HemiSec

HemiSec, our mobile biometric scanner turns a quick palm scan into a vein-feature signature for fast, spoof-resistant authentication on everyday hardware to secure your sensitive information.

Comment

Inspiration

We wanted a biometric that’s hard to spoof, works fast, and is portable. Palm veins are compelling because they’re sub-surface (harder to fake than fingerprints) and can be captured with inexpensive near-IR/monochrome setups. Our goal was to prove you can build a practical palm-vein authenticator with a lightweight vision pipeline and a clean enrollment/verification flow.

What it does

Palm Vein Mobile Biometric Scanner turns a palm image into a vessel-enhanced feature map, extracts a compact feature representation, and uses it for authentication:

  • Enroll: capture a few samples, build a stable template
  • Verify: capture a new sample, compute similarity, accept/reject
  • Debug visualization: optional red overlay tracing detected vessels for rapid tuning

How we built it

Vision pipeline

  • Hand/palm segmentation: threshold + connected components to isolate the palm region and ignore background.
  • Illumination correction: background subtraction / high-pass filtering to remove slow lighting gradients.
  • Local contrast enhancement: CLAHE to bring out subtle vein contrast.
  • Vessel enhancement: multi-orientation Gabor filter bank (and multi-scale variants when useful) to produce a vessel-likelihood response map.
  • Final product feature map: the vessel-enhanced response map is what we feed into feature extraction and matching.

Authentication logic

  • Template creation: aggregate multiple enrollment captures into a robust reference representation.
  • Matching: compare verification features to enrolled template(s) using a similarity score (e.g., cosine similarity / normalized correlation) with a tunable threshold.

Engineering

  • CLI pipeline that outputs the final feature map + intermediate stages for debugging.
  • Parameter tuning guided by quick visualization and rough hand-marked vein traces.

Challenges we ran into

  • Noise vs. sensitivity tradeoff: making vessels “pop” can also amplify skin texture and artifacts.
  • Illumination variation: small changes in lighting/exposure can dominate subtle vein contrast.
  • Scale mismatch: veins vary in width, so a single filter scale can miss detail or overfit to texture.
  • No ground truth: debugging required strong visual tooling and careful iteration.

Accomplishments that we're proud of

  • Produced a consistent vessel-enhanced feature map that highlights vein structure from raw palm images.
  • Built an end-to-end enrollment → verification loop that demonstrates real authentication behavior.
  • Created transparent debug outputs (overlay + intermediate saves) that made iteration fast.

What we learned

  • Preprocessing (segmentation + illumination correction) largely determines success.
  • Vessel enhancement is about structure, not just contrast—orientation-selective ridge filters work well when tuned properly.
  • Once the feature map is reliable, downstream steps (template building, similarity metrics, thresholds) become much easier to reason about.

What's next for Palm Vein Mobile Biometric Scanner

  • Robustness: exposure control, session-to-session normalization, improved artifact suppression.
  • Template stability: quality checks and better aggregation across enrollment samples.
  • Better matching: evaluate stronger descriptors/embeddings on top of the vessel feature map.
  • Mobile deployment: optimize runtime and integrate into a clean UX for enrollment + verification.
  • Security: explore liveness/spoof-resistance checks and multi-factor flows.

Built With

Share this project:

Updates