Image Blending via ROI-Based Weight Attenuation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image blending techniques face difficulties in aligning images with noise or active motion without decreasing the dynamic range or causing motion artifacts, especially when using optical flow techniques, which increase complexity and reduce matching levels.
Innovation Solution
A method involving region of interest (ROI)-based image alignment, detection of incompletely aligned regions, and blending using weights attenuated by spatial distance from the boundary, with additional ghost detection and cascaded image blenders to improve alignment and reduce artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If optical flow technique is used to solve motion artifacts, then motion artifacts are reduced, but computational complexity increases and matching level decreases
Solution Approach 1:
The patent divides the image into multiple regions of interest (ROIs) based on motion detection, treating different spatial regions differently. Motion-affected regions are identified and processed separately from stable regions, allowing selective application of blending techniques that reduce computational complexity while maintaining artifact reduction effectiveness.
Solution Approach 2:
The patent applies different blending weights and processing characteristics to different regions of the image. Regions with detected motion have different weight attenuation characteristics compared to stable regions, allowing optimized local processing that reduces overall computational complexity while maintaining quality where needed.
2Object-affected harmful factors
If optical flow technique is used to solve motion artifacts, then motion artifacts are reduced, but matching level decreases
Solution Approach 1:
The patent performs preliminary motion detection and ROI identification before the blending process. By pre-identifying which regions need special attention and calculating appropriate weights in advance, the system avoids the need for complex iterative optimization that would reduce matching level, instead using direct weight-based blending that preserves alignment accuracy.
Solution Approach 2:
The patent introduces weight maps as an intermediary between the aligned images and the final blended output. These weight maps, derived from motion detection results, mediate the blending process by selectively combining regions from different images, achieving artifact reduction without requiring complex re-alignment operations that would degrade matching precision.
3Use of energy by moving object
If images with noise or active motion are blended, then dynamic range is maintained, but motion artifacts increase
Solution Approach 1:
The patent dynamically adjusts blending weights based on detected motion characteristics. The weight attenuation is not fixed but adapts to the specific motion patterns detected in each ROI, allowing the system to maintain dynamic range while reducing motion artifacts by optimizing the blend in real-time based on actual motion content.
Solution Approach 2:
The patent changes the blending parameters (weights) based on the detected motion state and noise characteristics of different regions. By adjusting weight attenuation factors according to local motion intensity and noise levels, the system maintains dynamic range information while suppressing motion artifacts through parameter adaptation rather than fixed processing.
Data Source
AI summary
A method and apparatus may include aligning a reference image and an input image through ROI-based image alignment; detecting an ROI associated with an incompletely aligned region in the aligned image; and blending the aligned reference image and input image by using a plurality of weights which are attenuated according to a spatial distance from a boundary of the detected ROI. In addition, the method and apparatus may include aligning a reference image and an input image through image alignment based on an nth region of interest (ROI) which is detected at a previous stage; detecting another ROI associated with an incompletely aligned region in the aligned image; and blending the aligned image in the other ROI and an (n+1)th blended image, which is input from a next stage, in a region outside the other ROI, and outputting the last blended image to the previous stage.