CNN training with graph-based sample preselection: application to handwritten character recognition

Frédéric Rayar; Masanori Goto; Seiichi Uchida

CNN training with graph-based sample preselection: application to handwritten character recognition

Machine Learning 2018-03-07 v3

Authors: Frédéric Rayar , Masanori Goto , Seiichi Uchida

Abstract

In this paper, we present a study on sample preselection in large training data set for CNN-based classification. To do so, we structure the input data set in a network representation, namely the Relative Neighbourhood Graph, and then extract some vectors of interest. The proposed preselection method is evaluated in the context of handwritten character recognition, by using two data sets, up to several hundred thousands of images. It is shown that the graph-based preselection can reduce the training data set without degrading the recognition accuracy of a non pretrained CNN shallow model.

Keywords

convolutional neural network handwritten character recognition graph neural network

Cite

@article{arxiv.1712.02122,
  title  = {CNN training with graph-based sample preselection: application to handwritten character recognition},
  author = {Frédéric Rayar and Masanori Goto and Seiichi Uchida},
  journal= {arXiv preprint arXiv:1712.02122},
  year   = {2018}
}

Comments

Paper of 10 pages. Minor spelling corrections brought regarding the v2. Accepted as an oral paper in the 13th IAPR Internationale Workshop on Document Analysis Systems (DAS 2018)

CNN training with graph-based sample preselection: application to handwritten character recognition

Abstract

Keywords

Cite

Comments

Related papers