Intelligent image processing system and method based on AI
By using an AI-based intelligent image processing system, lightweight neural network units are dynamically selected and executed to process images, solving the problems of generalization ability and unstable results of traditional methods in complex scenes, and achieving efficient and flexible image quality improvement.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTHWEAT UNIV OF SCI & TECH
- Filing Date
- 2026-01-21
- Publication Date
- 2026-05-05
AI Technical Summary
Traditional image processing methods have poor generalization ability and unstable results when dealing with complex and ever-changing natural scenes and strong noise interference. They are also difficult to cope with unknown types of degradation, and the processing flow is rigid and the task is limited.
An AI-based intelligent image processing system is adopted, including an image input interface, an analysis module, a dynamic task routing pool, and a lightweight processing module chain. Through multi-dimensional feature extraction and task decision-making, multiple lightweight neural network units are dynamically selected and executed for image processing, supporting a variety of image quality improvement tasks.
It achieves flexible adaptation to different image degradation types, improves processing efficiency and effect, adapts to images from different sources, has strong robustness and generalization ability, and is easy to deploy on mobile or embedded devices.
Smart Images
Figure CN121981900A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, specifically to an AI-based intelligent image processing system and method. Background Technology
[0002] Image processing refers to the manipulation and analysis of digital images through a series of algorithms and techniques, aiming to improve image quality, enhance useful information, or transform them into a form more suitable for human-computer recognition. It is a fundamental key technology in many fields such as computer vision, photogrammetry, and medical imaging. Typical image processing tasks include noise removal, blur restoration, resolution enhancement, and contrast and color enhancement. These processes are crucial for improving the visual appeal of images and enhancing the input quality for subsequent advanced visual tasks.
[0003] Traditional image processing methods, such as filtering based on fixed operators, edge detection, and histogram equalization, essentially rely on manually designed features and prior models. When dealing with complex and ever-changing natural scenes, strong noise interference, or highly diverse image content, they have poor generalization ability, unstable results, and difficulty in dealing with unknown types of degradation. Existing AI processing technologies tend to have a single task, a fixed processing flow, and rigid scenarios.
[0004] To address the aforementioned issues, we propose an AI-based intelligent image processing system and method. Summary of the Invention
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an AI-based intelligent image processing system and method, including an image input interface, an image analysis module, multi-dimensional features of the image, a dynamic task routing pool, a lightweight processing module chain, and an image output interface. The image input interface is used to acquire the original image. The image analysis module is connected to the image input interface and is used to extract multi-dimensional features of the original image and generate task decision vectors and intensity control parameters. The dynamic task routing pool is connected to the image analysis module and can select and sort at least two sub-tasks from a plurality of preset image processing sub-tasks based on the task decision vector to form a processing sequence. The lightweight processing module is linked to a dynamic task routing pool and includes multiple pluggable neural network units, each corresponding to an image processing subtask, for sequentially executing processing according to the processing sequence. The image output interface is chained to the lightweight processing module and is used to output the processed enhanced image.
[0006] As a preferred embodiment of the present invention, the image analysis module includes a feature extraction backbone network, a task decision branch, and a parameter prediction branch, wherein the task decision branch and the parameter prediction branch are parallel; the multi-dimensional features include noise level, ambiguity, contrast, color distribution, and semantic content complexity.
[0007] As a preferred embodiment of the present invention, the dynamic task routing pool is configured with logical dependencies and priority rules between subtasks; the image processing subtasks include various types such as denoising, deblurring, super-resolution reconstruction, contrast enhancement, and color correction.
[0008] As a preferred embodiment of the present invention, each neural network unit is an independently trained and optimized model, which is only for a single image processing task. The neural network unit structure adopts a lightweight design that includes an attention mechanism and depthwise separable convolution.
[0009] An AI-based intelligent image processing method, which can be deployed on cloud servers, mobile terminals, and embedded image acquisition devices, includes the following steps: S1. Obtain the original image to be processed; S2. Analyze the original image, extract multi-dimensional features, and generate task decision vectors and intensity control parameters; S3. Based on the task decision vector, select and sort at least two sub-tasks from a plurality of preset image processing sub-tasks to form a processing sequence; S4. According to the processing sequence, the corresponding lightweight neural network units are called sequentially to process the image; S5. Output the final enhanced image.
