active_face_liveness

Releases

1. ▸▾0.2.02 Oct 2026

   Add opt-in built-in mouth opening and detector-backed tongue challenges, including rest/action/return validation, per-gesture timeouts, typed evidence

   • Add opt-in built-in mouth opening and detector-backed tongue challenges, including rest/action/return validation, per-gesture timeouts, typed evidence, localized prompts and an example mouth toggle. Both default to disabled; tongue requires an application-supplied model adapter.

   Behavior changes

   • Migration: direct classifyLiveness callers must explicitly pass all required evidence flags. Omitted flags now default to false. Incomplete scans return unknown rather than photo.
   • A sustained squint no longer satisfies the blink challenge; require a full open–closed–open sequence. Optional mouth/tongue challenges default to disabled.

   Reliability improvements

   • Fail closed on omitted liveness evidence; classify incomplete scans as unknown.
   • Require continuously observed head holds and invalidate interrupted sessions.
   • Use monotonic camera receipt times and rendered pulse windows, timer-driven illumination, stale-frame rejection and bounded inference draining.
   • Require an open–closed–open blink on the same signal; remove squint-only pass.
   • Track RGB skin patches with landmarks and camera rotation; correlate per-frame colour responses with equal weighting per pulse.
   • Limit replay-loop evidence to dynamic windows during head movement.
   • Extract pure Dart evidence primitives and add regression scenarios.
   • Document heuristic-score limits and add a labelled-session evaluation tool.

   Integration improvements

   • Added the one-line ActiveFaceLiveness.verify privacy-first API.
   • Added stable failure codes, public scan steps, and step callbacks.
   • Added configurable scanner colours, visibility controls, and header labels.
   • Added a translucent outside-face-guide overlay.
   • Added runtime configuration validation and privacy-safe JSON serialization.
   • Added a complete integration example.

   Validation

   • Static analysis passes and 33 pure-Dart regression scenarios pass.
   • Full Flutter tests remain blocked by the macOS dartcv4/OpenCV native build.
   • Detection thresholds and accuracy still require labelled real-device trials; the combined score is an uncalibrated heuristic, not a probability.