Image Processing Device Resolving Resolution-Artifact Trade-off
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques that combine multiple images to increase resolution often result in artifacts in regions where the subject has moved, due to folding noise and difficulty in accurately calculating correlations in stationary regions with fine patterns.
Innovation Solution
An image processing device and method that aligns and combines multiple time-series images in high and low-resolution image spaces, calculates feature quantities between low-resolution combined images, and corrects high-resolution images based on these quantities to suppress artifacts and enhance resolution in stationary regions while handling moving subjects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple time-series images are combined in a high-resolution image space to increase resolution, then folding noise is removed and high-frequency components are reproduced, but artifacts occur in regions where the subject has moved
Solution Approach 1:
The patent divides the image processing into multiple resolution levels. It segments the combination process into low-resolution preprocessing and high-resolution final combination. The low-resolution combined images are used to identify moving regions, while the high-resolution combination is applied primarily to stationary regions, thus preventing artifacts in moving areas while maintaining high resolution in stationary areas.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image based on their motion characteristics. Stationary regions undergo high-resolution combination to maximize detail, while moving regions are handled differently to avoid artifacts. This local differentiation resolves the contradiction by optimizing each region according to its specific characteristics.
2Object-generated harmful factors
If the degree of similarity between images is calculated to control combining ratios, then artifacts from subject motion are prevented, but the processing complexity increases
Solution Approach 1:
The patent segments the processing into two stages: first generating low-resolution combined images to assess motion, then using these to guide the high-resolution combination. This segmentation reduces complexity by performing the computationally intensive similarity calculations at low resolution, where fewer pixels require processing.
Solution Approach 2:
The patent performs similarity calculations partially - only at low resolution and only to the extent needed to identify moving regions. Full high-resolution similarity calculations are avoided, as the low-resolution assessment is sufficient for determining which regions require artifact suppression.
3Measurement precision
If low-resolution combined images are generated by dividing time-series images into groups, then correlation calculation accuracy is improved in stationary regions, but additional processing steps are required
Solution Approach 1:
The patent performs preliminary processing by generating low-resolution combined images from grouped time-series images before the final high-resolution combination. This preliminary action prepares the data by removing folding noise and establishing baseline correlations, making the subsequent high-resolution processing more accurate and efficient.
Solution Approach 2:
The patent transitions from high-resolution processing directly to a two-dimensional approach: first reducing to low-resolution dimensions for accurate correlation assessment, then returning to high-resolution for final output. This dimensional change allows accurate measurement without the computational burden of high-resolution processing throughout.
Data Source
AI summary
An image processing device includes one or more processors configured to: generate a high-resolution combined image by aligning the plurality of images with each other in a high-resolution image space based on an amount of displacement between the plurality of images, and combining the plurality of images; generate at least two low-resolution combined images by generating at least two groups each composed of at least two images by dividing the plurality of images in the time direction, aligning the at least two images in each of the groups with each other in a low-resolution image space based on the amount of displacement, and combining the at least two images through weighted addition; calculate, in each region, a feature quantity pertaining to a correlation between the generated at least two low-resolution combined images; and correct the high-resolution combined image based on the calculated feature quantity.