Skip to content

armgit / Filee

Public Imported from GitHub
0

Free, open-source file converter. Convert 263 formats offline by drag & drop — images, PDF, Word, Excel, PowerPoint, HWP/HWPX, video, audio, e-books and archives. No upload, no Office needed.

OCR: a scan with a stamp is still recognized

cre8ive-marketinge77c45706c5c
diff --git a/src/Filee.Engines/Ocr/TesseractConverter.cs b/src/Filee.Engines/Ocr/TesseractConverter.cs
index 242fee1d..a5c8f9f5 100644
--- a/src/Filee.Engines/Ocr/TesseractConverter.cs
+++ b/src/Filee.Engines/Ocr/TesseractConverter.cs
@@ -17,6 +17,7 @@ using PdfSharp.Pdf;
using PdfSharp.Pdf.IO;
using PDFtoImage;
using SkiaSharp;
+using UglyToad.PdfPig.Core;
using UglyToad.PdfPig.Exceptions;
// PDFtoImage marks its API for the platforms PDFium ships for; Filee only targets desktop platforms, all covered.
@@ -201,13 +202,17 @@ public sealed class TesseractConverter : IConverter
}
}
- /// <summary>For each page: true when it has any text, visible or not (an OCR layer counts).</summary>
+ /// <summary>
+ /// For each page: true when it has text of its own. An OCR layer (invisible text) always counts. A scan with a
+ /// stamp does not: a picture over most of the page whose only text is a little visible text (a page number or a
+ /// Bates number the scanner software added, under <see cref="StampCoverage"/> of the page).
+ /// </summary>
internal static List<bool> PagesWithText(string path)
{
try
{
using var pdf = UglyToad.PdfPig.PdfDocument.Open(path, new UglyToad.PdfPig.ParsingOptions { UseLenientParsing = true, SkipMissingFonts = true });
- return [.. pdf.GetPages().Select(page => page.Letters.Any(l => !string.IsNullOrWhiteSpace(l.Value)))];
+ return [.. pdf.GetPages().Select(HasText)];
}
catch (PdfDocumentEncryptedException)
{
@@ -215,6 +220,23 @@ public sealed class TesseractConverter : IConverter
}
}
+ /// <summary>The share of a page that visible text may cover on a scan and still be only a stamp.</summary>
+ internal const double StampCoverage = 0.01;
+
+ private static bool HasText(UglyToad.PdfPig.Content.Page page)
+ {
+ var letters = page.Letters.Where(l => !string.IsNullOrWhiteSpace(l.Value)).ToList();
+ if (letters.Count == 0)
+ return false;
+ if (letters.Any(l => l.RenderingMode is TextRenderingMode.Neither or TextRenderingMode.NeitherClip))
+ return true;
+ var area = page.Width * page.Height;
+ if (area <= 0 || letters.Sum(l => l.GlyphRectangle.Area) >= StampCoverage * area)
+ return true;
+ // A page drawn with little text and no picture under it (a cover, a short note) is not a scan.
+ return !page.GetImages().Any(image => image.Bounds.Area >= area / 2);
+ }
+
/// <summary>
/// True when the picture says how large it is meant to be. Without that Tesseract guesses 70 dpi (a page of a
/// metre and more); 300 dpi is what scanners write.
diff --git a/tests/Filee.Engines.Tests/OcrTests.cs b/tests/Filee.Engines.Tests/OcrTests.cs
index 66aeadb5..b8582e33 100644
--- a/tests/Filee.Engines.Tests/OcrTests.cs
+++ b/tests/Filee.Engines.Tests/OcrTests.cs
@@ -3,6 +3,7 @@
// language data they read the first time (English, German).
using System.IO.Compression;
+using System.Text;
using Filee.Core.Conversion;
using Filee.Core.Presets;
using Filee.Engines.Infrastructure;
@@ -43,6 +44,62 @@ public class OcrLanguageListTests
}
}
+/// <summary>Which pages count as having text, so OCR leaves them alone (no Tesseract needed).</summary>
+public class OcrPagesWithTextTests(EngineFixture fx) : IClassFixture<EngineFixture>
+{
+ /// <summary>
+ /// A one-page A4 PDF written by hand: optionally a picture over the whole page (a scan), then lines of Helvetica
+ /// text, visible or invisible (render mode 3, as an OCR layer is written).
+ /// </summary>
+ private string Pdf(bool scan, int lines, string text = "ABC-000123", bool invisible = false)
+ {
+ var content = new StringBuilder();
+ if (scan)
+ content.Append("q 595 0 0 842 0 0 cm /Im1 Do Q\n");
+ for (var i = 0; i < lines; i++)
+ content.Append($"BT /F1 10 Tf {(invisible ? "3" : "0")} Tr 40 {800 - i * 12} Td ({text}) Tj ET\n");
+ string[] objects =
+ [
+ "<< /Type /Catalog /Pages 2 0 R >>",
+ "<< /Type /Pages /Kids [3 0 R] /Count 1 >>",
+ "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Contents 4 0 R /Resources << /Font << /F1 5 0 R >> /XObject << /Im1 6 0 R >> >> >>",
+ $"<< /Length {content.Length} >>\nstream\n{content}endstream",
+ "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>",
+ "<< /Type /XObject /Subtype /Image /Width 1 /Height 1 /ColorSpace /DeviceGray /BitsPerComponent 8 /Length 1 >>\nstream\n\u00c8\nendstream",
+ ];
+ var pdf = new StringBuilder("%PDF-1.4\n");
+ var offsets = new List<int>();
+ for (var i = 0; i < objects.Length; i++)
+ {
+ offsets.Add(pdf.Length);
+ pdf.Append($"{i + 1} 0 obj\n{objects[i]}\nendobj\n");
+ }
+ var xref = pdf.Length;
+ pdf.Append($"xref\n0 {objects.Length + 1}\n0000000000 65535 f \n");
+ foreach (var offset in offsets)
+ pdf.Append($"{offset:D10} 00000 n \n");
+ pdf.Append($"trailer\n<< /Size {objects.Length + 1} /Root 1 0 R >>\nstartxref\n{xref}\n%%EOF\n");
+ var path = Path.Combine(fx.NewFolder(), "page.pdf");
+ File.WriteAllBytes(path, Encoding.Latin1.GetBytes(pdf.ToString()));
+ return path;
+ }
+
+ [Fact]
+ public void A_scan_with_only_a_stamp_is_recognized()
+ {
+ Assert.Equal([false], TesseractConverter.PagesWithText(Pdf(scan: true, lines: 1)));
+ Assert.Equal([false], TesseractConverter.PagesWithText(Pdf(scan: true, lines: 0)));
+ }
+
+ [Fact]
+ public void An_ocr_layer_a_page_of_text_and_a_drawn_note_are_left_alone()
+ {
+ Assert.Equal([true], TesseractConverter.PagesWithText(Pdf(scan: true, lines: 1, invisible: true)));
+ Assert.Equal([true], TesseractConverter.PagesWithText(Pdf(scan: true, lines: 40, text: "The quick brown fox jumps over the lazy dog again and again")));
+ Assert.Equal([true], TesseractConverter.PagesWithText(Pdf(scan: false, lines: 1)));
+ }
+}
+
public class OcrTests(EngineFixture fx) : IClassFixture<EngineFixture>
{
private static readonly bool Installed = TesseractConverter.Locate() is not null;