Nearest Neighbor Field Motion Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional algorithms for computing nearest neighbor fields between images are computationally expensive and fail to enforce spatial coherency, leading to noisy estimates of optical flow and incorrect motion fields, especially when large motions are involved.
Innovation Solution
A method that determines matching patches between two digital images using a nearest neighbor field algorithm, followed by a motion optimization algorithm to identify predominant motion components based on statistics, which are then used to generate a spatially coherent motion field, accounting for large motions and eliminating noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional exhaustive search algorithms are used to compute nearest neighbor fields, then matching accuracy is improved, but computational cost and time consumption increase significantly
Solution Approach 1:
The patent segments the exhaustive search process by organizing patches into hierarchical levels. Instead of comparing every patch against all possible matches, the algorithm divides the search space into coarse-to-fine levels, where each level processes a subset of patches. This segmentation reduces the computational complexity from O(N²) to approximately O(N log N) while preserving matching accuracy through progressive refinement.
Solution Approach 2:
The patent applies preliminary action by pre-sorting and pre-organizing patches based on their visual characteristics and spatial relationships before the actual matching process. Patches are grouped by similarity metrics and spatial coherence properties in advance, allowing the algorithm to quickly eliminate non-matching candidates without performing exhaustive comparisons. This preliminary organization significantly reduces the search space while maintaining accuracy.
2Stability of the object's composition
If conventional optical flow algorithms are used to compute motion fields, then spatial coherency is enforced, but the algorithms fail to handle large motions and produce noisy estimates
Solution Approach 1:
The patent applies dynamics by making the matching process adaptive to local motion characteristics. Instead of assuming small displacements as conventional optical flow algorithms do, the system dynamically adjusts the search window size and matching criteria based on the estimated motion magnitude in each region. This allows the algorithm to handle both small and large motions appropriately, maintaining spatial coherency while improving accuracy for large displacements.
Solution Approach 2:
The patent changes key parameters of the matching algorithm based on local image characteristics and motion estimates. The search window size, patch comparison metrics, and coherence enforcement strength are adjusted dynamically according to the local content and motion magnitude. This parameter adaptation allows the system to maintain spatial coherency in static regions while accurately tracking large motions in dynamic regions, resolving the contradiction between coherency and accuracy.
3Productivity
If reconstruction-based algorithms are used to compute nearest neighbor fields, then computational efficiency is improved, but spatial coherency of matching patches is not enforced
Solution Approach 1:
The patent introduces spatial coherence constraints as an intermediary mechanism that mediates between the efficient reconstruction-based matching and the requirement for spatially coherent results. The algorithm first performs efficient patch matching using reconstruction techniques, then applies a coherence enforcement step that adjusts matches to ensure spatial consistency. This intermediary coherence check corrects mismatches while preserving the computational efficiency of the underlying reconstruction algorithm.
Data Source
AI summary
In embodiments of statistics of nearest neighbor fields, matching patches of a nearest neighbor field can be determined at image grid locations of a first digital image and a second digital image. A motion field can then be determined based on motion data of the matching patches. Predominant motion components of the motion field can be determined based on statistics of the motion data to generate a final motion field. The predominant motion components correspond to a motion of objects as represented by a displacement between the first and second digital images. One of the predominant motion components can then be assigned to each of the matching patches to optimize the final motion field of the matching patches.