BFMatcher

BFMatcher, short for Brute-Force Matcher, is a simple descriptor matcher in OpenCV’s features2d module. It pairs descriptors by exhaustively computing distances between every descriptor in one set and every descriptor in another, selecting the best matches according to a chosen distance metric. This makes it a straightforward baseline method for matching features across images.

BFMatcher supports different descriptor types by using corresponding distance norms. For floating-point descriptors such as SIFT or SURF, the L2 (Euclidean) distance is common, while binary descriptors like ORB, BRISK, or BRIEF use Hamming distances. The distance metric is specified at construction via normType, for example NORM_L2 or NORM_HAMMING. A crossCheck flag can be enabled to enforce mutual best matches, which improves precision at the cost of potentially fewer matches.

In practice, BFMatcher is used with either match to find the closest neighbor for each descriptor, or knnMatch to retrieve the k nearest neighbors for subsequent filtering, such as Lowe’s ratio test, to improve match quality. Compared to more complex or approximate matchers like FLANN, BFMatcher is exact and simple but can be slower on large descriptor sets. It remains a common choice in traditional feature-based pipelines that use detectors such as SIFT, SURF, and ORB.