--- title: "Fingerprint API" description: "API reference for spectral fingerprinting, verification, and similarity search." published: 2026-02-18T18:06:47.821654+00:00 updated: 2026-02-18T18:06:47.821654+00:00 tags: ["api", "fingerprinting"] url: https://xiobjects.com/docs/xio/api/fingerprinting source: XI Objects --- # Fingerprint API ## Overview The fingerprint API provides image and video fingerprinting, composite feature verification, and similarity search via the Orbital network. ## Fingerprint Extraction XI Objects generates fingerprint vectors from image content through a dual-pipeline approach: - **Forensic pipeline (LWA)**: Luminance Waveform Analysis for manipulation detection and coherence scoring - **Discriminative pipeline**: Perceptual hashing, color, spatial, and block-level analysis for content identity matching ### Result A fingerprint extraction produces: | Field | Description | |-------|-------------| | Forensic fingerprint | Waveform-based fingerprint for manipulation detection | | Discriminative fingerprint | Multi-signal identity fingerprint with search vector | | Composite feature set | Combined descriptor for detailed verification | | Source dimensions | Original image width and height | | Extraction time | Processing duration | ## Composite Feature Set (Phase 2) For high-confidence verification, the composite feature set provides detailed structural comparison across four independent descriptors: ### Components The composite feature set uses the forensic and discriminative pipelines together to provide high-confidence structural comparison. The forensic pipeline detects modifications through waveform coherence analysis, while the discriminative pipeline confirms content identity through perceptual hashing, color distribution, spatial layout, and block-level analysis. ## Fingerprint Comparison ### Phase 1: Cosine Similarity Spectral fingerprints are compared via cosine similarity, returning a score from 0.0 to 1.0: | Score | Interpretation | |-------|---------------| | ≥ 0.85 | High confidence match | | 0.60 – 0.85 | Likely derived from same source | | < 0.60 | Content does not match | The comparison result includes cosine similarity, distance metrics, and a match determination based on the configured threshold. ### Phase 2: Unified Verification The unified verification provides a combined assessment: | Field | Description | |-------|-------------| | Verified | Whether content is verified | | Combined score | Weighted composite score (0.0–1.0) | | Status | Classification of result | | Phase 1 similarity | Spectral similarity score | | Phase 2 result | Agreement ratio and per-component scores | ### Verification Status Values | Status | Description | |--------|-------------| | `Verified` | Strong match across all components | | `ProbableMatch` | High confidence match with minor component variation | | `Inconclusive` | Insufficient evidence to determine | | `PossibleManipulation` | Components disagree; potential tampering detected | | `VerificationFailed` | Content does not match | ## Similarity Search ### Orbital Endpoint ```http POST /search/fingerprint Content-Type: application/json ``` Search the Orbital network for fingerprints similar to a given vector. Returns ranked matches with cosine similarity scores. ## Video Fingerprinting The video fingerprinting library provides dual-track (visual + audio) analysis for video identification, comparison, and manipulation detection: - **Forensic fingerprinting**: Frequency-domain waveforms from video frames and audio with hierarchical aggregation (window → segment → master) - **Discriminative fingerprinting**: Per-frame perceptual hashing, color histogram, and spatial grid analysis for content identity matching - **Forensic analysis**: A/V sync analysis, temporal coherence, compression generation estimation, and motion analysis - **Comparison**: Temporal alignment (landmark + cross-correlation), delta detection with modification type classification, and transform-aware identity matching