Dear friends, the open source project we share with you today is an ultra-lightweight Chinese OCR.

With the development of artificial intelligence, image recognition is more and more commonly used, life application scenarios such as: invoice, train ticket, verification code recognition, search App used to identify the test on the book and so on more application scenarios, so, today’s recommended open source project is: OCR related.

This project is the OCR project of Chineseocr_lite, which is an ultra-lightweight Chinese OCR to detect and recognize Chinese characters in natural scenes. This project supports vertical text recognition and NCNN reasoning. The total model of PsenET (8.5m) + CRNN (6.3m) + Anglenet (1.5m) is only 17M.

This project is based on ChineseOCR and Psenet to realize Chinese natural scene text detection and recognition.

Realize the function

  1. Provides lightweight Backone detection models PSENET (8.5m), CRNN_LSTM_lite (9.5m) and Line text direction classification network (1.5m)
  2. Arbitrary direction text detection, identify the line text direction
  3. CRNN \ CRNN_lite LSTM \ Dense
  4. Supports vertical text recognition
  5. NCNN implementation (support LSTM) NIhui big man implementation of CRNN_LSTM reasoning specific operation details: detailed record ultra-light Chinese OCR LSTM model NCNN implementation
  6. Provide vertical text samples and font libraries (font rotated 90 degrees)

Effect of the project

Vertical font styles:

Example of vertical text generated by vertical text:

Web service startup 

cd chineseocr_lite## Go to chineseocr
python app.py 8080 ##8080 Port number. Any port can be set
Copy the code

Access the service

http:/ / 127.0.0.1:8080 / OCR
Copy the code

Identification result Display

NCNN Detection and Identification Display (x86 CPU single process)

Project links:Github.com/ouyanghuiyu…

Do you like today’s recommendation? If you like the words, please leave a message and like at the bottom of the article, to show my support, your message, like and forward attention is my motivation to keep updating oh!