DeepCard

A bank card number recognition system based on deep-learning.

Usage

Docker

Pull Image

1. Pull and run from Docker Hub.

docker run -it --rm -p 8000:80 --name dc stardustdl/deepcard:cpu

2. Visit URL http://localhost:8000 to see DeepCard website.

Rebuild Image

1. Build DeepCard.Host (needs .NET Core SDK):

cd ./src/DeepCard.Host
dotnet publish -c Release -r linux-x64

2. Build docker image:

docker build -t deepcard:cpu -f ./docker/cpu/Dockerfile .

2. Start docker container (will use more than 3GB memory):

docker run -it --rm -p 8000:80 --name dc deepcard:cpu

4. Visit URL http://localhost:8000 to see DeepCard website.

Host

If you want to run directly on the host, use this method.

1. Install dependences:

chmod +x ./build.sh
./build.sh

If you need to build GUI from the source code, you need to install .NET Core SDK first. These commands are commentted in build.sh.

2. Run the server:

chmod +x ./run.sh
./run.sh

3. Visit URL http://localhost:8000 to see DeepCard website.

4. If you want to batch images, copy your images into ./src/DeepCard.API/test_images and use these commands:

cd ./src/DeepCard.API
python3 ./demo.py

The results will be at ./src/DeepCard.API/test_result/result.txt, each line contains a file name and its recognization result.

Development

Use different shells to run these command, because they need long-run.

1. Copy the trained models to the directories. Or copy src/DeepCard.API/server.py to the independent OCR directory, and run the file there.

2. Start the OCR API server.

cd ./src/DeepCard.API ; python server.py

3. Start the website host server

dotnet run -p ./src/DeepCard.Host

Then visit the URL http://localhost:5000. You can see the website.

Hints: If you want to debug host without real OCR, start the demo API server with command: dotnet run -p ./src/DeepCard.API.Demo --urls="http://localhost:4000"

Dependences

1. Python
2. OpenCV
3. Tensorflow
4. Keras
5. Pytorch
6. .NET Core 2.2

Reference

• chinese_ocr

Collaborators

• @courao
• @mitdalao
• @GreatTMZ