Image Super-Resolution Reconstruction with Feature-Aware Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face image super-resolution reconstruction methods, whether signal reconstruction-based or machine learning-based, suffer from loss of detail information and high-frequency information, resulting in blurred edges and jagged edges due to reliance on pixel mean square error in training super-resolution models.
Innovation Solution
An image reconstruction method that incorporates a newly constructed super-resolution model trained using a combination of pixel mean square error and image feature mean square error, including texture, shape, and spatial relationship features, to improve the accuracy of the model and reduce high-frequency information loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a super-resolution model is trained using only pixel mean square error, then the training process is simple and easy to implement, but the reconstructed image loses high-frequency information and produces blurred edges
Solution Approach 1:
The patent combines multiple loss functions (pixel mean square error loss, feature mean square error loss, and gradient loss) into a comprehensive error loss function. This merging of different loss components allows the model to simultaneously optimize pixel-level accuracy and preserve high-frequency information, resolving the contradiction between training simplicity and information preservation.
Solution Approach 2:
The error loss function acts as a composite structure integrating multiple loss components with different weights. This composite approach enables the training process to balance between simple pixel-level reconstruction and complex high-frequency detail preservation, achieving both ease of implementation and information retention.
2Ease of manufacture
If a super-resolution model is trained using only pixel mean square error, then the training process is straightforward, but the reconstructed image exhibits obvious jagged edges and lost detail information
Solution Approach 1:
The patent merges pixel-level loss with feature-level loss and gradient-based loss to create a multi-component error loss function. This combination enables the model to achieve high reconstruction quality by optimizing both pixel accuracy and structural fidelity simultaneously, while maintaining a relatively straightforward training process through unified loss aggregation.
Solution Approach 2:
The feature mean square error loss and gradient loss act as intermediaries that bridge the gap between simple pixel-level optimization and complex reconstruction quality requirements. These intermediary loss components guide the model to preserve edges and details without complicating the overall training framework.
3Device complexity
If a super-resolution model is trained using only pixel mean square error, then the model structure remains simple, but the reconstructed image lacks texture details and high-level semantic features
Solution Approach 1:
The error loss function serves as a composite structure that integrates pixel-level, feature-level, and gradient-level losses. This composite approach enables the relatively simple model structure to capture complex texture details and semantic information by multi-level optimization, without significantly increasing model complexity.
Solution Approach 2:
The patent extends the optimization from a single pixel-level dimension to multiple dimensions by incorporating feature-space loss and gradient-space loss. This dimensional extension allows the simple model structure to recover lost information by optimizing across different representation levels simultaneously.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments of this application provide an image reconstruction method and device. The method includes: inputting a first image into a newly constructed super-resolution model to obtain a reconstructed second image, where a resolution of the second image is higher than that of the first image; the newly constructed super-resolution model is obtained by training an initial super-resolution model by using an error loss; the error loss includes a pixel mean square error and an image feature mean square error; and an image feature includes at least one of a texture feature, a shape feature, a spatial relationship feature, and an image high-level semantic feature. According to the embodiments of this application, quality of a reconstructed image can be improved.