Super-Resolution Image Model Training with Frequency-Domain Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep neural networks for image/video super-resolution struggle to effectively learn and reconstruct high-frequency information, leading to performance issues in generating sharp high-resolution images.
Innovation Solution
A method that transforms super-resolution images into frequency-domain representations and optimizes the initial model using frequency-domain loss to align high-frequency components with real images, thereby recovering missing information and improving the prediction effect.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If deep neural networks are used for image/video super-resolution, then the model can process and generate images, but the network struggles to effectively learn and reconstruct high-frequency information, resulting in poor sharpness and quality in generated high-resolution images
Solution Approach 1:
The patent transforms the super-resolution problem from the spatial domain to the frequency domain by introducing a frequency-domain loss function. This dimensional change allows the model to specifically target and reconstruct high-frequency information that was previously difficult to capture, thereby improving image sharpness and quality without losing critical high-frequency details
Solution Approach 2:
The patent modifies the training objective by changing the loss function parameters to operate in the frequency domain rather than the spatial domain. By applying Fourier transformation to the loss calculation, the model optimizes different parameters (frequency components) that directly correspond to image sharpness and detail, enabling better reconstruction of high-frequency information
Data Source
AI summary
Embodiments of the present disclosure provide a method, a device, and a computer program product for generating a super-resolution image model. The method includes acquiring a first image with a first resolution and a second image with a second resolution, the first image corresponding to the second image; generating a first super-resolution image with a first super resolution and a second super-resolution image with a second super resolution according to an initial super-resolution image model based on the first image; transforming the first super-resolution image into a first frequency-domain representation; transforming the second super-resolution image into a second frequency-domain representation; and generating a trained super-resolution image model based on a loss between the first frequency-domain representation and the second frequency-domain representation and a reference frequency-domain representation of the second image.