Home // VISUAL 2016, The First International Conference on Applications and Systems of Visual Paradigms // View article

Full Incremental Learning for Along Classification of Textual Images

Authors:
Vincent Poulain d'Andecy
Aurélie Joseph
Saddok Kebairi

Keywords: incremental classification, text-based vector, shape-based vector, BSM, A2ING, Document Image Processing

Abstract:
Incremental classification is still a challenge with an important industrial impact by allowing a class training process simplification. Recently, works on Incremental Growing Neural Gas (IGNG) have demonstrated the ability of this technology to cope with this challenge for Optical Character Recognition(OCR)-based image classification. Previous proposals focused on the classifier itself but didn’t deal with descriptors which were not in the scope of these studies taking an a priori fixed descriptors set. This assumption is not applicable in real-life when the environment is progressive and the incremental system doesn’t know a priori the image content to learn. In this paper we proposed an enhancement of an incremental system based on an IGNG extension (A2ING) with a combination of graphical, re-using the Blurred Shape Model (BSM), and a novel strategy based on incremental textual descriptors. Performance achievement shows a better precision with an acceptable recall than predefined descriptors. The benefit is to not require a prior descriptors selection.

Pages: 45 to 49

Copyright: Copyright (c) IARIA, 2016

Publication date: November 13, 2016

Published in: conference

ISSN: 2519-8645

ISBN: 978-1-61208-520-3

Location: Barcelona, Spain

Dates: from November 13, 2016 to November 17, 2016