Super-resolution Image Reconstruction via Inverse Problem Solving
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing super-resolution image processing methods, such as interpolation-type approaches, provide limited digital magnification and are not effective in enhancing low-resolution images beyond a certain level, often resulting in images with low spatial resolution due to a low number of photodetectors in camera focal plane arrays.
Innovation Solution
The proposed system and method utilize inverse problem solving, including image registration and reconstruction processes, specifically using back-projection and inverse filtering steps, to convert a set of low-resolution images into high-resolution images, with the ability to project low-resolution images onto a high-resolution grid and remove back-projection effects to achieve improved image resolution.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If interpolation-type image processing is used to combine low-resolution images, then image resolution is improved to some extent, but digital magnification is limited to less than 2×
Solution Approach 1:
Instead of using interpolation to enhance resolution (forward approach), the patent applies inverse problem solving and deconvolution methods to recover the original high-resolution image from low-resolution observations. This inversion approach allows achieving digital magnification up to 20× by mathematically reversing the blurring and downsampling processes that created the low-resolution images.
Solution Approach 2:
The patent changes the mathematical parameters and algorithms used in image processing. Rather than relying on simple interpolation formulas, the system employs iterative deconvolution algorithms, regularization techniques, and inverse filtering that fundamentally alter how resolution enhancement is achieved, enabling much higher magnification factors.
2Device complexity
If the number of photodetectors in camera focal plane array is reduced, then device complexity and cost are reduced, but spatial resolution becomes low
Solution Approach 1:
The patent creates multiple copies of the same scene from different positions and combines them through computational processing. By capturing the same object from multiple viewpoints with the low-resolution camera and then fusing these copies through inverse problem solving, the system achieves high-resolution output without requiring high-resolution hardware.
Solution Approach 2:
The patent transitions from spatial dimension (physical photodetector density) to computational dimension (image processing operations). Instead of improving resolution through hardware density, the system uses multiple low-resolution images captured from different positions and processes them through deconvolution algorithms to achieve high-resolution results in the computational domain.
3Measurement precision
If multiple low-resolution images are combined through back-projection, then image resolution is enhanced, but computational complexity increases
Solution Approach 1:
The patent employs iterative deconvolution algorithms that continuously refine the image reconstruction through multiple processing steps. Each iteration progressively improves the resolution by removing back-projection effects and enhancing fine details, maintaining useful computational action throughout the process to achieve convergence on a high-resolution image.
Solution Approach 2:
The system incorporates feedback mechanisms in its iterative deconvolution process, where the results of each processing stage are fed back into subsequent iterations for refinement. This feedback loop allows the algorithm to progressively improve the image quality and resolution while managing computational complexity through controlled iteration.
Data Source
AI summary
Exemplary super-resolution methods and systems may generate, or create, a super-resolution based on a plurality of low-resolution images. Such exemplary methods and systems may utilize image registration and back-projection to provide intermediate imaging data, and then use inverse problem solving to remove any back-projection effects as well as noise to generate a super-resolution image.