aleksandrov-dima / tabula-sharp

Extract tables from PDF files (port of tabula-java)

tabula-sharp

tabula-sharp is a library for extracting tables from PDF files — it is a port of tabula-java

• Supports .NET 5, .NET Core 3.1, .NET Standard 2.0, .NET Framework 4.5, 4.51, 4.52, 4.6, 4.61, 4.62, 4.7
• No java bindings

NuGet packages available on the releases page and on www.nuget.org:

• Tabula
• Tabula.Json
• Tabula.Csv

Differences with tabula-java

• Uses PdfPig, and not PdfBox.
• Coordinate system starts from the bottom left point (going up) of the page, and not from the top left point (going down).
• The NurminenDetectionAlgorithm is replaced by SimpleNurminenDetectionAlgorithm, because it requieres an image management library.
• Table results might be different because of the way PdfPig builds Letters bounding box.

Usage

Stream mode - BasicExtractionAlgorithm

using (PdfDocument document = PdfDocument.Open("doc.pdf", new ParsingOptions() { ClipPaths = true }))
{
	ObjectExtractor oe = new ObjectExtractor(document);
	PageArea page = oe.Extract(1);
	
	// detect canditate table zones
	SimpleNurminenDetectionAlgorithm detector = new SimpleNurminenDetectionAlgorithm();
	var regions = detector.Detect(page);
	
	IExtractionAlgorithm ea = new BasicExtractionAlgorithm();
	List<Table> tables = ea.Extract(page.GetArea(regions[0].BoundingBox)); // take first candidate area
	var table = tables[0];
	var rows = table.Rows;
}

Lattice mode - SpreadsheetExtractionAlgorithm

using (PdfDocument document = PdfDocument.Open("doc.pdf", new ParsingOptions() { ClipPaths = true }))
{
	ObjectExtractor oe = new ObjectExtractor(document);
	PageArea page = oe.Extract(1);

	IExtractionAlgorithm ea = new SpreadsheetExtractionAlgorithm();
	List<Table> tables = ea.Extract(page);
	var table = tables[0];
	var rows = table.Rows;
}

Results

Stream mode - BasicExtractionAlgorithm

Lattice mode - SpreadsheetExtractionAlgorithm

HELP WANTED

• The original java implementation uses STR trees in RectangleSpatialIndex. This is not the case here so it might be a bit slower. Any help implementing a similar approach is welcome.