adriaanbd / kamtutecs-api

A Dockerized API for OCR and NLP using Tesseract, OpenCV, and spaCy.

Kamtutecs

Getting Started

Instructions

1. Setup Backend
2. Setup Frontend
3. Go to localhost:3000
4. Upload image with horizontal text in it
5. Draw a bounding box around the desired text to extract it
6. Submit
7. Open Developer Tools (F12)
8. Look at the Console to see the response

Backend

Docker

$ git clone https://github.com/adriaanbd/kamtutecs-api.git
$ cd kamtutecs-api
$ docker-compose build
$ docker-compose up

Without Docker

$ apt-get -y install tesseract-ocr tesseract-ocr-spa
$ apt-get -y install libtesseract-dev
$ apt-get -y install libleptonica-dev
$ git clone https://github.com/adriaanbd/kamtutecs-api.git
$ cd kamtutecs-api
$ pip install -r requirements.txt
$ uvicorn app.main:app

Frontend

$ git clone https://github.com/adriaanbd/kamtutecs.git
$ cd kamtutecs
$ npm install
$ npm start

kamtutecs-api's Issues

Extracting text from bottom left corner gives OpenCV error

Error with Stakeholders image extracting text from bottom left corner:

File "./app/main.py", line 99, in preprocess_img
kamtutecs |     grey_img = cv2.cvtColor(cropped_image, cv2.COLOR_BGR2GRAY)
kamtutecs | cv2.error: OpenCV(4.4.0) /tmp/pip-req-build-a98tlsvg/opencv/modules/imgproc/src/color.cpp:182: error: (-215:Assertion failed) !_src.empty() in function 'cvtColor'

Translate OCR Text Error

When a different tesseract configuration is applied that produces an invalid OCR Text or an invalid OCR Text is passed on to translate(ocr_text) it produces the following error:

web_1  |   File "./app/main.py", line 158, in textract
web_1  |     translation = translate(ocr_text)
web_1  |   File "./app/main.py", line 122, in translate
web_1  |     language = detect_lang(text)
web_1  |   File "./app/main.py", line 64, in detect_lang
web_1  |     possible_lang = translator.detect(img_str)
web_1  |   File "/usr/local/lib/python3.8/dist-packages/googletrans/client.py", line 255, in detect
web_1  |     data = self._translate(text, 'en', 'auto', kwargs)
web_1  |   File "/usr/local/lib/python3.8/dist-packages/googletrans/client.py", line 78, in _translate
web_1  |     token = self.token_acquirer.do(text)
web_1  |   File "/usr/local/lib/python3.8/dist-packages/googletrans/gtoken.py", line 194, in do
web_1  |     self._update()
web_1  |   File "/usr/local/lib/python3.8/dist-packages/googletrans/gtoken.py", line 62, in _update
web_1  |     code = self.RE_TKK.search(r.text).group(1).replace('var ', '')
web_1  | AttributeError: 'NoneType' object has no attribute 'group'

We could unit test this function and perhap passing it None to see if that's the issue indeed (because I'm not sure it is). A try/catch block would fix this temporarily as well.

OCR Text not entirely clean

While the translated text is cleaned (most of the time), the extracted text is not.

Check out the text at the end of the sentence:

ocr_text: "Improve the local economy by ensuring local businesses win city contracts. \u000c"