[0010] As a preferred technical solution of the present invention, when selecting and sorting each subtask, the pre-set logical dependencies and priority rules between subtasks must be followed.
[0011] As a preferred embodiment of the present invention, each lightweight neural network unit is independently trained using a specialized dataset for the corresponding task; at the same time, the parameters of each neural network unit are fixed, and the image analysis module is trained end-to-end using a comprehensive dataset containing multiple degradation types, and the dynamic task routing strategy is optimized.
[0012] The beneficial effects of this invention are as follows: This AI-based intelligent image processing system and method dynamically determines the processing flow and intensity based on the characteristics of the image itself through a front-end analysis module, avoiding excessive deletion and optimization; through a dynamic routing mechanism, multiple lightweight sub-modules are flexibly combined to collaboratively solve various image quality problems in a single forward propagation, improving overall efficiency and effectiveness; at the same time, each sub-module adopts lightweight designs such as depthwise separable convolution and channel attention, resulting in a small number of overall model parameters, fast inference speed, and easy deployment on mobile or embedded devices. The modular design and dynamic combination strategy enable the system to better adapt to images from different sources and with different degradation types, exhibiting stronger robustness and generalization ability. Attached Figure Description
[0013] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall architecture diagram of an AI-based intelligent image processing system according to the present invention; Figure 2 This is a structural diagram of the image analysis module of an AI-based intelligent image processing system according to the present invention; Figure 3 This is a flowchart of the dynamic task routing and processing of an AI-based intelligent image processing system and method according to the present invention. Figure 4 This is a schematic diagram of a lightweight processing module chain of an AI-based intelligent image processing system and method according to the present invention. Figure 5 This is a system training and deployment architecture diagram of an AI-based intelligent image processing system according to the present invention; Detailed Implementation
[0014] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.
[0015] Example: Figures 1-5 As shown, an AI-based intelligent image processing system includes an image input interface, an image analysis module, multi-dimensional image features, a dynamic task routing pool, a lightweight processing module chain, and an image output interface. The image input interface is used to acquire the original image. The image analysis module is connected to the image input interface to extract multi-dimensional features of the original image and generate task decision vectors and intensity control parameters; The dynamic task routing pool connects to the image analysis module and can select and sort at least two sub-tasks from multiple preset image processing sub-tasks based on the task decision vector to form a processing sequence. The lightweight processing module chain connects to the dynamic task routing pool and includes multiple pluggable neural network units, each corresponding to an image processing subtask, which is used to execute processing sequentially according to the processing sequence. The image output interface is chained with the lightweight processing module to output the enhanced image after processing.
[0016] The image analysis module includes a feature extraction backbone network, a task decision branch, and a parameter prediction branch, with the task decision branch and parameter prediction branch operating in parallel. Multi-dimensional features include noise level, blurriness, contrast, color distribution, and semantic content complexity.
[0017] The dynamic task routing pool is configured with logical dependencies and priority rules between subtasks; image processing subtasks include various types such as denoising, deblurring, super-resolution reconstruction, contrast enhancement, and color correction.
[0018] Each neural network unit is an independently trained and optimized model designed for a single image processing task. The neural network unit structure employs a lightweight design that incorporates attention mechanisms and depthwise separable convolutions.
[0019] An AI-based intelligent image processing method, which can be deployed on cloud servers, mobile terminals, and embedded image acquisition devices, includes the following steps: Obtain the original image to be processed; The original image is analyzed to extract multi-dimensional features and generate task decision vectors and intensity control parameters. Based on the task decision vector, at least two sub-tasks are selected and sorted from multiple preset image processing sub-tasks to form a processing sequence; Based on the processing sequence, the corresponding lightweight neural network units are called sequentially to process the image; Output the final enhanced image.
[0020] When selecting and sorting subtasks, the predefined logical dependencies and priority rules between subtasks must be followed.
