This is based on a real use case in my work. The main idea is to locate the area in a picture taken by users and identify letters in this area.

The methods we used include two parts:

• MSER to get the candidate area from the picture taken by users.
• RCNN model to get the final letter sequences

In Nov. 2018, we use about 1500 pictures and get acceptable results. So open its source here and want to get more helps and suggestions