Panoramic Image Stitching Seam Artifact Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image stitching methods, particularly transform-based approaches, produce artifacts like ghosting or double impressions in panoramic images when dealing with seam regions that have depth variations, making it difficult to effectively handle overlap regions captured by multiple cameras.
Innovation Solution
A method that extracts the seam region from overlapping images, applies convergence values to generate strips, divides these strips into vertical segments, computes an optimal convergence using a distance matrix via dynamic programming, and applies a smoothness parameter to minimize cumulative error and guide convergence transitions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If transform-based stitching is used, then stitching speed is improved, but artifacts like ghosting appear in seam regions with depth variations
Solution Approach 1:
The patent divides the image into multiple strips along the seam region, with each strip processed independently using dynamic programming to find optimal convergence values. This segmentation allows the system to handle depth variations locally in each strip while maintaining overall stitching efficiency, resolving the contradiction between speed and accuracy by parallelizing processing across segments.
Solution Approach 2:
The patent dynamically adjusts convergence values for each strip based on local depth variations detected in the seam region. By changing the convergence parameter adaptively rather than applying a global transform, the system maintains stitching speed while eliminating artifacts caused by depth variations, thus resolving the accuracy-speed contradiction.
2Device complexity
If conventional stitching handles narrow seam regions, then processing complexity is reduced, but artifact minimization becomes difficult
Solution Approach 1:
The patent performs preliminary extraction of the seam region and identification of depth variations before applying convergence adjustments. This preliminary action prepares the narrow seam region for optimized processing, allowing the system to handle these challenging regions with appropriate complexity while minimizing artifacts, thus resolving the contradiction between processing complexity and artifact reduction.
3Manufacturing precision
If dynamic programming is applied to compute optimal convergence, then stitching accuracy is improved, but computational time increases
Solution Approach 1:
The patent applies dynamic programming separately to each strip rather than to the entire image at once. This segmentation of the computational task reduces the overall computational time while maintaining the accuracy benefits of dynamic programming, as each strip is processed independently and efficiently. This resolves the contradiction between optimization accuracy and computational time.
Solution Approach 2:
The patent applies dynamic programming only to the seam region strips where depth variations exist, rather than to the entire panoramic image. This partial action concentrates computational resources on the problematic areas, achieving optimal convergence where needed while minimizing overall computational time, thus resolving the accuracy-time contradiction.
Data Source
AI summary
A method, apparatus and computer program products are provided for reducing artifacts in a seam region when stitching overlapping images. One example method includes extracting the seam region from the overlapping images, wherein the overlapping images comprise a first image captured by a first image capturing device and a second image captured by a second image capturing device, the overlapping images sharing the seam region, the seam region being a region of each of the first image and the second images depicting a common captured area, applying a set of convergence values to the seam region generating a plurality of strips corresponding to the seam region, dividing the plurality of strips into multiple vertical segments, and computing an optimal convergence for each of the multiple vertical segments.