Title: Word Recognition using Embedded Prototype Subspace Classifiers on a New Imbalanced Dataset
Authors: Hast, Anders
Vats, Ekta
Citation: Journal of WSCG. 2021, vol. 29, no. 1-2, p. 39-47.
Issue Date: 2021
Publisher: Václav Skala - UNION Agency
Document type: článek
URI: http://wscg.zcu.cz/WSCG2021/2021-J-WSCG-1-2.pdf
ISSN: 1213-6972 (print)
1213-6980 (CD-ROM)
1213-6964 (on-line)
Keywords: podprostory;vestavěné prototypy;clustering;hluboké učení;samoorganizující se mapy;t-SNE;rozdělení dat
Keywords in different language: subspaces;Embedded Prototypes;clustering;deep learning;Self Organising Maps;t-SNE;Data splits
Abstract in different language: This paper presents an approach towards word recognition based on embedded prototype subspace classification.The purpose of this paper is three-fold. Firstly, a new dataset for word recognition is presented, which is extractedfrom the Esposalles database consisting of the Barcelona cathedral marriage records. Secondly, different clusteringtechniques are evaluated for Embedded Prototype Subspace Classifiers. The dataset, containing 30 different classesof words is heavily imbalanced, and some word classes are very similar, which renders the classification task ratherchallenging. For ease of use, no stratified sampling is done in advance, and the impact of different data splits isevaluated for different clustering techniques. It will be demonstrated that the original clustering technique based onscaling the bandwidth has to be adjusted for this new dataset. Thirdly, an algorithm is therefore proposed that findskclusters, striving to obtain a certain amount of feature points in each cluster, rather than finding some clustersbased on scaling the Silverman’s rule of thumb. Furthermore, Self Organising Maps are also evaluated as both aclustering and embedding technique.
Rights: © Václav Skala - UNION Agency
Appears in Collections:Volume 29, Number 1-2 (2021)

Files in This Item:
File Description SizeFormat 
J41.pdfPlný text3,49 MBAdobe PDFView/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/11025/44947

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.