Module features 

Corner detection and binary descriptors on analysis-resolution grey images.

The scoring is the minimum eigenvalue of the 3×3 structure tensor computed from exact integer rolling row sums; ties and ordering follow the historical implementation so keyframe words, matches and downstream motion hypotheses are reproducible across releases.

Structs

Feature

Match

Constants

DESCRIPTOR_WORDS

MAX_FEATURES_DEFAULT

Functions

extract_features

Spatially balanced minimum-eigenvalue corners with 256-bit BRIEF descriptors, strongest first.

feature_words

Four independent 12-bit vocabulary bands per descriptor; deduplicated in first-seen order.

hamming

match_features

Mutual-nearest distinct matches plus explicitly flagged ambiguous alternatives, in a order.

smooth

3×3 binomial blur; the one-pixel border is copied unchanged.