Lyo.Pdf.Ocr 2.0.0

Lyo.Pdf.Ocr

PNG-renders a PDF page with Lyo.Pdf.Rendering (PDFium), runs an IOcrEngine from Lyo.Images.Ocr, then lifts OCR pixel boxes into PDF points.

Targets net10.0.

Examples

Add Pdf OCR in DI

services.AddTesseractOcrEngineFromConfiguration(configuration); // or any IOcrEngine
services.AddPdfOcr();

Read one page

public sealed class Worker(PdfOcrService pdfOcr, IPdfService pdfService)
{
    public async Task<PdfOcrDocumentPage> RunAsync(byte[] pdfBytes, CancellationToken ct)
    {
        await using var reader = await pdfService.OpenFromBytesAsync(pdfBytes, ct);
        var result = await pdfOcr.ReadPageAsync(reader, pageNumber1Based: 1, dpi: 200, cancellationToken: ct);
        return result.ValueOrThrow();
    }
}

ReadPageAsync pipeline

PdfOcrService has one method:

Task<Result<PdfOcrDocumentPage>> ReadPageAsync(
    IPdfReader pdfReader,
    int pageNumber1Based,
    int dpi,
    OcrReadRequest? ocrRequest = null,
    string? pdfPassword = null,
    CancellationToken cancellationToken = default);

Steps:

1. pdfReader.GetPageSizePoints(pageNumber1Based) for the PDF page dimensions.
2. IPdfPageRasterizer.RenderPageToPngAsync(pdfReader.SourceBytes, …) for the pixel raster (and bitmap width/height).
3. IOcrEngine.ReadAsync(pngStream, ocrRequest, …) for the per-word text and pixel-space bounding boxes (Y-up).
4. OcrCoordinateTransforms.MapPixelBoxToPdfPoints(box, pageWidthPts, pageHeightPts, widthPx, heightPx) to lift each OcrWord into a PdfWord.

PdfOcrDocumentPage holds the original OcrPageResult, the projected IReadOnlyList<PdfWord> WordsInPdfPoints, and the source page size.

Either stage can fail as Result<PdfOcrDocumentPage>.Failure; unexpected exceptions are tagged with PdfOcrErrorCodes.ReadFailed ("PDF_OCR_READ_FAILED").

DI registration

AddPdfOcr adds PdfOcrService as a singleton and, if IPdfPageRasterizer is missing, calls AddPdfPageRasterizer. Register IOcrEngine yourself.

When OCR is worth it

Selectable-text PDFs should use IPdfReader.Text (PdfPig) instead. OCR is useful only when the PDF has no embedded text layer.

Dependencies

Generated from ProjectReference / PackageReference (same model as docs/Lyo.ProjectGraph.html).

• Lyo.Exceptions (direct, lyo)
• Lyo.Images.Ocr (direct, lyo)
• Lyo.Pdf.Models (direct, lyo)
• Lyo.Pdf.Rendering (direct, lyo)
• Microsoft.Extensions.DependencyInjection.Abstractions 10.0.5 (direct, microsoft)
• Microsoft.Extensions.Logging.Abstractions 10.0.5 (direct, microsoft)
• Lyo.Common.Core (transitive, lyo)
• Lyo.Common.Metadata (transitive, lyo)
• Lyo.Configuration (transitive, lyo)
• Lyo.DataTable.Models (transitive, lyo)
• Lyo.Metrics (transitive, lyo)
• Lyo.Result (transitive, lyo)
• Microsoft.Bcl.AsyncInterfaces 10.0.5 (transitive, microsoft, netstandard2.0)
• Microsoft.Extensions.Configuration.Binder 10.0.5 (transitive, microsoft)
• PDFtoImage 5.2.1 (transitive, third-party)
• SixLabors.ImageSharp 3.1.12 (transitive, third-party)
• System.Memory 4.6.3 (transitive, microsoft, netstandard2.0)
• System.Text.Json 10.0.5 (transitive, microsoft, netstandard2.0)