Image Super-Resolution With Local Channel Self-Attention Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image super-resolution models based on convolutional neural networks face issues with large model parameters leading to overfitting and slow computation speeds for high-resolution images, while channel attention models neglect local information, resulting in blurred images.
Innovation Solution
An image super-resolution method utilizing a channel attention network with multi-level cascaded local channel self-attention layers that divide input features into blocks, recalibrate them using a channel self-attention mechanism, and combine these blocks to enhance texture details while maintaining computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If convolutional neural networks are used for image super-resolution, then model completeness is improved, but model parameters increase leading to overfitting and slow computation
Solution Approach 1:
The patent divides the feature processing into block-based local channel self-attention layers, where each layer processes local feature blocks separately. This segmentation reduces the computational complexity and number of parameters compared to global attention mechanisms, while maintaining model effectiveness through localized feature recalibration.
2Productivity
If channel attention models are used for image super-resolution, then computation speed is improved, but local information is neglected resulting in blurred images
Solution Approach 1:
The patent implements local channel self-attention layers that operate on divided feature blocks rather than global features. This local quality approach allows the model to preserve and enhance local texture details while maintaining computational efficiency, directly addressing the blurring issue of global channel attention models.
3Reliability
If global channel self-attention is applied to high-resolution images, then computation time increases significantly
Solution Approach 1:
The patent segments the feature map into multiple local blocks and applies self-attention independently to each block. This segmentation strategy reduces the computational time for high-resolution images while maintaining the effectiveness of the attention mechanism through localized feature recalibration.
Data Source
AI summary
Embodiments include an image super-resolution method and apparatus. The method includes: performing feature extraction on a to-be-super-resolved image to obtain a first image feature; processing the first image feature by using a channel attention network to obtain a second image feature, where the channel attention network includes multi-level cascaded local channel self-attention layers, any one of the local channel self-attention layers is configured to divide an input feature of the local channel self-attention layer into multiple first feature blocks, separately recalibrate the multiple first feature blocks based on a channel self-attention mechanism to obtain a second feature block corresponding to each first feature block, combine second feature blocks corresponding to the multiple first feature blocks to obtain a combined feature, and obtain an output feature based on the combined feature; and generating, based on the second image feature and the to-be-super-resolved image, a super-resolution image corresponding to the to-be-super-resolved image.