环境
PSENET 编译
cd psenet/pse
rm -rf pse.so
make
实现功能
2020.03.16更新
2020.03.12更新
- 升级crnn_lite_lstm_dw.pth模型crnn_lite_lstm_dw_v2.pth , 精度更高
竖排字体样式:
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/vertical_text_fonts/imgs/test.jpg?raw=true)
竖排生成的竖排文本样例:
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/vertical_text_fonts/imgs/00156360.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/vertical_text_fonts/imgs/00000027.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/vertical_text_fonts/imgs/00156365.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/vertical_text_fonts/imgs/00187940.jpg?raw=true)
web服务启动
cd chineseocr_lite## 进入chineseocr目录
python app.py 8080 ##8080端口号,可以设置任意端口
访问服务
http://127.0.0.1:8080/ocr
识别结果展示
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/test_imgs/5_res.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/test_imgs/4_res.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/test_imgs/1_res.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/test_imgs/2_res.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/test_imgs/3_res.jpg?raw=true)
ncnn检测识别展示(x86 cpu 单进程)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/ncnn_project/ocr/res_imgs/res_3.jpg?raw=true)
![](https://github.com/ouyanghuiyu/chineseocr_lite/raw/master/ncnn_project/ocr/res_imgs/res_2.jpg?raw=true)
参考
- ncnn https://github.com/Tencent/ncnn
- crnn https://github.com/meijieru/crnn.pytorch.git
- chineseocr https://github.com/chineseocr/chineseocr
- Psenet https://github.com/WenmuZhou/PSENet.pytorch
- 语言模型实现 https://github.com/lukhy/masr