In this paper we present a novel saliency-based technique for the automatic extraction of relevant subjects in digital images. We use enhanced saliency maps to determine the most relevant parts of the images and an image cropping technique on the map itself to extract one or more relevant subjects. The contribution of the paper is two-fold as we propose a technique to enhance the standard GBVS saliency map and a technique to extract the most salient parts of the image. The GBVS saliency map is enhanced by applying three filters particularly designed to optimize the performance for the task of relevant subjects extraction. The extraction of relevant subjects is demonstrated on a manually annotated dataset and results are encouraging. A variation of the same technique has also been used to extract the most significant region of an image. This region can then be used to obtain a thumbnail keeping most of the relevant information of the original image and discarding nonsignificant background. Experimental results are reported also in this case.
Greco, L., La Cascia, M., Lo Cascio, F. (2013). Fully automatic saliency-based subjects extraction in digital images. In Proceedings of the 10th International Conference on Signal Processing and Multimedia Applications (pp. 129-136).
Fully automatic saliency-based subjects extraction in digital images
GRECO, Luca;LA CASCIA, Marco;
2013-01-01
Abstract
In this paper we present a novel saliency-based technique for the automatic extraction of relevant subjects in digital images. We use enhanced saliency maps to determine the most relevant parts of the images and an image cropping technique on the map itself to extract one or more relevant subjects. The contribution of the paper is two-fold as we propose a technique to enhance the standard GBVS saliency map and a technique to extract the most salient parts of the image. The GBVS saliency map is enhanced by applying three filters particularly designed to optimize the performance for the task of relevant subjects extraction. The extraction of relevant subjects is demonstrated on a manually annotated dataset and results are encouraging. A variation of the same technique has also been used to extract the most significant region of an image. This region can then be used to obtain a thumbnail keeping most of the relevant information of the original image and discarding nonsignificant background. Experimental results are reported also in this case.File | Dimensione | Formato | |
---|---|---|---|
SigmapCR.pdf
Solo gestori archvio
Tipologia:
Versione Editoriale
Dimensione
1.89 MB
Formato
Adobe PDF
|
1.89 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.