the present invention. DETAILED DESCRIPTION
[0060] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0061] Example 1
[0062] Reference Figure 1-3 , an image processing method based on AI recognition, comprising the following steps:
[0063] S1: Preprocess the input original image;
[0064] S2: Use the pre-trained deep convolutional neural network model to extract features from the pre-processed image and generate feature vectors;
[0065] S3: Input the extracted feature vector into the classifier, use the AI recognition algorithm to classify and identify the target in the image, and obtain preliminary recognition results;
[0066] S4: Based on the recognition results, the image is segmented and labeled using a semantic segmentation algorithm;
[0067] S5: Post-process the segmented and annotated images and output the processed images. By combining AI technologies such as deep convolutional neural networks, classifiers, and semantic segmentation algorithms, efficient and accurate image processing is achieved, which is conducive to improving the degree of automation and quality of image processing and meeting the image processing needs of different application scenarios.
[0068] S1 includes the following steps:
[0069] S11: performing denoising processing on the input original image to remove noise in the image;
[0070] S12: performing enhancement processing on the denoised image to improve the contrast and clarity of the image;
[0071] S13: performing normalization processing on the enhanced image, scaling the pixel values to a preset range;
[0072] S14: Grayscale the normalized image, converting the color image into a grayscale image. These operations effectively remove noise interference from the image, enhance the image's useful information, and bring the image's pixel values into a range suitable for model processing. This reduces the amount of data and computational complexity. This provides high-quality image data for subsequent feature extraction and recognition, improving the accuracy and robustness of the entire image processing method.
[0073] S1 also includes image resizing to unify the image size for easier processing. This ensures smooth processing and avoids wasted computing resources and inconsistent results caused by image size differences. This improves the efficiency and stability of the image processing system, enabling it to better adapt to image data from different sources and sizes.
[0074] S2 includes:
[0075] S21: Select a pre-trained deep convolutional neural network model, which can be VGGNet or ResNet.
[0076] S22: Input the preprocessed image into the selected deep convolutional neural network model;
[0077] S23: Perform convolution, activation, and pooling operations on the image through a deep convolutional neural network model to extract deep features of the image;
[0078] S24: Generate feature vectors containing deep image features. VGGNet is simple in structure, easy to implement, and effective in extracting image features. ResNet, by introducing residual connections, effectively solves the training challenges of deep networks and can learn richer image features. These models provide powerful tools for image feature extraction, making the extracted feature vectors more representative and discriminative, thereby improving the accuracy of image classification and recognition.
[0079] S3 includes:
[0080] S31: Select the classifier, which is the Softmax classifier;
[0081] S32: input the feature vector into the selected classifier;
[0082] S33: Use AI recognition algorithm to process the feature vector and calculate the probability distribution of the image belonging to different categories;
[0083] S34: Determine the category of the object in the image based on the probability distribution and obtain preliminary recognition results. The Softmax classifier is suitable for multi-classification problems. It can calculate the probability distribution of the image belonging to different categories based on the feature vector, thereby accurately classifying and identifying the objects in the image. It provides a reliable classification method for image processing methods, helping to improve the credibility and accuracy of recognition results.
[0084] S4 includes:
[0085] S41: Select the semantic segmentation algorithm, which can be U-Net or Mask R-CNN.
[0086] S42: inputting the recognition result and the pre-processed image into the selected semantic segmentation algorithm;
[0087] S43: Divide the image into different semantic regions through semantic segmentation algorithm;
[0088] S44: Annotate the segmented image, using different labels or colors to identify different semantic regions. U-Net excels in fields such as medical image segmentation, accurately demarcating semantic regions within an image. Mask R-CNN excels in object detection and segmentation, simultaneously achieving both object classification and pixel-level segmentation. The application of these semantic segmentation algorithms enables image processing methods to analyze and understand images in greater detail, providing richer information for subsequent image processing and analysis.
[0089] S5 includes:
[0090] S51: performing boundary optimization processing on the segmented and annotated image to remove noise points and false targets on the segmentation boundary;
[0091] S52: Filling the gaps in the segmented area to make the segmentation result more complete;
[0092] S53: The optimized image is format converted and encoded to generate the final processing result, which is then output. These operations further improve the accuracy and quality of image segmentation, making the segmentation result more complete and accurate. The processed image is also converted into a format suitable for output and subsequent applications. This helps enhance the readability and practicality of the final output image, improving the performance and effectiveness of the entire image processing system.
