Ferns
lastmod 2026-11-10created 2022-07-122 edits
Ferns (random ferns) is a method that solves keypoint recognition by recasting it as a classification problem: it uses many small groups of binary tests (ferns), each combining the results of brightness comparisons between pixel pairs, and combines them in a naive Bayes fashion. Training takes time, but recognition is very fast, so it has been used for real-time object tracking and augmented reality; it can be seen as a simplified variant of random trees.