Mobile App Development for Document Scanning

Mobile Application Development for Document Scanning

Document scanning app — not just camera wrapper. User expects result indistinguishable from tablet scanner: straight edges, readable text, correct orientation, acceptable PDF size. When this fails — app deleted after first passport scan with finger shadow.

Document Detection and Correction

Most important — find document boundaries in frame and remove perspective distortion. On iOS chain: AVCaptureSession → VNDetectRectanglesRequest → CIPerspectiveCorrection.

VNDetectRectanglesRequest tuned for task:

let request = VNDetectRectanglesRequest { request, _ in
    guard let results = request.results as? [VNRectangleObservation],
          let rect = results.first else { return }
    DispatchQueue.main.async {
        self.highlightDetectedDocument(rect)
    }
}
request.minimumAspectRatio = 0.5
request.maximumAspectRatio = 1.0
request.minimumSize = 0.3        // minimum 30% frame area
request.quadratureTolerance = 20 // tolerance for non-straight corners

After capture — CIPerspectiveCorrection with four corner points. Critical: give user manual corner adjustment when auto-detect misses. Without this mode app useless on crumpled or partially covered documents.

On Android — similar path via ML Kit Document Scanner API (appeared 2023) or custom implementation via OpenCV findContours → approxPolyDP. ML Kit Document Scanner easier to integrate but requires Google Play Services — not option for GMS-less devices.

Image Post-processing

Detection without enhancement gives mediocre result. After perspective correction apply:

• Adaptive thresholding — for black-and-white mode (passports, contracts). CIColorMonochrome + custom kernel or OpenCV adaptiveThreshold. Not CIPhotoEffectNoir — gives uneven results on documents with pale text.
• Noise reduction — CINoiseReduction with 0.02 noise level and 0.4 sharpness parameters for typical office documents.
• Background leveling — uneven lighting (hand shadow, window in frame) removed via top-hat transform in OpenCV.

Color mode — for documents with stamps and signatures. Important not to over-compress contrast — color stamps become muddy with aggressive post-processing.

PDF Assembly and OCR

Multi-page PDF assembled via PDFKit on iOS — straightforward:

let pdfDocument = PDFDocument()
for (index, image) in scannedPages.enumerated() {
    let pdfPage = PDFPage(image: image)!
    pdfDocument.insert(pdfPage, at: index)
}
pdfDocument.write(to: outputURL)

On Android — iText7 or PdfDocument from Android SDK.

OCR to document (searchable PDF) — separate feature. VNRecognizeTextRequest on iOS with revision3 gives acceptable quality for Russian and English. Results embedded as hidden text layer via PDFKit. For serious tasks (archives, legal documents) — Google Cloud Document AI or Tesseract with LSTM engine.

Storage and Synchronization

Documents take space. A4 PDF pages at 200 DPI — about 200–400 KB after JPEG compression at quality 85. Without compression — 2–5 MB. Storage strategy: locally in Application Support (not Documents — iCloud otherwise auto-captures all scans), sync via iCloud Drive or custom backend on user demand.

For Google Drive / Dropbox integration use official SDKs not REST directly — they handle token refresh and partial upload on connection break.

Implementation Process

Important to define immediately: need OCR, what export formats (PDF, JPEG, DOCX), require cloud storage, legal validity requirements for scans.

Development: camera with real-time detection → manual adjustment → post-processing → OCR (optional) → PDF assembly → export.

Timeline Estimates

Basic scanner with auto-detect and PDF — 3–4 weeks. With OCR, searchable PDF, cloud sync and multi-page mode — 6–9 weeks.