Frequency-Domain Blur Kernel Estimation for Blind Image Super-Resolution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image super-resolution technologies struggle with poor applicability when blur kernels are unknown, leading to information loss and inaccurate restoration, especially in blind super-resolution methods, and spatial domain-based blur kernel estimation is difficult for special blur kernels like motion blur.
Innovation Solution
Perform frequency domain transformation on images to estimate blur kernels using spectral features, enabling accurate prediction and subsequent super-resolution processing, applicable to both synthetic and real images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If blind super-resolution transforms an image with unknown blur kernel into a bicubic blurred image domain, then the non-blind super-resolution technology can be used for restoration, but image information is lost making it impossible to restore to obtain a high-quality high-definition image
Solution Approach 1:
The patent changes the domain parameter from spatial domain to frequency domain for blur kernel estimation. By performing frequency domain transformation on the input image and analyzing spectral features, the method estimates the blur kernel without transforming the image into a bicubic blurred domain, thus avoiding information loss while enabling accurate restoration.
2Ease of manufacture
If blur kernel estimation is performed based on spatial domain, then the process is simple, but it is difficult to predict special blur kernels like motion blur
Solution Approach 1:
The patent transitions from spatial domain to frequency domain for blur kernel estimation. By performing frequency domain transformation and analyzing spectral features in the frequency dimension, the method achieves accurate prediction of special blur kernels like motion blur that cannot be effectively predicted in the spatial domain.
3Reliability
If non-blind super-resolution is used, then restoration can be performed when blur kernel is known, but applicability is poor when blur kernel is unknown
Solution Approach 1:
The patent performs preliminary blur kernel estimation in the frequency domain before executing the super-resolution restoration process. By estimating the blur kernel from spectral features of the input image with unknown degradation, the method enables non-blind super-resolution to be applied to blind super-resolution scenarios, significantly improving applicability while maintaining restoration accuracy.
Data Source
AI summary
An image super-resolution method includes performing frequency domain transformation on a first image to obtain a spectral feature of the first image, the spectral feature representing a distribution of a grayscale gradient in the first image. The method further includes performing blur kernel prediction based on the spectral feature to obtain a blur kernel of the first image, the blur kernel being a convolution kernel. The method also includes performing super-resolution processing on the first image based on the blur kernel to generate a super-resolved image, a definition of the super-resolved image being higher than a definition of the first image.