OfficeIMO.Reader.Ocr.Tesseract 3.3.0

OfficeIMO.Reader.Ocr.Tesseract - Tesseract OCR provider

OfficeIMO.Reader.Ocr.Tesseract is an optional IOfficeOcrEngine backed by an installed Tesseract command-line executable. It does not bundle native binaries or trained language data.

Install

dotnet add package OfficeIMO.Reader.Ocr.Tesseract
dotnet add package OfficeIMO.Reader.Pdf

Install Tesseract separately for the host operating system. TesseractOcrEngine.CreateDefault() discovers an explicit path, OFFICEIMO_TESSERACT_PATH, TESSERACT_PATH, the process PATH, and common platform locations:

TesseractOcrEngine engine = TesseractOcrEngine.CreateDefault();

You can verify the executable and required languages directly:

tesseract --version
tesseract --list-langs

Tesseract 5 is the current stable major line. Its command contract supports image input, language expressions, and TSV output; see the official Tesseract manual.

Recognize Reader candidates

using OfficeIMO.Reader;
using OfficeIMO.Reader.Ocr.Tesseract;
using OfficeIMO.Reader.Pdf;

OfficeDocumentReader reader = new OfficeDocumentReaderBuilder()
    .AddPdfHandler()
    .Build();
OfficeDocumentReadResult source = reader.ReadDocument("scanned.pdf");
var engine = TesseractOcrEngine.CreateDefault(new TesseractOcrEngineOptions {
    Language = "eng+pol",
    PageSegmentationMode = 3,
    Timeout = TimeSpan.FromMinutes(1)
});

OfficeDocumentOcrExecutionResult execution = await source.ApplyOcrAsync(
    engine,
    new OfficeDocumentOcrExecutionOptions {
        MaxCandidates = 25,
        MaxDegreeOfParallelism = 2,
        MaxTotalInputBytes = 64L * 1024L * 1024L
    });

foreach (OfficeDocumentOcrRecognition recognition in execution.Recognitions) {
    foreach (OfficeOcrTextSpan span in recognition.Result.Spans) {
        Console.WriteLine($"{span.Level}: {span.Text} ({span.Confidence:P0})");
    }
}

The provider parses Tesseract TSV into line and word spans with pixel bounding boxes and normalized confidence. Tesseract TSV does not expose character boxes, so SupportsCharacterSpans is false. A process or delegate engine can still return character spans through the shared core contract.

GetVersionAsync() and GetLanguagesAsync() provide explicit installation evidence. TesseractLanguageData.EnsureAsync("eng+pol") can provision any model in the facade's 28-language catalog, plus orientation data, into a versioned user cache. Downloads come from one immutable official tessdata_fast commit and must match package-pinned lengths and SHA-256 digests. Missing executables, unavailable trained data, unsupported input formats, and nonzero process exits surface as engine failures; ApplyOcrAsync converts them to structured diagnostics when ContinueOnError is enabled.

Per-request payload and output files use owner-only Unix directories and permissions. Temporary files are deleted by default; enable KeepTemporaryFiles only for controlled diagnostics.

Targets and licenses

• Targets: netstandard2.0, net8.0, net10.0 (net472 is also included on Windows builds).
• OfficeIMO provider license: MIT.
• Tesseract is an external dependency distributed under its own Apache 2.0 license.

Dependency footprint

• External: An installed Tesseract CLI. Language data is not bundled; callers may use their system data or explicitly invoke the checksum-pinned provisioner.
• OfficeIMO: OfficeIMO.Reader.Core and OfficeIMO.Reader.Ocr.Process own the OCR contract, bounded process execution, TSV projection, and diagnostics.

See the complete OfficeIMO package map for related formats and conversion paths.