Skip to content

Latest commit

 

History

History
20 lines (9 loc) · 633 Bytes

README.md

File metadata and controls

20 lines (9 loc) · 633 Bytes

Resnet U-net based Scanned Document Table Segmentation

Overview

Data

ICDAR 2013 Table Competition you can download the train and test data from the below link.

https://roundtrippdf.com/en/downloads/

Pre-processing

The dataset contains 27 PDFs, first step is to convert the PDF into images. When we trained the model with original RGB images the segmented result is not good. Applied the distance transform on RGB images before passing into the model. It improved the overall segmentation accuracy.

Proposed method

alt text