Skip to content
FaceOff Technologies

Deep Fraud Guard

Forensic-grade deepfake, document-forgery & identity-verification platform. Combines AI image detection, document tampering forensics, and Fraud ID Guard for Aadhaar & PAN authentication.

Request a demoOpen DeepFraudGuard Console
Live Product Walkthrough

Platform Forensics

Support & Compliance: Confidential Enterprise System Architecture & Continuous Feedback Learning Loop
Engine A: Deepfake Image Detection (REAL / FAKE)
Engine B: Document Forgery Detection (AUTHENTIC / SUSPICIOUS / FORGED)
Fraud ID Guard: Aadhaar & PAN Verification (AUTHENTIC / SUSPICIOUS / UNUSUAL)
Explainable Evidence: ELA recompression maps & attention overlays
Developer REST API & Asynchronous Job Queues
Core forensic engines: Deepfake Image, Document Forgery & Fraud ID Guard

3

Core forensic engines: Deepfake Image, Document Forgery & Fraud ID Guard

Supported media uploads across JPG, PNG, WEBP, HEIC, PDF & TIFF formats

< 20MB

Supported media uploads across JPG, PNG, WEBP, HEIC, PDF & TIFF formats

Cryptographic Aadhaar signature decode & PAN format check-digit validation

Secure-Code

Cryptographic Aadhaar signature decode & PAN format check-digit validation

Key-authenticated endpoints with asynchronous processing & structured JSON

REST API

Key-authenticated endpoints with asynchronous processing & structured JSON

High-Level Architecture

DeepFraudGuard platform data flow

Submissions undergo automatic intake, classification, and routing to specialized image or document engines, while identity documents are redirected to Fraud ID Guard.

  1. 01

    Intake & Validation

    File-type & size checks, secure storage, unique job ID assigned.

  2. 02

    Content Routing

    Automatic routing to Image Engine, Document Engine, or Fraud ID Guard.

  3. 03

    Asynchronous Analysis

    Multi-signal forensic panel, recompression ELA & VLM reasoning pass.

  4. 04

    Weighted Fusion

    Signals combined into a calibrated score, verdict & confidence level.

  5. 05

    Evidence & Reporting

    Attention maps, tamper overlays & exportable forensic PDF reports.

Every decision is evidence-backed with plain-language rationale and human-in-the-loop continuous learning feedback.

Forensic Services & Engines

Three specialized detection pathways

Each upload is routed to the exact engine trained on its specific media class.

Engine A · Deepfake Image Detection

Vision-SLM · DCT Frequency Analysis · Noise-Residual & Compression Forensics

  • Identifies AI-generated and digitally manipulated photographs via multi-signal ensemble
  • Detects generation artifacts, face swaps, local manipulation, and recompression traces
  • Computes visual attention maps, frequency spectrum rolloff, and noise residual views
  • Returns calibrated REAL or FAKE verdicts with stability-tested uncertainty metrics

Engine B · Document Forgery Detection

Document-Agnostic · Structural PDF Forensics · Recompression ELA · Content Rules

  • Detects tampering, splices, content inconsistencies, and synthetic post-creation edits
  • Content & Consistency: Validates arithmetic logic, date sequences, layout alignment & erasures
  • Recompression Error-Level Analysis (ELA): Amplifies differences on scanned/photographed pages
  • Structural PDF Forensics: Examines incremental edits, authoring tool traces & embedded overlays

Fraud ID Guard · Identity Verification

Secure-Code Decode · Check-Digit Validation · Front-to-Back Reconciliation

  • Purpose-built for Indian identity documents (Aadhaar PVC/e-Aadhaar/mAadhaar & PAN)
  • Secure-Code & Cryptographic Verification: Decodes QR codes & verifies signatures against issuing authority keys
  • Cross-Side Reconciliation: Reconciles front-printed fields against back-side secure-code demographics
  • Returns AUTHENTIC (≥73%), SUSPICIOUS (35-72%), or UNUSUAL (<35%) confidence verdicts

Developer REST API & Platform Security

Key Authentication · Asynchronous Worker Queue · Rate Limiting · JSON Outputs

  • Key-authenticated REST endpoints supporting background polling via unique job IDs
  • Transparent job lifecycle: PENDING → PROCESSING → COMPLETED (or FAILED)
  • Strict tenant isolation, zero raw frame retention, and encrypted transport
  • Exportable PDF audit reports and plain-language evidence narrative

Fraud ID Guard operates as a dedicated identity-forensic service with front-to-back cross-side reconciliation for Aadhaar and PAN documents.

Verdict & Confidence Bands

Calibrated verdict bands across engines

Standardized decision thresholds tailored for specific document and image categories.

Domain / EngineVerdict ThresholdsOperational Guidance
Deepfake Image EngineREAL / FAKE (Binary Probability)Supported by uncertainty metrics & known-media hash overrides
Document Forgery EngineAUTHENTIC (<35) · SUSPICIOUS (35-64) · FORGED (≥65)Corroborated by ELA recompression & structural PDF edit logs
Fraud ID Guard (Identity)AUTHENTIC (≥73%) · SUSPICIOUS (35-72%) · UNUSUAL (<35%)Hard rule engine overrides for secure-code signature failures & field mismatches

Evidentiary Standard

Why enterprise security trusts DeepFraudGuard

Four pillars of DeepFraudGuard's forensic evidence framework.

Multi-signal forensic panel
Combines frequency analysis, noise residuals, ELA recompression, and Vision-SLM semantic reasoning so no single forgery passes unnoticed.
Cryptographic secure-code verification
Decodes Aadhaar QR codes and verifies digital signatures against issuing authority keys across all historical key periods.
Document-agnostic forgery detection
Works seamlessly across invoices, bank statements, payslips, contracts, certificates, and official identity cards.
Continuous human-in-the-loop learning
Reviewed human feedback continuously recalibrates the system on confirmed difficult edge cases.

The rest of the line

More in Synthetic Media Guard

Inspect images, documents & identity credentials.

Launch the DeepFraudGuard console or request developer API documentation for system integration.

Request a demoTalk to sales