BlockMatching aims to provide a sophisticated implementation on common block matching
algorithms for image processing and
computer vision tasks. Block matching is a data and computational intense algorithm, performance is
of high priority for this package.
🚧 This is still a WIP project.
Two functions are provided as the standard API:
best_match: finds the best matching candidate. This is also known as nearest neighbor search.multi_match: sort the similarities of all candidates and return the smallest K results. This is sometimes known as K nearest neighbor search or top-k selection.
Available block matching strategies:
FullSearch(brute force): search among all possible candidates. This gives the most accurate result but is computationally intensive. CUDA is supported for commonly used distances defined in Distances.jl.