I am using tesseract for OCR, mainly on invoices. However, tesseract requires to specify the language before it starts processing a file.
I thought I am going to perform ocr based on a predefined default language. Then I'd like use the resulting text to check which language is used. If it is not the default language, I process it again in order to get a better result from tesseract.
But how can I implement a language detection algorithm? Is there a C++ library I could use?
I am not sure if this would help as the library is in Java. But I found it really cool as it is able to detect around 50 languages from the given text and with a pretty good precision level. You may like to have a look at it and as it is open source, you may rewrite the code in C++ and give it back to the open source community if your application requires to be written only in C++.
Here is the link for the same:
http://code.google.com/p/language-detection/
Note: It uses the Apache Nutch and Tika libraries for analysis.
This paper "Natural Language Identification for OCR Applications" describes techniques involved in identification tasks similar to your requirements.
You might want to read my paper The WiLI benchmark dataset for written language identification and try lidtk.
TL;DR: Give CLD-2 a try.