I am doing character recognition in ID cards in opencv-2.x using C++. I tried Tesseract OCR, but it didn't give me better results than my own neural network training and all. Although, I am still struggling with segmentation of characters. I was wondering if it is possible to get the character or word position from Tesseract OCR Api in C++ and use my neural network for prediction. Any suggestions?
Simply, I need bounding box of each character in ID using Tesseract.
You can retrieve the position of each word using the run function:
virtual void cv::text::OCRTesseract::run(Mat& image,
std::string& output_text,
std::vector<Rect>* component_rects = NULL,
std::vector<std::string>* component_texts = NULL,
std::vector<float>* component_confidences = NULL,
int component_level = 0);
where component_rects will provide a list of Rects for the individual text elements found, and component_level = OCR_LEVEL_WORD will find single words.
opencv-3.x, I am using opencv-2.9.x. And I am quite sure there is no OCRTesseract in my verson of opencv.
ResultIterator. Sorry but I can't write a working example right now.Use HOCR format for your results instead of plain text. I think the argument is -hocr or just hocr. Embedded in the HOCR results (which are returned as XHTML, but it's easy to convert to JSON) are the page coordinates of every word.
Shameless plug, you can also try my online OCR service http://OCRestful.com, which offers OCR as a service via a RESTful API. Just POST your doc and get back JSON with word coordinates and word-by-word confidence scores. There's a permanently-free tier as well as paid plans.
You mentioned your data is sensitive--you can force the document to auto-delete after it's OCR-processed, and you control the lifetime of the extracted OCR data.
Matt
