Loading...
A Unified Framework for Saliency Segmentation
Author(s)
Kakanopas, Donyarut
Date Issued
January 1, 2021
Type
Conference Paper
Abstract
Saliency segmentation is an extendable saliency detection that can detect and segment the most interested object(s) of an image. During the past decade, many saliency detection methods have been proposed. Those methods can correctly identify salient locations, but they provide the detected saliency maps with a diversity of intensity values, highlighting only high contrast edges, unevenly salient regions, and ill-defined salient boundaries, thus making them unextendible to saliency segmentation. Therefore, this paper proposes a unified framework for saliency segmentation. This framework consists of three main processes: (i) saliency feature extraction, implemented with a set of i-Hola filters, our recently proposed method, (ii) saliency map selection, using a multi-solution technique in selecting the optimal saliency map, and (iii) saliency segmentation, implemented with an iterative approach. Based on a challenging dataset divided into seven categories with different characteristics, the experimental results showed that our proposed method outperformed the baselines in almost all categories in terms of IOU performance.
Citation
Lecture Notes in Networks and Systems, 251, 191-200, 2021