[0093] Example 2
[0094] An image processing system based on AI recognition is applicable to an image processing method based on AI recognition, comprising:
[0095] Image input module, used to receive and read raw image data in various formats;
[0096] Preprocessing module, which performs preprocessing operations on the input image;
[0097] The feature extraction module extracts features from the pre-processed image based on the pre-trained deep convolutional neural network model and generates feature vectors;
[0098] The AI recognition module uses a classifier and AI recognition algorithm to process the extracted feature vectors to achieve classification, recognition, and semantic understanding of the targets in the image, and outputs preliminary recognition results;
[0099] The segmentation and annotation module uses the semantic segmentation algorithm to segment the image and annotate it according to the recognition results;
[0100] The post-processing module performs post-processing operations on the segmented and annotated images, optimizes the segmentation results, improves the image quality, and finally generates the processed image output;
[0101] The result output module visually displays the processed image, along with related recognition results, segmentation, and annotation information, to the user. Through the collaborative work of these functional modules, automatic and efficient image processing is achieved. The system completes the entire process from image input to result output, providing users with a more comprehensive and convenient image processing solution. This will help promote the application and development of image processing technology in various fields, improving work efficiency and quality.
[0102] The preprocessing modules include:
[0103] A denoising unit, used for performing denoising processing on the input original image;
[0104] An enhancement unit, used for enhancing the denoised image;
[0105] A normalization unit, used for normalizing the enhanced image;
[0106] The grayscale unit is used to grayscale the normalized image. This modular design facilitates the optimization and improvement of the preprocessing process, can better meet the preprocessing requirements of different image data, further improve the preprocessing effect, and lay a solid foundation for subsequent image processing.
[0107] From the above, we can know that:
[0108] The present invention addresses the following technical issues: Currently, traditional image processing methods have poor adaptability to image data. Images from different sources and under different imaging conditions have different characteristics, such as lighting conditions, noise levels, and resolution. Traditional methods generally require parameter adjustment and optimization for specific data sets or application scenarios, making it difficult to achieve universal processing of different image data. For example, images captured in outdoor scenes and images captured in indoor scenes have significantly different lighting conditions and noise characteristics. Traditional image processing methods may require the design of different processing flows and parameter settings to achieve better processing results, which increases the complexity and maintenance costs of the system.
[0109] With the rise of deep learning technology, image processing methods based on artificial intelligence have gradually become a research hotspot. These methods can automatically learn deep feature representations in images by constructing deep neural network models, and have stronger feature extraction capabilities and adaptability. However, most of the existing AI-based image processing methods focus on a single task, such as simple image classification, target detection or image segmentation, and lack a complete solution that can organically combine multiple image processing tasks. In practical applications, users often need to perform multiple processing operations on images. For example, in a security monitoring system, it is necessary not only to detect and identify targets on the monitoring images, but also to segment and label the target areas in order to more intuitively display the monitoring results. Existing methods usually need to call different models and algorithms separately to complete these tasks, which increases the complexity and computational cost of the system; by adopting the technical solutions of the above embodiments and through the above settings, this application will inevitably solve the above technical problems and achieve the following technical effects:
[0110] 1. This invention uses a deep convolutional neural network to automatically extract deep image features and combines it with a semantic segmentation algorithm to accurately divide semantic regions, making image classification, target detection, and segmentation results more accurate. Preprocessing optimizes image quality and unifies the format. The modules work together to achieve pipeline processing. The parallel computing capability of the deep learning model accelerates feature extraction and recognition to meet real-time processing requirements.
[0111] 2. This invention can automatically learn the characteristics of different image types, adapt to complex scenarios and changing conditions, and has good versatility for images from different sources, reducing system complexity. Its high degree of automation reduces manual intervention and reliance on professional operators, simplifies subsequent maintenance, and saves time and resources. This effectively addresses the lack of accuracy and adaptability of traditional image processing methods in complex scenarios, providing an efficient, accurate, and reliable solution for the widespread application of image processing technology in security monitoring, autonomous driving, medical imaging, industrial inspection, and other fields, with significant application value and market prospects.
