Volume XLII-2/W12
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W12, 137-142, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W12-137-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-2/W12, 137-142, 2019
https://doi.org/10.5194/isprs-archives-XLII-2-W12-137-2019
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

  09 May 2019

09 May 2019

NEURAL NETWORKS FOR SHAPE RECOGNITION BY MEDIAL REPRESENTATION

N. Lomov and S. Arseev N. Lomov and S. Arseev
  • Lomonosov Moscow State University GSP-1, Leninskie Gory, Moscow, 119991, Russian Federation

Keywords: Deep Learning, Convolutional Neural Networks, Medial Representation, Skeleton

Abstract. The article is dedicated to the development of neural networks that process data of a special kind – a medial representation of the shape, which is considered as a special case of an undirected graph. Methods for solving problems that complicate the processing of data of this type by traditional neural networks – different length of input data, heterogeneity of its structure, unordered constituent elements – are proposed. Skeletal counterparts of standard operations used in convolutional neural networks are formulated. Experiments on character recognition for various fonts, on classification of handwritten digits and data compression using the autoencoder-style architecture are carried out.