tesseract-ocr widely disseminated, required top, but Chinese is not very good deal, if it is natural scenes, the basic Oh, and
Reference Links: https://github.com/tesseract-ocr
This is an AI Daniel individual test procedures of Beijing, Chinese good results, I also have to participate, you can try it with their own scenario on this basis, to expand,
Reference link: https: //github.com/senlinuc/caffe_ocr
Pictures of synthesis of links, we can progress in AI research team sent Tong concern I have read their paper, they are first advocated the use of composite picture data, and has made the team better effect, frozen chicken,
Reference link: https: //github.com/tongpi
If you have marked data can be shared, please leave a message in the comments section, or private letter I would like to initiate a shared annotation data to build a picture library.
In addition to image processing, the mainstream has OpenCV, PIL, leptonica, I go with the flow, and chose to OpenCV, but few people know that before there's OpenCV Leptonica, if you use a Linux system, you'll know her function powerful and at an alarming rate, Tesseract-OCR processing source images is to choose Leptonica.