PdfPecker
Docs 1.20.0: Text extraction
Download PDF Changelog

This is the manual for PdfPecker 1.20.0. Read it for the latest version, 1.25.0.

Text extraction

Changed in 1.20.0: Images, Document properties.

PdfText extracts the text of any PDF, not only those PdfPecker made, in reading order: as one string, or page by page with every line and word and where it is on the page. It reads what the PDF draws, using PdfPecker's own PDF parser and font engines, so it needs no other package either.

// All the text, pages separated by a form feed (\f) as pdftotext does.
string text = PdfText.Extract(pdfBytes);
string report = PdfText.ExtractFile("report.pdf");

// Some pages of an encrypted PDF.
string summary = PdfText.Extract(pdfBytes, new PdfTextOptions { Password = "pw", Pages = PdfPageRange.Parse("1-3") });

// Lines and words with their positions, such as for finding a total on an invoice.
foreach (PdfTextPage page in PdfText.ExtractPages(pdfBytes))
    foreach (PdfTextLine line in page.Lines)
        if (line.Text.StartsWith("Total"))
            Console.WriteLine($"Page {page.PageNumber}, {line.Y:F0} pt from the top: {line.Words[^1].Text}");

PdfTextPage first = PdfText.ExtractPage(pdfBytes, 1);

Extract, ExtractPages and ExtractPage also take a Stream, and ExtractPagesFromFile reads the pages of a file.

Pages, lines and words

PdfTextPage.Text is the page's text, with lines separated by a line feed and paragraphs by an empty line. Its Lines go from top to bottom, and each PdfTextLine has its Text, its Words in reading order and a bounding box (X, Y, Width, Height). Each PdfTextWord has its Text, its box, its FontSize and its FontName (such as Helvetica-Bold). Positions are in points from the top-left corner of the page as it is displayed, after its rotation, with y growing downward, like the signature box and the stamp offsets. Boxes span the font's usual ascent and descent, so they are close to, not exactly, the inked area. Text that runs up or down the page gets its own lines, with their Angle.

What is read

The characters come from the fonts' ToUnicode maps, encodings and glyph names, for the standard fonts and for embedded TrueType, CFF, Type 1 and Type 3 fonts, and /ActualText in marked content stands for the glyphs it covers. Where a PDF draws no space between words but leaves a gap, the gap becomes a space; kerning and letter spacing do not. Ligatures such as fi come out as their letters unless ExpandLigatures is false. Right-to-left text such as Arabic and Hebrew comes out in logical order, and PdfTextLine.RightToLeft says when most of a line is. Invisible text, such as the recognised text layer of a scanned page, is included, and text hidden by optional content (a layer that is off) is not. The values shown in form fields and the text of stamps are included unless IncludeAnnotations is false.

Pages are read in parallel (MaxDegreeOfParallelism, default one per processor), MaxPages (default 100,000) limits the work, and inputs are handled as untrusted, as with merging. A damaged file or page throws InvalidDataException, a wrong password PdfPasswordException.

Columns side by side are read across, line by line, not one column after the other, and table cells on one row make one line. Text drawn only as outlines or images has no characters to extract; a scanned page needs OCR first. See Limitations.

Images New in 1.20.0

PdfImages extracts the images the pages of any PDF draw: photos, logos, scans and the pictures inside forms and annotations, each once, in the order they are first drawn, with every page and place it is drawn.

foreach (PdfExtractedImage image in PdfImages.Extract(pdfBytes))
{
    image.Save($"image-{image.Placements[0].PageNumber}{image.FileExtension}");
    Console.WriteLine($"{image.Width}x{image.Height}, drawn {image.Placements.Count} times, camera: {image.Exif?.Model}");
}

A JPEG that a viewer shows as it is comes out as its original bytes, with no loss and with its EXIF data (KeepJpeg = false decodes it instead); every other image (Flate, JPEG 2000, CCITT, JBIG2 and the rest) is decoded with its mask as alpha and written as PNG. Each image has its Exif summary (camera, dates, orientation, GPS and the whole EXIF block) and the XMP metadata the PDF attaches to it. Placements are in points from the top-left corner, like the text's. Images in hidden layers, soft masks, tiling patterns and Type 3 glyphs are not returned. Pages, Password, IncludeAnnotations, MaxImages and MaxPages work as for text.

Document properties New in 1.20.0

PdfDocumentProperties.Read reads what a PDF says about itself, from its document information or, where that is missing or older, its XMP metadata (so PDF/A-4 and PDF 2.0 files without document information read too):

var p = PdfDocumentProperties.ReadFile("received.pdf");
Console.WriteLine($"{p.Title} by {p.Author}, created {p.CreationDate}, PDF {p.Version}, {p.PageCount} pages");
Console.WriteLine($"PDF/A: {p.PdfAConformance}, tagged: {p.IsTagged}, linearized: {p.IsLinearized}, encrypted: {p.IsEncrypted}");

It gives the title, author, subject, keywords, creator, producer, dates, custom properties and language, the PDF version, every page's displayed size and rotation, encryption and permissions, the PDF/A and PDF/UA conformance the metadata claims (a claim, not a check), whether the file is tagged or linearized, and the XMP packet. It also takes a Stream and a password.