Perspective Preserving Image Stitching via Local Homography
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image stitching techniques face challenges in seamlessly blending overlapping images, especially in the presence of parallax and lens distortion, and struggle to maintain natural appearance and accuracy, particularly in non-overlapping regions.
Innovation Solution
A method that combines local homography and global similarity transformations, linearizing homography in non-overlapping regions and smoothly transitioning to global similarity, using weighted linear transformations to mitigate perspective distortion and create a fully continuous stitching field.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional image stitching techniques are used to blend overlapping images, then seamless panorama creation is achieved, but perspective distortion and parallax errors occur in non-overlapping regions
Solution Approach 1:
The patent applies different transformation models to different regions of the image: local homography for overlapping regions and global similarity transformation for non-overlapping regions. This region-specific approach allows each area to be processed with the most appropriate transformation, reducing perspective distortion while maintaining stitching accuracy.
Solution Approach 2:
The patent segments the image into overlapping and non-overlapping regions, applying different stitching strategies to each segment. This segmentation allows the system to handle perspective distortion in non-overlapping regions separately from the seamless blending requirements of overlapping regions.
2Reliability
If local homography is used for image stitching, then overlapping regions blend seamlessly, but parameter selection becomes complex and manual tuning is required
Solution Approach 1:
The patent implements automatic parameter selection where the system autonomously determines the optimal transformation model and parameters for each region based on image content analysis. This self-service approach eliminates manual parameter tuning while maintaining reliable seamless blending in overlapping regions.
Solution Approach 2:
The patent dynamically changes transformation parameters based on the specific characteristics of each image region. By automatically adjusting parameters such as transformation type and weighting factors, the system achieves reliable blending without requiring complex manual parameter selection.
3Object-affected harmful factors
If global similarity transformation is applied to entire image, then perspective distortion is reduced, but local variations and parallax effects are not accounted for
Solution Approach 1:
The patent segments the transformation application into global similarity for non-overlapping regions and local homography for overlapping regions. This segmentation ensures that global perspective correction is applied where needed while preserving local alignment accuracy in regions where multiple images provide detailed information.
Solution Approach 2:
The patent applies different transformation qualities to different regions: global similarity transformation provides perspective correction in non-overlapping regions, while local homography provides precise alignment in overlapping regions. This local quality differentiation resolves the contradiction between global perspective correction and local alignment precision.
Data Source
AI summary
A method and system of stitching a plurality of image views of a scene, including grouping matched points of interest in a plurality of groups, and determining a similarity transformation with smallest rotation angle for each grouping of the matched points. The method further includes generating virtual matching points on non-overlapping area of the plurality of image views and generating virtual matching points on overlapping area for each of the plurality of image views.