Seçim Tutanak OCR API

An endpoint was developed with Python FastAPI for scanning election minutes, detecting and reading QR, detecting table fields with vote numbers and reading values with OCR.

Running

To start API without DOCKER

• uvicorn secim_tutanak_ocr_api.main:app --reload --host 0.0.0.0 --port 8080

To start API with Dockerfile

1. Build
   • docker build -t secim-tutanak-ocr-api .
2. RUN
   • docker run -p 8080:8080 -it --rm --gpus all -v secim-tutanak-ocr-vol:/code/ -t secim-tutanak-ocr-api

To start API with Docker-Compose in Production

1. Build & Run * docker-compose -f docker-compose.dev.yml up --build

Using

OCR Endpoint POST:

http://127.0.0.1:8080/api/ocr/predimg"

Usage Example Python:

file_path = '17013_bulk_999999999_1684082326979111104.jpeg'

#read image as byte
with open(file_path, 'rb') as f:
    img_file = f.read()


#send image as byte with file tag
files = {'file': ('custom_test_img.jpg',img_file,'multipart/form-data')}


# [POST] Request to api
response_predict = requests.post(url, files=files)

Libraries Used

Document Scanner (PreProcessing)

Model files

• Model - MobileNetV3-Large backend

Reference: https://github.com/spmallick/learnopencv/tree/master/Document-Scanner-Custom-Semantic-Segmentation-using-PyTorch-DeepLabV3

QR Dedection and Decode

Reference: https://github.com/NaturalHistoryMuseum/pyzbar/

Table Dedection

Reference: Demo: https://huggingface.co/TahaDouaji/detr-doc-table-detection

OCR

Referance: https://github.com/PaddlePaddle/PaddleOCR