[0112] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection, or communication; direct connection or indirect connection through an intermediate medium; internal communication between two elements or interaction between two elements, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.
[0113] Obviously, the embodiments described above are only some embodiments of the present invention, rather than all embodiments. The preferred embodiments of the present invention are given in the accompanying drawings, but they do not limit the patent scope of the present invention. The present invention can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive. Although the present invention has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present invention specification and drawings, directly or indirectly used in other related technical fields, is also within the scope of patent protection of the present invention.
Claims
1. An image processing method based on AI recognition, characterized in that: The following steps are involved: S1: Preprocess the input original image; S2: Use the pre-trained deep convolutional neural network model to extract features from the pre-processed image and generate feature vectors; S3: Input the extracted feature vector into the classifier, use the AI recognition algorithm to classify and identify the target in the image, and obtain preliminary recognition results; S4: Based on the recognition results, the image is segmented and labeled using a semantic segmentation algorithm; S5: Perform post-processing operations on the segmented and labeled image, and finally output the processed image.
2. The image processing method based on AI recognition according to claim 1, characterized in that: Said S1 comprises the following steps: S11: performing denoising processing on the input original image to remove noise in the image; S12: performing enhancement processing on the denoised image to improve the contrast and clarity of the image; S13: performing normalization processing on the enhanced image, scaling the pixel values to a preset range; S14: grayscale processing is performed on the normalized image to convert the color image into a grayscale image.
3. The image processing method based on AI recognition according to claim 2, characterized in that: The step S1 further includes resizing the image to unify the image size for ease of subsequent processing.
4. The image processing method based on AI recognition according to claim 1, characterized in that: The S2 includes: S21: Select a pre-trained deep convolutional neural network model, wherein the deep convolutional neural network model is VGGNet or ResNet; S22: Input the preprocessed image into the selected deep convolutional neural network model; S23: performing convolution, activation, and pooling operations on the image through the deep convolutional neural network model to extract deep features of the image; S24: Generate a feature vector containing deep features of the image.
5. The image processing method based on AI recognition according to claim 1, characterized in that: The S3 includes: S31: Select a classifier, which is a Softmax classifier; S32: input the feature vector into the selected classifier; S33: Use AI recognition algorithm to process the feature vector and calculate the probability distribution of the image belonging to different categories; S34: Determine the category of the target in the image based on the probability distribution and obtain a preliminary recognition result.
6. The image processing method based on AI recognition according to claim 1, characterized in that: The S4 includes: S41: Select a semantic segmentation algorithm, which may be U-Net or Mask R-CNN. S42: inputting the recognition result and the pre-processed image into the selected semantic segmentation algorithm; S43: Divide the image into different semantic regions through semantic segmentation algorithm; S44: Annotate the segmented image and use different labels or colors to identify different semantic areas.
7. The image processing method based on AI recognition according to claim 1, characterized in that: The S5 includes: S51: performing boundary optimization processing on the segmented and annotated image to remove noise points and false targets on the segmentation boundary; S52: Filling the gaps in the segmented area to make the segmentation result more complete; S53: Perform format conversion and encoding on the optimized image to generate a final processing result, and then output the processed image.
8. An image processing system based on AI recognition, applicable to an image processing method based on AI recognition according to any one of claims 1 to 7, characterized in that: include: Image input module, used to receive and read raw image data in various formats; Preprocessing module, which performs preprocessing operations on the input image; The feature extraction module extracts features from the pre-processed image based on the pre-trained deep convolutional neural network model and generates feature vectors; The AI recognition module uses a classifier and AI recognition algorithm to process the extracted feature vectors to achieve classification, recognition, and semantic understanding of the targets in the image, and outputs preliminary recognition results; The segmentation and annotation module uses the semantic segmentation algorithm to segment the image and annotate it according to the recognition results; The post-processing module performs post-processing operations on the segmented and annotated images, optimizes the segmentation results, improves the image quality, and finally generates the processed image output; The result output module displays the processed image and related recognition results and segmentation annotation information to the user in a visual manner.
9. The image processing system based on AI recognition according to claim 8, characterized in that: The pre-processing module comprises: A denoising unit, used for performing denoising processing on the input original image; An enhancement unit, used for enhancing the denoised image; A normalization unit, used for normalizing the enhanced image; The grayscale unit is used to perform grayscale processing on the normalized image.