dobai2019document.md

A document detection technique using convolutional neural networks for optical character recognition systems

Lorand Dobai, Mihai Teletin

Browse

@article{dobai2019document,
   title={A document detection technique using convolutional neural networks for optical character recognition systems},
   ISBN={9782875870650},
   journal={ESANN 2019 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence
and Machine Learning},
   author={Dobai, Lorand and Teletin, Mihai},
   year={2019},
   pages={547-552}
}

Pipeline

Receipt detection	Receipt localization	Receipt normalization	Text line segmentation	Optical character recognition	Semantic analysis
❌	✔️	✔️	❌	❌	❌

Receipt localization

• detect the key points of the document

• MobileNet backbone

• Given x, the model computes the location of 4 points in the image domain representing the corners of the cash receipt.

• the pairwise angles formed by the estimated points have to be as close as possible to the ground truth.

• loss function which combines mean squared error and the angular error

Receipt normalization

• skew correction (deskewing), the process of aligning the document in order to make the lines of text as horizontally straight as possible

• OpenCV was used for the projective transformation

Notes

• jointly detect and deskew documents