[0021] Each lightweight neural network unit is trained independently using a specialized dataset for the corresponding task. Meanwhile, the parameters of each neural network unit are fixed, and the image analysis module is trained end-to-end using a comprehensive dataset containing multiple degradation types, and the dynamic task routing strategy is optimized.
[0022] The image input interface supports multiple image input formats: JPEG, PNG, BMP, and RAW. It can interface with cameras, image sensors, network streams, or local storage devices, and provides image preprocessing functions (such as normalization and resizing). The image analysis module uses a lightweight convolutional neural network backbone: MobileNetV3 or EfficientNet-Lite, combined with a multi-scale feature fusion mechanism. The task decision branch outputs a multi-dimensional vector to represent the execution priority of each subtask, and the parameter prediction branch outputs intensity adjustment parameters to adjust the processing intensity. The dynamic task routing pool has built-in logical dependencies and priority rules, such as denoising → deblurring → super-resolution, and supports dynamic sequence generation based on task decision vectors. It can also support parallel processing optimization of concurrent subtasks. Each module in the lightweight processing module chain is an independently trained single-task model, with the overall parameter count controlled within 5M. It supports INT8 quantization deployment, real-time loading and unloading of modules, and facilitates system expansion. The image output interface supports outputting as image files, video streams, or direct transmission to display devices, and can output metadata such as processing sequence, intensity parameters, and processing time.
[0023] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An AI-based intelligent image processing system, comprising an image input interface, an image analysis module, multi-dimensional image features, a dynamic task routing pool, a lightweight processing module chain, and an image output interface, characterized in that, The image input interface is used to acquire the original image; The image analysis module is connected to the image input interface and is used to extract multi-dimensional features of the original image and generate task decision vectors and intensity control parameters. The dynamic task routing pool is connected to the image analysis module and can select and sort at least two sub-tasks from a plurality of preset image processing sub-tasks based on the task decision vector to form a processing sequence. The lightweight processing module is linked to a dynamic task routing pool and includes multiple pluggable neural network units, each corresponding to an image processing subtask, for sequentially executing processing according to the processing sequence. The image output interface is chained to the lightweight processing module and is used to output the processed enhanced image.
2. The AI-based intelligent image processing system according to claim 1, characterized in that, The image analysis module includes a feature extraction backbone network, a task decision branch, and a parameter prediction branch, with the task decision branch and parameter prediction branch operating in parallel. The multi-dimensional features include noise level, ambiguity, contrast, color distribution, and semantic content complexity.
3. The AI-based intelligent image processing system according to claim 1, characterized in that, The dynamic task routing pool is configured with logical dependencies and priority rules between subtasks; the image processing subtasks include various types such as denoising, deblurring, super-resolution reconstruction, contrast enhancement, and color correction.
4. The AI-based intelligent image processing system according to claim 1, characterized in that, Each neural network unit is an independently trained and optimized model designed for a single image processing task. The neural network unit structure employs a lightweight design that incorporates attention mechanisms and depthwise separable convolutions.
5. An AI-based intelligent image processing method, applied to the AI-based intelligent image processing system according to any one of claims 1-4, characterized in that, The system can be deployed on cloud servers, mobile terminals, and embedded image acquisition devices.
6. The AI-based intelligent image processing method according to claim 5, characterized in that, Includes the following steps: S1. Obtain the original image to be processed; S2. Analyze the original image, extract multi-dimensional features, and generate task decision vectors and intensity control parameters; S3. Based on the task decision vector, select and sort at least two sub-tasks from a plurality of preset image processing sub-tasks to form a processing sequence; S4. According to the processing sequence, the corresponding lightweight neural network units are called sequentially to process the image; S5. Output the final enhanced image.
7. The AI-based intelligent image processing method according to claim 6, characterized in that, When selecting and sorting subtasks, the predefined logical dependencies and priority rules between subtasks must be followed.
8. The AI-based intelligent image processing method according to claim 7, characterized in that, Each lightweight neural network unit is trained independently using a specialized dataset for the corresponding task. Meanwhile, with the parameters of each neural network unit fixed, the image analysis module is trained end-to-end using a comprehensive dataset containing multiple degradation types, and the dynamic task routing strategy is optimized.