CN-121981900-A - Intelligent image processing system and method based on AI

CN121981900ACN 121981900 ACN121981900 ACN 121981900ACN-121981900-A

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent image processing system and method based on AI, comprising an image input interface, an image analysis module, multi-dimensional characteristics of images, a dynamic task routing pool, a lightweight processing module chain and an image output interface, wherein the image input interface is used for acquiring original images. The AI-based intelligent image processing system and the AI-based intelligent image processing method dynamically determine the processing flow and intensity according to the characteristics of the image by a front-end analysis module to avoid excessive deletion and optimization, and improve the overall efficiency and effect by a dynamic routing mechanism, meanwhile, each sub-module adopts a lightweight design such as depth separable convolution, channel attention and the like, the overall model has small parameter, high reasoning speed and easy deployment on a mobile end or embedded equipment, and the system can be better adapted to images with different sources and different degradation types by a modularized design and dynamic combination strategy, so that the robustness and generalization capability are stronger.

Inventors

CHEN ZHE
HE YE
HUANG XINGNAN
CHEN YUXIN
LIU YING

Assignees

西南科技大学

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (8)

1. An AI-based intelligent image processing system comprises an image input interface, an image analysis module, multi-dimensional characteristics of images, a dynamic task routing pool, a lightweight processing module chain and an image output interface, and is characterized in that the image input interface is used for acquiring original images; the image analysis module is connected with the image input interface and is used for extracting the multidimensional characteristics of the original image and generating a task decision vector and an intensity regulation parameter; the dynamic task routing pool is connected with the image analysis module, and can select and sort at least two subtasks from a plurality of preset image processing subtasks based on the task decision vector to form a processing sequence; the lightweight processing module chain is connected with the dynamic task routing pool and comprises a plurality of pluggable neural network units, and each unit corresponds to one image processing subtask and is used for sequentially executing processing according to a processing sequence; the image output interface is connected with the lightweight processing module chain and is used for outputting the processed enhanced result image.
2. The AI-based intelligent image processing system of claim 1, wherein the image analysis module includes a feature extraction backbone network, task decision branches and parameter prediction branches, the task decision branches and parameter prediction branches being parallel, the multi-dimensional features including noise level, ambiguity, contrast, color distribution, and semantic content complexity.
3. The AI-based intelligent image processing system of claim 1, wherein logical dependencies and priority rules between subtasks are provided in the dynamic task routing pool, wherein the image processing subtasks include a plurality of denoising, deblurring, super-resolution reconstruction, contrast enhancement, and color correction.
4. The AI-based intelligent image processing system of claim 1, wherein each of the neural network elements is an independently trained and optimized model that uses a lightweight design that includes an attention mechanism and a depth separable convolution for only a single image processing task.
5. An AI-based intelligent image processing method applied to the AI-based intelligent image processing system described in any one of claims 1-4, wherein the system can be deployed in a cloud server, a mobile terminal and an embedded image acquisition device.
6. The AI-based intelligent image processing method of claim 5, comprising the steps of: s1, acquiring an original image to be processed; S2, analyzing the original image, extracting multidimensional features, and generating task decision vectors and intensity regulation parameters; s3, selecting and sequencing at least two subtasks from a plurality of preset image processing subtasks based on the task decision vector to form a processing sequence; S4, according to the processing sequence, sequentially calling corresponding lightweight neural network units to process the image; S5, outputting a final obtained enhanced result image.
7. The AI-based intelligent image processing method of claim 6, wherein a preset logical dependency relationship and priority rule between sub-tasks is followed when selecting to order the sub-tasks.
8. The AI-based intelligent image processing method of claim 7 wherein each lightweight neural network unit is independently trained using a specialized dataset of a corresponding task while fixing parameters of each neural network unit, training the image analysis module end-to-end using a comprehensive dataset comprising multiple degradation types, and optimizing the dynamic task routing strategy.

Description

Intelligent image processing system and method based on AI Technical Field The invention relates to the technical field of artificial intelligence, in particular to an intelligent image processing system and method based on AI. Background The image processing refers to the operation and analysis of digital images through a series of algorithms and techniques, aims at improving the image quality, enhancing useful information, or converting the digital images into a form more suitable for human-computer recognition, and is a bottom key technology in various fields such as computer vision, photogrammetry, medical imaging and the like. Typical image processing tasks include noise removal, blur restoration, resolution enhancement, contrast and color enhancement, and the like. These processes are critical to improving the visual perception of the image and improving the input quality of subsequent advanced visual tasks. The traditional image processing method such as filtering, edge detection, histogram equalization and the like based on fixed operators is essentially dependent on manually designed characteristics and prior models, and has poor generalization capability and unstable effect when processing complex and changeable natural scenes, strong noise interference or highly diversified image contents, and is difficult to cope with unknown type degradation, and the existing AI processing technology has the disadvantages of biased task singleness, solidified processing flow and stiff scenes. To solve the above problems, we propose an intelligent image processing system and method based on AI. Disclosure of Invention In order to solve the technical problems, the invention provides the following technical scheme: the invention provides an AI-based intelligent image processing system and method, comprising an image input interface, an image analysis module, multi-dimensional characteristics of images, a dynamic task routing pool, a lightweight processing module chain and an image output interface, wherein the image input interface is used for acquiring original images; the image analysis module is connected with the image input interface and is used for extracting the multidimensional characteristics of the original image and generating a task decision vector and an intensity regulation parameter; the dynamic task routing pool is connected with the image analysis module, and can select and sort at least two subtasks from a plurality of preset image processing subtasks based on the task decision vector to form a processing sequence; the lightweight processing module chain is connected with the dynamic task routing pool and comprises a plurality of pluggable neural network units, and each unit corresponds to one image processing subtask and is used for sequentially executing processing according to a processing sequence; the image output interface is connected with the lightweight processing module chain and is used for outputting the processed enhanced result image. As a preferable technical scheme of the invention, the image analysis module comprises a feature extraction backbone network, task decision branches and parameter prediction branches, wherein the task decision branches and the parameter prediction branches are parallel, and the multidimensional features comprise noise level, ambiguity, contrast, color distribution and semantic content complexity. As an optimal technical scheme, the dynamic task routing pool is internally provided with a logic dependency relationship and a priority rule among subtasks, and the image processing subtasks comprise a plurality of denoising, deblurring, super-resolution reconstruction, contrast enhancement and color correction. As a preferred technical scheme of the invention, each neural network unit is an independently trained and optimized model, and the model is only used for a single image processing task, and the neural network unit structure adopts a lightweight design comprising an attention mechanism and depth separable convolution. The AI-based intelligent image processing method comprises the following steps that the system can be deployed in a cloud server, a mobile terminal and an embedded image acquisition device: s1, acquiring an original image to be processed; S2, analyzing the original image, extracting multidimensional features, and generating task decision vectors and intensity regulation parameters; s3, selecting and sequencing at least two subtasks from a plurality of preset image processing subtasks based on the task decision vector to form a processing sequence; S4, according to the processing sequence, sequentially calling corresponding lightweight neural network units to process the image; S5, outputting a final obtained enhanced result image. As a preferable technical scheme of the invention, when selecting and sequencing all the subtasks, a preset logic dependency relationship and a priority rule among the subtasks are required to be followed. Acco