Table Learning Representation from Scanned PDF Documents Containing Some Red Stamps
Keywords: Table Learning Representation, Table detection, Table recognition, Scanned PDF Documents, Red Stamps
TL;DR: We propose a context-splitting Transformer-based table recognition method for recognizing the handwritten invoice amounts overlapped with a red stamp against the invoice amounts.
Abstract: Generally, it can be challenging to recognize the table contents from scanned PDF documents containing some red stamps and reconcile the recognized table contents.
In this paper, we address the reconciliation challenge involving matching the handwritten invoice amounts overlapped with a red stamp against the invoice amounts found in multi-page tables extracted from scanned PDF documents and we propose a context-splitting Transformer-based table recognition method for recognizing the handwritten invoice amounts overlapped with a red stamp against the invoice amounts.
Firstly, we recognize the layout structure of the table which is detected from the scanned PDF document containing some red stamps on the handwritten invoice amounts of the table.
Secondly, we represent the table cells as context-splitting embedding vectors which involve spatial context embedding, position context embedding, lexical context embedding, and colored context embedding.
Finally, we apply a stack of Transformer-based self-attention encoders to recognize the cross-modality table cells where we multiply the length of query vector and the length of key vector with the scaling factor of the original Transformer in order to make the training process more stable.
We improve the recognition accuracy of table cells with a red stamp on handwritten invoice amounts.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 10770
Loading