My Problem is the following. I scan a pdf inside my Flutter App and then need to process around 6/7 tables on the scanned Page. The tables all have a visible Border and more or less look similar. Because on Device OCR isn't really possible for this use case (or at least I think it's the case for Flutter), I had the idea to create a Python Flask Server which would serve me this Task. I had done a lot of research regarding OCR in Python but i don't seem to find the right fit for my use case, because the accuracy is always so low.
I have tried these Technologies but none seem to fit my Application:
- Using the "img2table" python package in connection with TesseractOCR and PaddleOCR but the accuracy was rather low
- Using "img2table" for table detection and then the OCR.Space API for each extracted Table (did work a lot better, but way to slow)
- Using the "deepdoctection" Pipeline which seemed promising, but the serving and usage was too complex for me (maybe this is the right solution but i don't know how to set it up correctly)
Does somebody here know a way to do this (excluding paid Models like Amazon Textract)?
