Image Correspondence Verification Using Sliding Histogram Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image evaluation methods for deriving correspondences are memory-, computationally, and/or time-intensive, particularly with high-resolution images, necessitating a reduction in memory accesses, additions, and multiplications to accelerate the process.
Innovation Solution
A method involving the creation of a histogram-based evaluation of correspondence hypotheses, where a sliding window is used to update histograms for each element in the hypothesis matrix, allowing for conditional verification of hypotheses by examining their environment, and confirming or rejecting them based on threshold comparisons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional correspondence verification methods are used for high-resolution images, then measurement precision is improved, but productivity deteriorates due to high computational intensity and time consumption
Solution Approach 1:
The patent segments the verification process into two distinct stages: (1) a fast coarse verification stage that quickly eliminates obviously incorrect correspondence hypotheses using simplified criteria, and (2) a precise fine verification stage that applies rigorous verification only to the remaining candidate hypotheses. This segmentation reduces the overall computational burden while maintaining verification accuracy for high-resolution images.
Solution Approach 2:
The patent applies partial verification actions by selectively verifying only those correspondence hypotheses that meet preliminary criteria, rather than verifying all hypotheses with equal thoroughness. This allows the system to achieve sufficient verification accuracy for critical hypotheses while skipping or simplifying verification for less critical ones, thereby improving overall processing productivity.
2Reliability
If exhaustive correspondence verification is performed, then reliability is improved, but loss of time increases due to computational intensity
Solution Approach 1:
The patent performs preliminary filtering of correspondence hypotheses based on easily computable criteria such as brightness consistency and basic geometric constraints before conducting time-consuming verification. This preliminary action eliminates obviously incorrect hypotheses early in the process, ensuring that subsequent verification efforts are focused only on promising candidates, thus maintaining reliability while reducing total verification time.
Solution Approach 2:
The patent implements a continuous verification pipeline where hypotheses are verified in stages with results fed back into the process. Rather than performing discrete exhaustive verification batches, the system continuously refines and verifies hypotheses, maintaining useful action throughout the processing time and reducing overall loss of time while preserving reliability through iterative improvement.
3Measurement precision
If detailed environmental evaluation is performed for each correspondence hypothesis, then measurement precision is improved, but use of energy increases due to additional computational operations
Solution Approach 1:
The patent applies local quality evaluation by assessing the environmental context of each correspondence hypothesis individually based on its specific characteristics and location in the image. Rather than applying uniform detailed evaluation to all hypotheses, the system adapts the evaluation depth and type to the local requirements of each hypothesis, improving measurement precision where needed while reducing energy consumption in regions where simpler evaluation suffices.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method (S) for evaluating images (B1, B2) and in particular for evaluating correspondence hypotheses ([u, v]) of images (B1, B2), comprising the steps of (i) providing (S1) a hypothesis matrix (10) of correspondence hypotheses ([u, v]) between first and second images (B1, B2) respectively given as a corresponding image matrix, (ii) evaluating (S2) the hypothesis matrix (10) and, as required, verifying (S3) the image correspondence hypotheses ([u, v]) and (iii) providing (S4) verified image correspondence hypotheses ([u, v]) in a correspondence matrix (100) of image correspondences as evaluation result, wherein the evaluation (S2) of the hypothesis matrix (10) takes place in that for each element (1) of the hypothesis matrix (10) for at least one component (u, v) of the correspondence hypotheses ([u, v]) a histogram (63, 64) relating to the values of the component (u, v) is formed (S2-1) and evaluated (S2-2).