Composite Image Denoising Network for Low-Light Detail Preservation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image denoising algorithms using neural networks fail to achieve good denoising performance while considering the quality of the denoised image, particularly in low-light conditions, affecting image resolution and object recognition accuracy.
Innovation Solution
An image denoising method combining a U-shaped network, residual network, and dense network to utilize deep and shallow features, along with an adversarial network training process using generated and actual noise images to enhance denoising performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional neural network denoising algorithms are used, then noise removal is achieved, but image quality and detail preservation deteriorate
Solution Approach 1:
The network is divided into multiple stages with different denoising strengths. The first denoising network removes obvious noise with stronger denoising capability, while the second denoising network handles residual noise with weaker denoising capability, preserving image details better than a single strong denoising network
Solution Approach 2:
The patent combines multiple neural network architectures (U-Net, residual networks, dense networks) into a composite denoising system. This composite structure integrates the advantages of different network types to achieve both effective noise removal and detail preservation
2Object-affected harmful factors
If strong denoising is applied, then noise is removed, but image resolution and detail information are lost
Solution Approach 1:
The denoising process is segmented into two sequential networks with different denoising intensities. The first network applies strong denoising to remove prominent noise, while the second network applies gentle denoising to preserve fine details and resolution
Solution Approach 2:
The first denoised image serves as an intermediary between the noisy input image and the final output. It provides a cleaned intermediate result that the second network further processes with preserved details
3Device complexity
If single-stage denoising is used, then processing is simple, but denoising performance and image quality are insufficient
Solution Approach 1:
The single-stage denoising process is divided into two sequential stages with different network configurations. This segmentation improves denoising performance by handling different noise levels appropriately while maintaining manageable system complexity
Solution Approach 2:
The patent employs dynamic denoising where the second network's processing is adapted based on the output of the first network. This dynamic approach optimizes performance for different noise conditions without requiring a completely different system architecture
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present application relates to the field of image processing, and provides an image denoising method and apparatus, an electronic device and a storage medium. The image denoising method includes: acquiring an image to be processed, and inputting the image to be processed into an image denoising model to acquire a denoised image, wherein the image denoising model is a model formed by combining a U-shaped network, a residual network and a dense network.