Extract text from an image in C# · oneocr cookbook

Load the bundled OCR engine once, load the image, and call ToText. The program below reads a sample invoice, a 300 DPI PNG you can download to try it.

Terminal

dotnet new console -o ImageToText && cd ImageToText
dotnet add package RecursiveAI.OneOcr --version 1.0.0
curl -O https://recursiveai.org/oneocr/samples/invoice.png

Put this in Program.cs and run dotnet run:

using OneImage;
using OneOcr;
 
// Load once at startup; the engine is immutable and safe to share across threads.
OcrEngine engine = OcrEngine.LoadBundled();
 
Image image = Image.Load("invoice.png");   // PNG, JPEG, WebP, TIFF, BMP, GIF
Console.WriteLine(image.ToText(engine));

Text

Northwind Supply Ltd
14 Harbour Street
BS1 4DJ Bristol
United Kingdom
VAT ID: GB284193752
INVOICE
Bill to
Blue Heron Bakery
88 Elm Road
OX1 2JD Oxford
United Kingdom
Invoice no.	INV-2041
Date	14/10/2026
Due date	13/11/2026
Order no.	PO-7731
Description	Qty	Unit price	VAT	Amount
Flour, 25 kg bag	4 pcs	31.50	0%	126.00
Cane sugar, 10 kg	2 pcs	18.25	0%	36.50
Baking parchment roll	6 pcs	7.40	20%	44.40
Subtotal
VAT total
Total
£206.90
£8.88
£215.78
Payment terms: Payable within 30 days
BAN: GB33 BUKB 2020 1555 5555 55	BIC: BUKBGB22	Tel. +44 117 496 0321

That is the whole page, in reading order: lines end with a newline and the cells of a table are separated by a tab. One word is wrong: the footer’s “IBAN:” came back as “BAN:”, at 97% confidence, so check values that matter. The models ship inside the NuGet package, so there is nothing else to install, no tessdata folder, and nothing is sent over the network.

Words and their bounding boxes

Ocr returns the page model instead of a string: blocks, lines and words, each with a box in pixels from the top left and a confidence:

using OneImage;
using OneOcr;
 
OcrEngine engine = OcrEngine.LoadBundled();
OcrPage page = Image.Load("invoice.png").Ocr(engine);
 
Console.WriteLine($"{page.Width} x {page.Height} px, {page.Words.Count} words");
foreach (OcrWord word in page.Words.Take(5))
{
    OcrBox b = word.Box;
    Console.WriteLine($"{word.Text,-10} {word.Confidence:P0}  " +
        $"({b.Left:F0}, {b.Top:F0}) to ({b.Right:F0}, {b.Bottom:F0})");
}

Text

2481 x 3509 px, 98 words
Northwind  100%  (206, 205) to (457, 245)
Supply     100%  (485, 205) to (648, 249)
Ltd        92%  (673, 207) to (740, 245)
14         100%  (205, 280) to (242, 308)
Harbour    99%  (260, 278) to (403, 309)

Confidence is the calibrated probability that the word is exactly right; Get OCR confidence scores shows how to use it. Every word, line and block also has a Source saying where it came from.

Limits

Printed text in Latin, Greek and Cyrillic scripts. That includes the EU languages, Turkish and Vietnamese, all in one model with no language to choose. Chinese, Japanese, Korean, Arabic and Indic scripts, handwriting and text in street photos aren’t read yet.
One page per call. Ocr and ToText throw ArgumentException for an image with several frames; OcrPages reads every page of a multi-page TIFF.
.NET 10. The packages target net10.0.

For scanned PDFs, see OCR a PDF; for tables, Extract a table from an image. The full walk-through is in Getting started.