AI Document Verification via Mobile Camera

AI document verification via mobile camera is not just OCR—it's a full pipeline: document detection in the frame, quality assessment, field extraction, validation, and authenticity check. Each step can become a bottleneck, so we use a combination of methods proven on dozens of projects. A typical scenario: the user photographs a passport in poor lighting, the document is partially shadowed, edges cropped. Even if OCR reads the text, the MRZ may be damaged and data fails validation. Without guidance, the user makes 3–5 attempts, worsening UX, increasing drop-off by up to 30%, and raising support costs. We solve this comprehensively, reducing failed attempts to single digits and saving the budget on fraud control. In our practice of implementing verification systems for banks and fintech services, recognition accuracy reaches 99% with proper pipeline configuration. Our team has over 8 years of experience in mobile document verification solutions and has successfully launched more than 50 projects in this area. We guarantee >99% accuracy with proper setup—each project is accompanied by a detailed quality audit.

Why Raw OCR Doesn't Work

Attempting to read a passport via Vision framework (iOS) or ML Kit (Android) and parse the result with regex is the most common first approach. It yields 70–80% accuracy in the lab and 40–60% on real users: glare, shooting angle, crumpled pages, document wear—all of this breaks simple OCR.

The correct pipeline adds three layers:

• Image preprocessing. Perspective correction (document shot at an angle), contrast enhancement, glare removal. On iOS—CIFilter + CIPerspectiveCorrection. On Android—OpenCV via JNI or CameraX with custom ImageAnalysis.
• Specialized document OCR. Not general OCR, but models trained on documents: Microsoft Azure Document Intelligence, Google Document AI, Amazon Textract. They return structured fields—surname, given_names, date_of_birth, document_number.
• Machine Readable Zone (MRZ) parsing. Passports contain MRZ per ICAO 9303 standard. This is the most reliable source: standardized font, clear structure, checksum. Libraries: mrz-java, passport-reader for iOS, or custom implementation.

Why We Choose Azure Document Intelligence?

Azure Document Intelligence delivers >99% accuracy on standardized documents, supports MRZ out of the box, and provides confidence scores for each field. Compared to Google Document AI (~98% accuracy) and Amazon Textract (95%, no built-in MRZ), Azure offers the best price/quality ratio for mobile scenarios. For budget projects, Google Document AI works; for AWS integration, Amazon Textract; but for mobile high-load scenarios, Azure is more stable.

How We Integrate Azure Document Intelligence

// iOS — Swift
import AzureAIDocumentIntelligence

class DocumentVerificationService {
    private let client: DocumentIntelligenceClient

    func analyzePassport(imageData: Data) async throws -> PassportData {
        let request = AnalyzeDocumentRequest(
            urlSource: nil,
            base64Source: imageData.base64EncodedString()
        )

        let operation = try await client.beginAnalyzeDocument(
            "prebuilt-idDocument",
            analyzeRequest: request
        )

        let result = try await operation.waitForResult()
        guard let document = result.documents?.first else {
            throw DocumentError.noDocumentDetected
        }

        return PassportData(
            firstName: document.fields?["FirstName"]?.valueString,
            lastName: document.fields?["LastName"]?.valueString,
            documentNumber: document.fields?["DocumentNumber"]?.valueString,
            dateOfBirth: document.fields?["DateOfBirth"]?.valueDate,
            expiryDate: document.fields?["ExpirationDate"]?.valueDate,
            nationality: document.fields?["CountryRegion"]?.valueCountryRegion,
            mrz: document.fields?["MachineReadableZone"]?.valueString,
            confidence: document.confidence ?? 0
        )
    }
}

Confidence score is a critical parameter. If confidence < 0.8, we request a retake with user guidance (better lighting, hold flatter, don't cover edges).

Realtime Camera Guidance

The user should not make multiple attempts blindly. Real-time feedback during capture via Vision framework on iOS:

// Detects document bounds in realtime while camera is active
func detectDocumentInFrame(_ pixelBuffer: CVPixelBuffer) {
    let request = VNDetectRectanglesRequest { [weak self] request, error in
        guard let observation = request.results?.first as? VNRectangleObservation else {
            self?.cameraGuidance = .noDocumentFound  // "Point camera at document"
            return
        }
        let area = observation.boundingBox.width * observation.boundingBox.height
        if area < 0.4 {
            self?.cameraGuidance = .tooFar           // "Move closer"
        } else if area > 0.9 {
            self?.cameraGuidance = .tooClose         // "Move further away"
        } else {
            self?.cameraGuidance = .ready            // Automatic capture
        }
    }
    request.minimumAspectRatio = 0.5
    request.maximumAspectRatio = 1.0
    request.minimumConfidence = 0.7
    try? VNImageRequestHandler(cvPixelBuffer: pixelBuffer).perform([request])
}

Automatic capture when positioning is ideal removes the need to press a button—reducing the number of poor shots.

Forgery Detection: What We Check?

Basic anti-spoofing for mobile verification includes:

These methods ensure the document is real and belongs to the presenter.

How Is Personal Data Protection Ensured?

All data is processed in accordance with GDPR and local regulations. Document images are not stored after verification—only extracted fields. We use encryption at rest and in transit, and an isolated environment for model execution. For sensitive data, the solution can be deployed in the client's cloud (Azure / AWS).

Process of Work

1. Analyze which documents need to be supported
2. Choose OCR provider (Azure / Google / custom)
3. Implement realtime camera guidance
4. Integrate document analysis API
5. MRZ parsing and cross-validation of fields
6. Anti-spoofing and liveness check
7. Compliance check (GDPR for passport data storage)

What's Included in the Result

• Camera module with realtime guidance and automatic capture
• Integration with chosen OCR service (Azure, Google, or custom model)
• MRZ parsing and cross-validation of fields
• Anti-spoofing and liveness check (optional)
• API documentation and integration guide
• Access to source code and test examples
• Support during release on App Store / Google Play

Time Estimates

MVP with Azure Document Intelligence and basic guidance—2–3 weeks. Full system with liveness check, anti-spoofing, and support for multiple document types—4–6 weeks.

To find out how to implement AI verification in your app, contact us. Order a consultation to assess your project—we'll choose the optimal solution and timeline. Get a free assessment of your case and recommendations for choosing a provider. AI verification integration pays off by reducing manual document checking.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.