Reference-Guided Image Super-Resolution for Continuous-Scale Enlargement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image super-resolution methods struggle to produce sharp and clear high-resolution images and are limited to fixed-scale resolution enlargement, failing to achieve continuous-scale enlargement.
Innovation Solution
A method that constructs a model using a reference image and a low-resolution image during training, allowing the model to complement edge and texture information and achieve continuous-scale resolution enlargement by interpolating high-frequency and low-frequency information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If conventional image super-resolution methods are used, then image resolution is enlarged, but the image quality becomes blurry and lacks sharp details
Solution Approach 1:
The patent segments the image processing into two distinct streams: a low-frequency stream that processes the input low-resolution image to preserve structural information, and a high-frequency stream that processes a reference high-resolution image to extract detailed texture information. This segmentation allows each stream to specialize in different aspects of image quality, resolving the contradiction between resolution enlargement and image sharpness.
Solution Approach 2:
The patent applies local quality by treating different frequency components of the image differently. The low-frequency components (structural information) are processed through one pathway while high-frequency components (detail information) are processed through another pathway. This allows the system to maintain structural accuracy while enhancing local details, thereby improving overall image quality during resolution enlargement.
2Adaptability or versatility
If fixed-scale resolution enlargement is applied, then image processing is simplified, but adaptability to different client device resolutions is lost
Solution Approach 1:
The patent implements dynamics by making the reference image selectable and changeable based on the target resolution requirements. Instead of a fixed processing pipeline, the system can dynamically choose different reference images from a pre-trained set, each optimized for different scaling factors. This allows the model to adapt to various client device resolutions without requiring a completely different model architecture, balancing adaptability with manageable complexity.
Solution Approach 2:
The patent changes parameters by using multiple pre-trained models with different scaling factors (e.g., 2x, 4x, 8x) and selecting the appropriate model based on the desired output resolution. This parameter-based approach allows the system to achieve continuous-scale enlargement capability while keeping each individual model relatively simple, as each model is specialized for a specific scaling task rather than attempting to handle all scales universally.
Data Source
AI summary
Embodiments of the present disclosure relate to a method, a device, and a computer program product for image processing. The method includes: obtaining an encoding feature of a reference image and an encoding feature of an input image of a first resolution, wherein the reference image has a resolution greater than the first resolution. The method further includes: obtaining high-frequency information and low-frequency information on the input image by interpolating the input image; obtaining a first output feature based on the encoding feature of the reference image and the high-frequency information; and obtaining a second output feature based on the encoding feature of the input image and the low-frequency information. The method further includes: generating an output image of a second resolution based on the first output feature and the second output feature, wherein the second resolution is greater than the first resolution.