Stacked Hourglass Defogging for Local-Global Feature Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based image defogging methods suffer from low efficiency and poor quality due to the need to estimate a large number of parameters and insufficient richness of features, failing to effectively capture both local and global features.
Innovation Solution
A multi-scale fusion defogging method using a stacked hourglass network with a multi-scale jump connection module and hierarchical attention distillation module to extract and fuse features at different scales, optimizing network structure for improved feature representation and quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If deep learning-based defogging methods are used to directly output defogged images, then the processing speed is improved, but the defogging quality deteriorates due to insufficient feature richness and large number of parameters to estimate
Solution Approach 1:
The patent segments the defogging process into multiple parallel convolutional streams operating at different scales (1/2, 1/4, 1/8, 1/16 of original scale). Each stream extracts features independently at its specific scale, avoiding the need to estimate all parameters in a single large network, thus maintaining processing speed while improving feature richness and image quality.
Solution Approach 2:
The patent adds a scale dimension by processing features at multiple resolutions (different spatial dimensions). The multi-scale feature fusion combines information from 1/2, 1/4, 1/8, and 1/16 scale streams, enriching the feature representation without proportionally increasing parameter count, thereby improving defogging quality while maintaining efficiency.
2Device complexity
If a single network is used for defogging, then the device complexity is reduced, but the generalization performance deteriorates due to insufficient feature richness
Solution Approach 1:
The patent merges multiple convolutional streams at different scales into a unified feature fusion module. The multi-scale features from streams at 1/2, 1/4, 1/8, and 1/16 scales are combined through addition operations, achieving rich feature representation and good generalization performance while keeping the overall network structure relatively simple and manageable.
3Manufacturing precision
If multiple parameters are estimated in defogging models, then the image quality is improved, but the computational efficiency deteriorates due to the large number of parameters
Solution Approach 1:
The patent applies local quality by having different convolutional streams process different spatial scales with appropriate feature extraction. Each stream operates locally at its specific scale with a manageable number of parameters, avoiding the need to estimate all parameters globally at once, thus maintaining computational efficiency while achieving high image quality through multi-scale feature fusion.
Data Source
AI summary
Disclosed is a multi-scale fusion defogging method based on a stacked hourglass network, including inputting a foggy image into a preset image defogging network; and outputting a fogless image after the foggy image is processed by the image defogging network. The image defogging network includes a 7×7 convolutional layer, a stacked hourglass module, a feature fusion, a multi-scale jump connection module, a 1×1 convolutional layer, a 3×3 convolutional layer, a hierarchical attention distillation module, the 3×3 convolutional layer and the 1×1 convolutional layer connected sequentially.