Super-Resolution Image Generation Using Optical Flow Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating super-resolution (SR) images in non-visible spectral ranges face challenges in increasing image resolution without enhancing detector resolution, particularly when dealing with moving or changing objects, which can result in artifacts and require discarding image sequences.
Innovation Solution
The method calculates optical flow for individual images, segments them based on displacement vector fields, and optimizes the SR image using separate variation parameters for each segment, allowing adaptation to varying image content and improving robustness by masking irrelevant sections.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the resolution of the detector is increased to improve image resolution, then the image resolution is improved, but the production costs increase significantly
Solution Approach 1:
The patent divides the image into multiple segments or regions, and applies different processing strategies to different segments. This allows the system to focus computational resources on critical areas while using simpler methods for other areas, achieving high resolution without requiring a high-resolution detector across the entire image.
Solution Approach 2:
The patent changes parameters such as exposure time, illumination intensity, or processing algorithms to achieve better image quality without improving detector resolution. By adjusting these parameters, the system can extract more information from the same detector hardware.
2Measurement precision
If small deviations between individual images are used to generate SR images, then higher image resolution is achieved, but artifacts appear when objects move or change significantly
Solution Approach 1:
The patent applies different processing qualities or methods to different regions of the image based on local characteristics. For regions with significant object changes, more robust methods are applied, while for static regions, standard SR techniques are used. This local adaptation maintains reliability across varying conditions.
Solution Approach 2:
The patent introduces dynamic adaptation in the SR generation process, where the processing parameters and methods are adjusted based on the detected motion or changes in the scene. This allows the system to respond to object movements and changes, maintaining reliability while achieving high resolution.
3Device complexity
If a single optimization method is applied to all image areas, then the process is simple, but image areas with high variability are not processed optimally
Solution Approach 1:
The patent segments the image into multiple regions with different characteristics, and applies tailored optimization methods to each segment. This segmentation approach balances complexity and quality by focusing advanced processing only where needed.
Solution Approach 2:
The patent implements local quality optimization by adapting the processing method to the specific characteristics of each image region. This allows high-quality processing for variable regions while maintaining simplicity for uniform regions, achieving an optimal balance between complexity and precision.
Data Source
Figure 1
Figure 2~3
AI summary
In the case of a measuring device for recording a sequence of individual images (6, 7, 8) in a non-visible spectral range, a method for generating an SR image (11) having an image resolution that is higher than an image resolution of the individual images (6, 7, 8) is proposed, wherein, for the individual images (6, 7, 8), a displacement vector field (11, 12) is determined with a calculation of the optical flow and the individual images (6, 7, 8) are segmented into segments (20, 21, 22, 23, 24, 25) with regard to the values of the displacement vector field (11, 12), wherein an optimization method is carried out for calculating the SR image (11) from the individual images (6, 7, 8) with variation parameters individually assigned to the segments (20, 21, 22, 23, 24, 25).