Face Recognition

Overview

Face Recognition node detects faces in the input frame and matches each detected face against a library of known reference images.

Each match is scored by two complementary metrics — cosine similarity and L2-norm distance — so you can tune acceptance behavior for your use case. Use this node in identity verification flows, access control, or attendance tracking.

Input

Input Image

image required

The image frame to analyze. Connect this to a camera or upstream image output.

Known Faces

array required

Library of reference faces to match against. Each entry contains:

• img — reference image (base64 or image reference) showing the person's face.
• name — string label returned in the output when this face is matched.

Provide at least one clear, front-facing image per person for best accuracy.

Detection Model

string required

Face detection backend used to locate faces before recognition.

Values:

• ONNX_YUNET (default) — faster, recommended for most use cases.
• TFDNN — TensorFlow DNN-based detector.

Cosine Similarity Threshold

number required

Minimum cosine similarity score required to accept a match. Cosine similarity measures the angle between face feature vectors.

Range: 0.0 to 1.0
Default: 0.5

Higher values require closer feature alignment for a match to be accepted.

L2-Norm Similarity Threshold

number required

Maximum L2-norm distance allowed for an accepted match. L2-norm measures the Euclidean distance between feature vectors, so lower distances mean more similar faces.

Range: 0.0 to 2.0
Default: 1.0

Lower values require a closer match.

Overlay Results

boolean required

Whether to draw recognition results (matched name and bounding box) on the output frame. See Overlay Results.

Draw Landmarks

boolean required

When enabled, draws the five facial landmark points on each detected face in the overlay.

Output

Overlay Image

image

Output frame from the node. If overlays are enabled, each detected face is annotated with a bounding box and the matched name (or an "unknown" label if no match meets the threshold).

Recognized Faces

array

Array of recognition results. Each object contains:

• bbox array: Bounding box [x, y, width, height] of the detected face.
• name string: Name from the matched entry in Known Faces, or an empty string if unrecognized.
• cosineSimilarity number: Cosine similarity score of the best match.
• l2NormSimilarity number: L2-norm similarity score of the best match.