Super Resolution Model Reference Patch Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current super-resolution techniques face limitations in generating high-quality output images from low-resolution inputs, particularly in effectively utilizing external reference images to recover rich textures and overcoming the performance gaps between different image resolutions.
Innovation Solution
A processor-implemented method that adjusts the position of a reference patch in a reference image based on pixel values from a ground truth patch and a reference patch, using a super-resolution model to generate an SR image, and trains the model on the SR image and ground truth image to improve texture recovery and resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a reference patch is directly used from a reference image for super-resolution, then the process is simple and fast, but the texture recovery quality is insufficient due to misalignment between reference and target image regions
Solution Approach 1:
The patent performs preliminary position adjustment of the reference patch by comparing pixel values with the ground truth patch before generating the super-resolution image. This preliminary alignment action ensures that the reference patch is properly positioned to capture accurate texture information, thereby improving texture recovery quality without requiring complex real-time alignment during the main SR process.
Solution Approach 2:
The patent replaces complex mechanical or geometric alignment mechanisms with a pixel value-based comparison approach. Instead of using feature matching or transform-based alignment methods, the system directly compares pixel values between the reference patch, ground truth patch, and adjusts the reference patch position based on pixel value differences, simplifying the alignment process while maintaining accuracy.
2Measurement precision
If the reference patch position is adjusted based on pixel value comparison, then texture transfer accuracy is improved, but the computational complexity and processing time increase
Solution Approach 1:
The patent applies partial action by performing position adjustment only on the reference patch rather than the entire reference image. By focusing the pixel value comparison and position adjustment on a localized patch level, the system achieves accurate alignment without the computational burden of processing the full image, thereby reducing processing time while maintaining precision.
3Reliability
If the SR model is trained with accurately aligned reference patches, then the generated SR image quality is improved, but the training complexity and data preparation requirements increase
Solution Approach 1:
The patent implements self-service by enabling the SR model to automatically adjust reference patch positions during training based on pixel value comparisons with ground truth patches. The model learns to perform its own alignment task without requiring external preprocessing or manual intervention, thereby improving training reliability while keeping the training process relatively simple and integrated within the model learning framework.
Data Source
AI summary
A processor-implemented method includes generating an adjusted reference patch by adjusting a position of a reference patch in a reference image based on a pixel value of a ground truth (GT) patch of a GT image and a pixel value of the reference patch, wherein the GT patch corresponds to a specific region of an input image; generating a super-resolution (SR) image of the input image using a SR model provided an input that is based on the generated adjusted reference patch; and training the SR model based on the SR image and the GT image.