Layout Analysis of scanned pages of modern Books in Arabic (LABA)

This is a implementation of system LABA. download
The code is implemented in Visual Studio Community 2013 and opencv 3.10. User might want to build a similar environment to re-use this code.

The code is able to provide three main uses:

  • 1. detect and label bounding boxes in the given image(s) and compute features used for training SVMs in LABA.
  • 2. train and generate SVMs prediction based on bounding boxes found in 1.
  • 3. generate the final result based on a voting mechanism.

    • How to use the code:

    The main workflow is provided in Source.cpp, divided into three steps, corresponding to the three uses above. User may need to modify the code according to the comments in Source.cpp to allow for training on custom data. Most code modification is only needed to be done on Source.cpp for the three uses.
    The defaut status of the code, step 2 and step 3 especially, is to generate the result of LABA for the fourth fold of experiment mentioned in the LABA paper.
    User may want to run and test to be clear about how the files of input and output are organized. User should be able to run through step 2 and step 3 successfully without any modification.

    IF YOU ARE RUNNING ON YOUR OWN DATASET, PLEASE MAKE SURE FILE NAMES IN DIFFERENT XML FILES ARE CORRECTLY ALIGNED!

    The experiment results of the LABA paper are stored in folder experimentResultXML/ in .xml format (although their ".xml" suffix is not added).

    TODO: 1. add more comments to increase the readability of the code. 2. adding a auto cross-validation code in the current project.

    cite:

    @inproceedings{qin2018laba, title={LABA: Logical Layout Analysis of Book Page Images in Arabic Using Multiple Support Vector Machines}, author={Qin, Wenda and Elanwar, Randa and Betke, Margrit}, booktitle={2018 IEEE 2nd International Workshop on Arabic and Derived Script Analysis and Recognition (ASAR)}, pages={35--40}, year={2018}, organization={IEEE} }

    If you have any further question for the code, please contact wdqin@bu.edu