Image Alignment via Translation Invariant Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image alignment techniques are computationally expensive and difficult to achieve in real time, especially for panoramic images captured with wide field view lenses, which suffer from barrel distortion and require robust stabilization without additional components like gyroscopic sensors.
Innovation Solution
A computer-implemented method and system that uses translation invariant feature matching to align multiple overlapping images by determining local maxima and minima pixel points, performing iterative intra and inter image correlation, and selecting matching feature points to compensate for translation and rotation, thereby creating a panoramic or stabilized image efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image alignment techniques are used, then alignment accuracy is improved, but computational cost increases and real-time processing becomes difficult
Solution Approach 1:
The patent segments the image alignment process into distinct stages: feature detection (identifying corners and edges), feature matching (finding corresponding points between images), and transformation calculation (computing alignment parameters). This segmentation allows each stage to be optimized independently, reducing overall computational complexity while maintaining alignment accuracy.
Solution Approach 2:
The patent extracts key feature points (corners, edges, landmarks) from images to represent the entire image content. By working with these extracted features rather than full-image pixel data, the computational burden is dramatically reduced, enabling real-time processing while preserving alignment precision through careful feature selection and matching.
2Area of stationary object
If wide field view lenses are used to capture larger area, then coverage is improved, but image resolution deteriorates and barrel distortion increases
Solution Approach 1:
The patent applies local quality by detecting and matching features in different regions of the image with region-specific characteristics. Corner detection operates differently from edge detection, and features in the center of the image are treated differently from those at the periphery where distortion is more severe. This allows the system to maintain accurate alignment despite varying local image quality caused by wide-angle distortion.
3Reliability
If additional components like gyroscopic sensors are added for image stabilization, then stabilization accuracy is improved, but device complexity and cost increase
Solution Approach 1:
The patent replaces mechanical stabilization systems (gyroscopic sensors, floating lens elements) with a computational approach using image feature matching and transformation algorithms. By substituting mechanical complexity with software-based image processing, the system achieves comparable stabilization accuracy without adding physical components, thereby reducing device complexity, cost, and power consumption.
Data Source
AI summary
A computer implemented method and system is provided for aligning multiple overlapping images in real time using translation invariant feature matching. A user captures overlapping images comprising a first image and a second image using one or more image capture devices. An image aligning application determines one or more local maxima pixel points and local minima pixel points in the first image and the second image based on predetermined statistical criteria. The image aligning application performs iterative intra image correlation in the first image for selecting a predetermined number of feature points. The image aligning application performs iterative inter image correlation for the selected feature points, for determining a predetermined number of best correlated feature point pairs, and selects a matching feature point pair from the best correlated feature point pairs. The image aligning application aligns the first image and the second image using the selected matching feature point pair.