WindowsOCREngine

A small command-line and library tool that performs OCR on PDF pages and image files using Windows built-in OCR (Windows.Media.Ocr). The application renders PDF pages, preprocesses images (padding, inversion, scaling), and extracts text using the Windows OCR engine. The project is implemented in C# targeting .NET 8 and is designed to run on Windows (desktop) environments that support WinRT OCR APIs.

Features

• OCR for PDF documents (page-by-page rendering)
• OCR for common image formats: PNG, JPEG, BMP, TIFF, GIF
• Preprocessing tuned for better accuracy: padding, color inversion, uniform scaling
• Reusable OCR engine wrapper with language selection (example: ru-RU)
• Saves extracted text to a .txt file next to the input file

Prerequisites

• Windows 10 (version 1809+) or Windows 11 required for WinRT OCR APIs
• .NET 8 SDK
• Visual Studio 2022 (recommended) or newer with .NET desktop development workload

Note: The project uses WinRT APIs (Windows.Media.Ocr, Windows.Data.Pdf, Windows.Graphics.Imaging) and therefore must run on Windows. Running on non-Windows platforms is not supported.

Build

1. Clone the repository:

   git clone https://github.com/your/repo.git
   cd WindowsImagePdfOcr

2. Open the solution in Visual Studio 2022.

3. Restore NuGet packages and build the solution (Build > Build Solution).

Alternatively, build from the command line with the .NET SDK:

dotnet build -c Release

Run

Usage (console):

WindowsImagePdfOcrApp.exe [-lang ] [-pagelabel ] [-dpi ] "C:\path\to\file.pdf"

If no path is provided, the app uses a test path configured in Program.cs. All flags accept the forms -flag <value>, --flag <value>, -flag=<value> and --flag=<value>, in any position.

• -lang <tag> — recognition-language override. Without it, the app auto-selects the first installed recognizer in priority order kk → ru → en. Matched by primary subtag, so -lang ru, -lang RU and -lang ru-RU are equivalent. If the requested language is not installed, a warning is printed and the app falls back to auto-selection.
• -pagelabel <text> — word used in the PDF page separator --- <text> N ---. Default depends on the OCR language: Страница (ru), Бет (kk), Page (otherwise).
• -dpi <n> — PDF render resolution (default 150, clamped to 72–600). 150 is the sweet spot for typical A4 document scans (clean native render, best body-text recognition). Very high DPI enlarges glyphs and can actually hurt prose recognition, so raise it only for unusually small fonts. Affects PDF input only.

Examples:

• OCR a PDF:

  WindowsImagePdfOcrApp.exe "C:\Scans\document.pdf"

• OCR an image:

  WindowsImagePdfOcrApp.exe "C:\Images\scan.png"

• Force Kazakh recognition:

  WindowsImagePdfOcrApp.exe -lang kk-KZ "C:\Scans\document.pdf"

• Russian scan at 400 DPI with a custom page label:

  WindowsImagePdfOcrApp.exe -lang ru-RU -dpi 400 -pagelabel "Лист" "C:\Scans\document.pdf"

The extracted text will be saved to file.pdf.txt or scan.png.txt next to the source file.

Configuration

• Language: By default the app auto-selects the OCR recognizer in priority order kk → ru → en (suited to trilingual Kazakh/Russian/Latin content). Override it from the command line with -lang <tag> (e.g. -lang kk-KZ). If the requested/auto language is not installed, the engine falls back to an available recognizer.
• Page separator: Multi-page PDFs are split with --- <label> N ---. The label defaults to the OCR language (Страница/Бет/Page) and can be overridden with -pagelabel <text>.
• Render DPI: PDF pages are rasterized at -dpi <n> (default 150) before OCR. 150 gives the best balance on A4 document scans; higher values enlarge text and can degrade prose recognition with the Windows OCR engine, so only raise it for very small fonts. Configurable per run; the rasterization code lives in PdfProcessor.cs.
• Scaling: The engine applies a scaling factor (default 2.0) before recognition. You can change scaling or pre-processing in PowerOcrEngine.cs.

Project layout

• Program.cs CLI entry point, detects file type and orchestrates processing
• PdfProcessor.cs Loads PDF, renders pages via Windows.Data.Pdf, converts pages to bitmaps
• ImageProcessor.cs Validates and loads images
• PowerOcrEngine.cs Image preprocessing (pad, invert, scale), converts to SoftwareBitmap, calls OcrEngine

Performance and optimization tips

• Parallelize PDF page processing when working with multi-page PDFs. Use a limited concurrency (for example Environment.ProcessorCount) to avoid saturating CPU and IO.
• Reduce render DPI (or scale) for large documents if OCR quality remains acceptable � lower resolution reduces memory usage and recognition time.
• Avoid unnecessary conversions: prefer SoftwareBitmap/WinRT types where possible to reduce copying between managed and native buffers.
• Reuse OCR engine instances where thread-safety permits. If OcrEngine is not thread-safe, implement a pool of engines to reduce initialization overhead.
• Add timing instrumentation (Stopwatch) around rendering, conversion, and OCR steps to identify bottlenecks.

Troubleshooting

• File not found: Ensure the full absolute path is provided and the app has read permissions.
• Unsupported image: Only common image formats are supported. Corrupted images may throw ArgumentException during load.
• Missing OCR languages: Install the required language packs for Windows if your language is not available to OcrEngine.