Tesseract OCR — extract text from images and scans in 100+ languages with confidence scores, bounding boxes, and structured layout output.

Overview

Tesseract is the leading open-source OCR engine, originally developed by HP and now maintained by Google. It uses a neural network (LSTM) and supports 100+ languages including Latin, Arabic, Chinese, Japanese, and more. Beyond plain text, it can return structured results with per-word confidence scores and pixel-accurate bounding boxes.

Quick Start

uses tesseract;

{ One-shot — no handle needed }
var text := OCRFile('/tmp/scan.png');
Writeln(text);

{ With specific language }
var pt := OCRFileLang('/tmp/doc.jpg', 'por');

{ Only extract digits (e.g. meter readings, invoices) }
var nums := OCRDigits('/tmp/meter.png');

Full API — Confidence & Bounding Boxes

uses tesseract;

var api := NewAPI('', 'eng');  { '' = default tessdata path }
SetPageSeg(api, psmAuto);

SetImageFile(api, '/tmp/invoice.png');

var text := GetText(api);
var conf := GetMeanConfidence(api);
Writeln('Confidence: ' + FloatToStr(conf) + '%');

{ Iterate recognized words }
var words := GetWords(api);
for w in words do
begin
  if w.Confidence > 80 then
    Writeln(w.Text + ' at (' + IntToStr(w.Left) + ',' + IntToStr(w.Top) + ')');
end;

FreeAPI(api);

Multi-language & LSTM Mode

{ English + Spanish, neural net only }
var api := NewAPIEx('', 'eng+spa', oemLSTM);
SetImageFile(api, '/tmp/mixed.png');
var text := GetText(api);
FreeAPI(api);

Restrict to Region

var api := NewAPI('', 'eng');
SetImageFile(api, '/tmp/form.png');

{ Only read the "Total" field region (pixels) }
SetRectangle(api, 500, 800, 200, 40);
var total := GetText(api);

FreeAPI(api);

HOCR Output

{ HOCR = HTML with embedded bounding boxes — useful for PDF text layers }
var api := NewAPI('', 'eng');
SetImageFile(api, '/tmp/page.png');
var hocr := GetHOCR(api, 1);
{ Write hocr to a .hocr file for pdfbeads, hocr2pdf, etc. }
FreeAPI(api);

Page Segmentation Modes

API Reference

Lifecycle

Image Input

Recognition

Convenience