Utilizing 3D Information from Point Clouds to Support Document Image Binarization

EasyChair Preprint 9141

Abstract

One important step for the document processing domain is binarization for degraded document images. We introduce a new method that supports the network using 2D images. This support is based on new features of 3D point clouds converted from 2D images. Specifically, the original 2D images are fed to the LadderNet network to extract 2D information. Besides, the converted 3D point clouds from the original 2D images are fed to network for 3D point cloud input to get unique features. These features are useful and differ-ent from 2D information. So, fusing the output features from two of these architectures achieves a better result. We illustrate the benefit of this idea for binarizing documents.

Keyphrases: 3D point clouds, Fusion Network, Image Binarization, document

@booklet{EasyChair:9141,
  author    = {Quang-Vinh Dang and Guee-Sang Lee},
  title     = {Utilizing 3D Information from Point Clouds to Support  Document Image Binarization},
  howpublished = {EasyChair Preprint 9141},
  year      = {EasyChair, 2022}}