Image Stitching Path Bypassing Subjects to Eliminate Ghost Artifacts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image stitching methods often result in misaligned images and ghost images, especially when stitching real-site images with a common subject, due to errors in image registration algorithms and parallax, which can severely impair the visual quality of panoramas, particularly when the subject is a human face or other attention-catching features.
Innovation Solution
An image stitching method that calculates the optimal stitching path to bypass the subject using a subject detection algorithm and assigns a large error to the subject area, constructing a mask diagram to weight the fusion process, and performs localized registration and image deformation to ensure precise alignment of pixels in overlap areas, thereby avoiding misalignment and ghost images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional image registration algorithms are used to stitch overlap areas, then the stitching process can be completed, but misaligned images and ghost images occur due to registration errors and parallax
Solution Approach 1:
The patent segments the overlap area into subject area and non-subject area using a mask diagram. Different registration and fusion strategies are applied to each segment: the subject area uses localized registration with higher precision requirements, while the non-subject area uses standard registration. This segmentation allows the system to prioritize alignment precision for the subject area, reducing ghost images and misalignment artifacts in the final panorama.
Solution Approach 2:
The patent applies local quality by constructing a mask diagram that assigns different weights to different regions within the overlap area. The subject area is assigned higher weight for alignment precision, while other areas use standard weighting. This local differentiation ensures that the most visually important regions (subject areas) achieve superior alignment and minimal artifacts, directly improving the overall visual quality of the panorama.
2Productivity
If the stitching path passes through the subject area, then the fusion can be completed, but ghost images and misaligned images occur on the subject
Solution Approach 1:
The patent extracts the subject area from the overlap region by generating a mask diagram that identifies and isolates the subject. This extraction allows the stitching algorithm to take the subject area out of the standard fusion path and apply specialized localized registration and fusion operations specifically to this region. By separating the subject processing from the general stitching workflow, the system maintains high processing efficiency while achieving superior alignment precision for the subject area, eliminating ghost images without sacrificing overall stitching productivity.
3Device complexity
If standard fusion algorithm is used without mask weighting, then the stitching process is simple, but pixels in subject areas do not align perfectly
Solution Approach 1:
The patent performs preliminary action by pre-generating a mask diagram that identifies subject areas before the fusion process begins. This mask diagram is then used to weight the fusion operation, ensuring that subject pixels receive higher priority for precise alignment. By preparing the mask diagram in advance, the system integrates precision enhancement into the existing fusion workflow without requiring a complete redesign of the stitching algorithm, thus maintaining relative simplicity while achieving superior pixel alignment precision in subject areas.
Data Source
AI summary
The present invention is to provide a method for ensuring perfect stitching of a subject's images in a real-site image stitching operation, which enables an electronic device to read two real-site images which are taken respectively of different parts of the same site and each of which has an overlap area of a common subject; to determine and choose an optimal stitching path bypassing the subject; to construct a mask diagram such that, within a difference diagram of the overlap areas, the farther a pixel on one side of the optimal stitching path is from the optimal stitching path, the greater the pixel's value, and the farther a pixel on the other side of the optimal stitching path is from the optimal stitching path, the smaller the pixel's value; and to stitch and fuse the overlap areas together through a fusion algorithm, with the mask diagram serving as a weight.