Publication detail

Automatic Image Labelling using Similarity Measures

UHER, V. BURGET, R. KARÁSEK, J. MAŠEK, J. DUTTA, M.

Original Title

Automatic Image Labelling using Similarity Measures

English Title

Automatic Image Labelling using Similarity Measures

Type

conference paper

Language

en

Original Abstract

Scene classification based on global features. It can be used, for example, for annotating large databases of photos. The whole process has several steps. The first step is features extraction, and then the distance between a new image and reference images is calculated. A model is trained to classify new images based on this distance. The model was created using the Naïve Bayes classifier. To improve accuracy the forward selection was used, which optimizes the selection of a group of attributes. The overall performance on the testing dataset was 69.76%.

English abstract

Scene classification based on global features. It can be used, for example, for annotating large databases of photos. The whole process has several steps. The first step is features extraction, and then the distance between a new image and reference images is calculated. A model is trained to classify new images based on this distance. The model was created using the Naïve Bayes classifier. To improve accuracy the forward selection was used, which optimizes the selection of a group of attributes. The overall performance on the testing dataset was 69.76%.

Keywords

Scene classification; image labelling; machine learning; image processing

RIV year

2014

Released

12.01.2015

Publisher

IEEE

Location

Greater Noida

ISBN

978-1-4799-5096-6

Book

MEDCOM 2014 CD-ROM

Pages from

101

Pages to

104

Pages count

4

URL

BibTex


@inproceedings{BUT109408,
  author="Václav {Uher} and Radim {Burget} and Jan {Karásek} and Jan {Mašek} and Malay Kishore {Dutta}",
  title="Automatic Image Labelling using Similarity Measures",
  annote="Scene classification based on global features. It can be used, for example, for annotating large databases of photos. The whole process has several steps. The first step is features extraction, and then the distance between a new image and reference images is calculated. A model is trained to classify new images based on this distance. The model was created using the Naïve Bayes classifier. To improve accuracy the forward selection was used, which optimizes the selection of a group of attributes. The overall performance on the testing dataset was 69.76%.",
  address="IEEE",
  booktitle="MEDCOM 2014 CD-ROM",
  chapter="109408",
  doi="10.1109/MedCom.2014.7005984",
  howpublished="electronic, physical medium",
  institution="IEEE",
  year="2015",
  month="january",
  pages="101--104",
  publisher="IEEE",
  type="conference paper"
}