Exis.PdfOcr 1.0.6

Exis.PdfOcr

Local, no-cloud OCR for PDFs — turns scanned (image) PDF pages into searchable PDFs by adding an invisible, selectable text layer over the page image. Built on Exis.PdfEditor, so the output round-trips through the same find/replace, redaction, and extraction APIs as any digital PDF.

Designed for workflows like HIPAA medical-document batch processing: 100% on-device, no network calls, and per-page OCR confidence is surfaced so you can flag low-confidence pages for human review.

Packages

Install for Windows (Tesseract OCR):

dotnet add package Exis.PdfOcr
dotnet add package Exis.PdfOcr.Tesseract
dotnet add package Exis.PdfOcr.Windows

Install for macOS / iOS / Mac Catalyst (on-device Vision OCR, nothing extra to ship):

dotnet add package Exis.PdfOcr
dotnet add package Exis.PdfOcr.Apple

The core targets the same netstandard2.0;net8.0 surface as Exis.PdfEditor, so the same WPF (.NET), MAUI, and .NET Framework projects can consume it. On Apple platforms the OS-provided Vision text recognizer is used in place of Tesseract, so no tessdata is deployed and there is nothing to code-sign or notarize.

Usage

using Exis.PdfEditor;   // licensing
using Exis.PdfOcr;

// Exis.PdfOcr builds on the licensed Exis.PdfEditor — initialise it once at startup.
ExisLicense.Initialize();              // or ExisLicense.Initialize("XXXX-XXXX-XXXX-XXXX");

var result = await PdfOcr.MakeSearchableAsync(
    "input.pdf", "output.pdf", new OcrOptions { Languages = ["eng"] });

Console.WriteLine($"OCR'd {result.PagesProcessed}, skipped {result.PagesSkipped}, " +
                  $"avg confidence {result.AverageConfidence:P0}");

// Flag low-confidence pages for human review — confidence is never swallowed.
foreach (var page in result.Pages)
    if (page.WasOcrd && page.Confidence < 0.80)
        Console.WriteLine($"  Page {page.PageNumber} needs review (conf {page.Confidence:P0})");

• Digital and already-OCR'd pages pass through untouched — only true scans are rasterized and recognized, which protects quality and is fast on mixed documents.
• The output's text layer is invisible (PDF render mode 3) and positioned over the page image, so the page looks identical but text is now selectable and findable.

Progress reporting

var progress = new Progress<OcrProgress>(p =>
    Console.WriteLine($"[{p.Phase}] page {p.PageNumber}/{p.TotalPages}"));

await PdfOcr.MakeSearchableAsync("in.pdf", "out.pdf", new OcrOptions(), progress);

Languages

OcrOptions.Languages are Tesseract language codes ("eng", "spa", "vie", …). The Exis.PdfOcr.Tesseract package ships English (eng.traineddata). To add languages, drop the extra xxx.traineddata files into the package's tessdata folder (next to eng.traineddata in your app's output) and list them:

new OcrOptions { Languages = ["eng", "spa", "vie"] }

The invisible text layer embeds a Unicode font (DejaVu Sans), so accented and non-Latin names (e.g. Nguyễn, Peña) survive find/replace intact rather than being mangled.

MAUI / packaged apps

Where trained data isn't on disk as a plain file, ship it as a MauiAsset, extract it to a readable folder on first run, and point the provider at it:

var provider = new TesseractOcrProvider(tessdataPath: extractedTessdataDir);
await PdfOcr.MakeSearchableAsync("in.pdf", "out.pdf", new OcrOptions(), provider: provider);

Custom rasterizer / OCR engine / image redactor

All three seams are pluggable. Implement IPdfRasterizer, IOcrProvider, or IImageRedactor and pass it to MakeSearchableAsync / RedactScannedAsync. If you omit them on Windows, the core discovers the Windows rasterizer + redactor and Tesseract provider by reflection — so the core itself never references Windows.* or any native OCR binaries.

On macOS / iOS / Mac Catalyst the reflection-based defaults don't apply; pass the Apple implementations explicitly:

using Exis.PdfOcr.Apple;

await PdfOcr.MakeSearchableAsync("in.pdf", "out.pdf", new OcrOptions(), progress,
    provider:   new VisionOcrProvider(),
    rasterizer: new ApplePdfRasterizer());

await PdfRedactor.RedactScannedAsync("in.pdf", "safe.pdf",
    new RedactionOptions { Terms = new[] { "John Smith" } },
    progress,
    redactor: new AppleImageRedactor());

Note on scanned-page de-identification

Replacing text in the OCR layer changes the selectable text, not the underlying scanned pixels: the visible image still shows the original characters. To remove a name from the visible scan as well, follow OCR with a pixel-redaction pass over the matched word boxes (Exis.PdfEditor's PdfRedactor). The OCR layer is what makes those words findable.