Tessnet2 a .NET 2.0 Open Source OCR assembly using Tesseract engine

http://www.pixel-technology.com/freeware/tessnet2/

Tessnet2 a .NET 2.0 Open Source OCR assembly using Tesseract engine

Keywords: Open source, OCR, Tesseract, .NET, DOTNET, C#, VB.NET, C++/CLI

Current version : 2.04.0, 02SEP09 (see version history)

The big picture
Tesseract is a C++ open source OCR engine. Tessnet2 is .NET assembly that expose very simple methods to do OCR.
Tessnet2 is multi threaded. It uses the engine the same way Tesseract.exe does. Tessdll uses another method (no thresholding).

License
Tessnet2 is under Apache 2 license (like tesseract), meaning you can use it like you want, included in commercial products. You can read full license info in source file.

Quick Tessnet2 usage

1. Download binary here, add a reference of the assembly Tessnet2.dll to your .NET project.

2. Download language data definition file here and put it in tessdata directory. Tessdata directory and your exe must be in the same directory.

3. Look at the Program.cs sample

Note: Tessnet2.dll needs Visual C++ 2008 Runtime. When deploying your application be sure to install C++ runtime (x86, x64)

Tessnet2 usage

Bitmap image = new Bitmap("eurotext.tif");
tessnet2.Tesseract ocr = new tessnet2.Tesseract();
ocr.SetVariable("tessedit_char_whitelist", "0123456789"); // If digit only
ocr.Init(@"c: emp", "fra", false); // To use correct tessdata
List<tessnet2.Word> result = ocr.DoOCR(image, Rectangle.Empty);
foreach (tessnet2.Word word in result)
    Console.WriteLine("{0} : {1}", word.Confidence, word.Text);

Tessnet2 source code and recompiling

1. Download Tesseract source code here and expand it in a directory

2. Download Tessnet2 source code here and expand it in Tesseract source code root directory (it should create dotnet sub directory)

3. Open the project solution tessnet2.sln. It's a Visual Studio 2008 C++/CLI project

Memory leak

Tesseract C++ source code is full of memory leak. Using tessnet2 assembly several time will cause memory overflow. This is not tessnet2 leak, this is tesseract leak and I spent two days in tesseract source code trying to improve this with no success. See what I think about this.

Tessnet2 demo
In the Tessnet2 source code you have two C# demo project. TesseractOCR is a multi-tread WinForm demo with a progression bar. TesseractConsole is a console demo.

The confidence score is between braquets. < 160 mean not bad

Version History

07JUN08: First release on Tesserect 2.03

10JUN08: Version 2.03.1. Change Confidence behavior, now it's calculated from each word letter and not from the first letter. Type change from byte to double. 0 = perfect, 100 = reject

13JUN08 : Version 2.03.2

After 3 days in Tesseract code (urgh), here is Tessnet2 version 2.03.2
The corrections deals with the following problems
* Confidence was not very useful, the value was strange. This has been corrected, setting the variable tessedit_write_ratings=true. After many test I found this mode is the best for confidence accuracy. Value range from 0 (perfect) to 255 (reject) . When value goes over 160 this really mean the OCR was bad.
* Calling DoOCR twice was not giving the same result. It was, as expected, a problem with global variables. The problem is almost fixed, sometime it doesn’t work but right now I can’t find what is not correctly reinitialized.