Composite Image Mosaic Blending via Scale-Invariant Feature Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for creating mosaic images from partially overlapping images captured by flatbed devices face issues with incorrect blending and slow matching and blending speeds, leading to suboptimal image quality.
Innovation Solution
A new approach involving the analysis of input images to reveal matching points, determine descriptors, perform pair-wise comparisons, choose a reference image, recover rotation/translation parameters, and seamlessly integrate images with minimal seam visibility, using a fast matching algorithm that does not require user intervention and allows for arbitrary image placement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional feature matching methods (cross-correlation, Hu invariants) are used, then the alignment process can be performed, but the methods are not invariant to scale and rotation, leading to incorrect blending
Solution Approach 1:
The patent applies parameter changes by using scale-invariant and rotation-invariant feature descriptors (SIFT, SURF, ORB) to represent local image characteristics. These descriptors transform the feature extraction process to be invariant under scaling and rotation operations, allowing accurate matching even when images are captured at different scales or orientations. This resolves the contradiction by enabling reliable blending while maintaining precise alignment through robust feature representation.
Solution Approach 2:
The patent replaces traditional mechanical alignment methods (cross-correlation, simple invariant matching) with a computational geometry-based approach using homography transforms and RANSAC algorithm. This substitution enables the system to handle affine transformations and perspective distortions automatically, achieving both high alignment precision and blending accuracy without manual intervention.
2Ease of operation
If automated image stitching methods are used, then user input is eliminated, but the matching and blending process becomes slow
Solution Approach 1:
The patent applies preliminary action by pre-extracting and pre-computing feature descriptors (SIFT, SURF, ORB) from all input images before the matching process. It also pre-establishes a hierarchy of features and pre-computes descriptor hashes for rapid comparison. This preparation phase enables the subsequent matching and blending operations to proceed much faster, resolving the contradiction between automated operation and processing speed.
Solution Approach 2:
The patent segments the image matching process into distinct stages: feature extraction, feature matching, homography computation, and blending. Each stage operates independently and can be optimized separately. The segmentation allows parallel processing of different stages and enables the system to handle large numbers of images efficiently, maintaining both automation and productivity.
3Area of stationary object
If multiple images are stitched together to create a mosaic, then the image coverage area is expanded, but visible seams and artifacts appear
Solution Approach 1:
The patent introduces an intermediary blending mechanism that smoothly transitions between adjacent images. It uses blend masks with soft edges and gradient-based blending functions as intermediaries to bridge the gaps between stitched images. This intermediary layer masks the harsh boundaries and visible seams, creating a seamless composite image while maintaining expanded coverage area.
Solution Approach 2:
The patent applies local quality by adjusting blend masks and blending parameters according to the local characteristics of each image region. It computes different blend weights and transition functions for different areas of the mosaic, adapting the blending process to local content and geometry. This localized approach minimizes visible seams in critical areas while maintaining overall image quality and expanded coverage.
Data Source
AI summary
The method of creating image mosaics, comprising performing following operations: component elements of images are analyzed and descriptors of matching points are revealed; pair-wise comparison of descriptors is performed; descriptors are matched with final image and rotation/translation parameters are recovered; component elements one-by-one are complemented in final image, based on recovered rotation/translation parameters.