Napredna pretraga

Pregled bibliografske jedinice broj: 768444

Object Level vs. Scene Level Image Annotation


Ivašić-Kos, Marina; Pobar, Miran; Ipšić, Ivo
Object Level vs. Scene Level Image Annotation // Recent Advances in Electrical and Electronic Engineering, Proceedings of the 3rd International Conference on Circuits, Systems, Communications, Computers and Applications (CSCCA '14) / Mastorakis, Nikos E. ; Nakamatsu, Kazumi ; Paspalakis, Emmanuel (ur.).
Firenca: WSEAS Press, 2014. str. 162-168 (predavanje, međunarodna recenzija, cjeloviti rad (in extenso), znanstveni)


Naslov
Object Level vs. Scene Level Image Annotation

Autori
Ivašić-Kos, Marina ; Pobar, Miran ; Ipšić, Ivo

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Recent Advances in Electrical and Electronic Engineering, Proceedings of the 3rd International Conference on Circuits, Systems, Communications, Computers and Applications (CSCCA '14) / Mastorakis, Nikos E. ; Nakamatsu, Kazumi ; Paspalakis, Emmanuel - Firenca : WSEAS Press, 2014, 162-168

ISBN
978-960-474-399-5

Skup
3rd International Conference on Circuits, Systems, Communications, Computers and Applications (CSCCA '14)

Mjesto i datum
Firenca, Italija, 22-24.11.2014

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
Image annotation; multi-label classification; scene classification

Sažetak
Automatic annotation methods deal with visual features such as color, texture, structure, etc. that can be extracted from the raw image data, and can automatically assign keywords to an unlabeled image. The major goal is to bridge the so-called semantic gap between the available features and keywords that could be useful to humans for image retrieval. Although different people will most likely annotate the same image with different words, most people when searching for images use object or scene labels. Therefore, the aim of this paper is to annotate the images with both object and scene labels, and to compare the performance of automatic image annotation both levels. The assumption is that there can be many objects in each image, but an image can be classified into one scene. The same features sets composed of dominant colors and GIST descriptors for the both annotation levels were used, but different classification methods due to the multi-label classification problem present in object level annotation. The scene level annotation task was performed using the Naïve Bayes classifier and the object level annotation using the RAKEL and ML-kNN multi-label classification methods. The Naïve Bayes classifier was also used in this case, but on transformed data. Preliminary results of scene and object level annotations of outdoor images are compared using different feature subsets.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo, Informacijske i komunikacijske znanosti



POVEZANOST RADA


Ustanove
Sveučilište u Rijeci - Odjel za informatiku