Background

This is an example of exposing an Optical Character Recognition (OCR) service as a RESTful API.
As multiple OCR engines and models are running, the consumer can choose which engine/model to carry out the recognition - or execute against all models and parse the output.
This isn't production-ready, but hopefully is a useful starter for anyone looking to build a similar capability.

Components

I've run this on Ubuntu 22.04. Works both on an IaaS VM or in a container
Flask provides the web service
EasyOCR and Paddle for OCR

Image Size

The container image is REALLY large. The dependencies of EasyOCR are heavy, so removing that element would result in a significantly smaller image.
I tried using a more lightweight base image (Alpine) although wasn't able to get PaddlePaddle to function with it.

Usage

Container

Build using the Dockerfile. Listens on the default Flask port of 5000. As mentioned, the image is pretty large, and will take a while to push and pull.

Ubuntu Server

Clone this repo, change directory into the base directory and:

matt@ocr:~$ apt-get update
matt@ocr:~$ apt install python3 python3-pip libgl1-mesa-glx curl -y
matt@ocr:~$ pip install paddlepaddle==2.4.0rc0 paddleocr==2.6.0.1 imutils==0.5.4 Pillow==9.5.0 opencv-python==4.5.5.64 numpy==1.23.5 Flask==2.3.2 matplotlib==3.6.0 easyocr==1.7.0
matt@ocr:~$ mkdir -p /app/uploads
matt@ocr:~$ chmod 777 /app/uploads
matt@ocr:~$ cp * /app/
matt@ocr:~$ python3 /app/ocr_api.py

Calling the Service

An example call to the service is included in the repo as sendimage.py.
The service returns a JSON response such as the one below. To only run the image through a single model, specify it in the 'model' HTTP header. Details are in sendimage.py. In my testing, PPOCR-v3 has been the winner in terms of accuracy.

[
	[{
		"text": "HELLO WORLD",
		"confidence": 0.9998155695810481,
		"model": "easyocr",
		"processing_time_ms": 304.94,
		"correlation_id": "274a024c-1571-49db-af82-2df5d23df3bd"
	}],
	[{
		"text": "HELLOWORLD",
		"confidence": 0.9997454881668091,
		"model": "ppocr",
		"processing_time_ms": 75.31,
		"correlation_id": "9a00de4a-0b22-4c25-924b-35b4115f53ff"
	}],
	[{
		"text": "HELLOWORLD",
		"confidence": 0.9898713231086731,
		"model": "ppocrv2",
		"processing_time_ms": 85.71,
		"correlation_id": "bea9806d-da5d-4a8e-b506-2857173236e7"
	}],
	[{
		"text": "HELLO WORLD",
		"confidence": 0.9897682666778564,
		"model": "ppocrv3",
		"processing_time_ms": 137.64,
		"correlation_id": "97db6f59-20e9-40d4-bfc8-f221590e7612"
	}]
]