FLANN Matcher for Feature Correspondence

WHY IT EXISTS: Feature matching determines whether a query image appears inside a training image by comparing local descriptors. OpenCV solves this with matcher objects that take two descriptor sets and produce correspondences.

THE MENTAL MODEL: A matcher acts like a clerk who receives a descriptor from the query image and finds the best corresponding descriptor in the training image. The clerk returns a ticket called DMatch that records the distance between the pair, the index in the training set, the index in the query set, and which training image was used.

HOW IT WORKS: OpenCV offers two matchers: Brute-Force and FLANN. The source details Brute-Force as an exhaustive method that compares each query descriptor against every training descriptor using a distance norm. For binary string descriptors such as ORB, BRIEF, and BRISK, the source specifies Hamming distance. For floating-point descriptors such as SIFT and SURF, it specifies L2 norm. FLANN is introduced as the alternative matcher for the same feature matching pipeline. The source documents that matchers expose match and knnMatch methods. The match method returns the best correspondence for each query descriptor. The knnMatch method returns k best matches, which supports additional filtering like ratio tests. The output structure described is the DMatch object, which contains four attributes: distance measuring descriptor similarity with lower values being better, trainIdx for the training descriptor index, queryIdx for the query descriptor index, and imgIdx for the training image identifier.

WHEN TO USE IT: Use FLANN or the alternative Brute-Force matcher whenever you need to establish keypoint correspondences between two images in OpenCV, such as object detection, image stitching, or tracking pipelines.

WHEN NOT TO USE IT: Do not use a matcher if descriptors are not yet computed or if images lack sufficient texture to produce stable keypoints, because the pipeline depends on meaningful descriptor inputs.

ONE CANONICAL EXAMPLE: The source demonstrates Brute-Force matching with ORB descriptors between box.png and box_in_scene.png. After detecting keypoints and computing descriptors, the engineer creates a BFMatcher with NORM_HAMMING and crossCheck enabled, calls match, sorts the resulting DMatch list by ascending distance, and draws the top ten matches. FLANN is presented as the alternative matcher choice within this same OpenCV pipeline.