Original scientific paper

An improved saliency detection algorithm based on Itti’s model

Zhongshan Chen ; (1) School of Electronic Science & Engineering, Southeast University, NO. 2 Sipailou, 210096, Nanjing, China; (2) School of Information Engineering Yancheng Institute of Technology, NO. 211 Jianjun East Road, 224051, Yancheng, China
Yan Tu ; School of Electronic Science & Engineering, Southeast University, NO. 2 Sipailou, 210096, Nanjing, China
Lili Wang ; School of Electronic Science & Engineering, Southeast University, NO. 2 Sipailou, 210096, Nanjing, China

Abstract

Visual attention mechanism (VAM) automatically ignores the superfluous information and pays attention to the most significant objects when people are watching the pictures. There are numerous bottom-up visual attention computational models to detect the salient area of an image. In this paper, an improved visual attention computational model based on Itti’s model is proposed, which is comprised of three components. Firstly, the lower-level primitive image features s are extracted from CIELa*b* color space instead of RGB color space; secondly, the feature images are decomposed into wavelet pyramids by wavelet-based multi-scale transform. Thirdly, a new strategy is used to combine all conspicuity maps into a final saliency map with different weights, which are proportional to the contribution of each conspicuity map. Compared with Itti’s models, subjective experiments prove that the approach proposed in this paper is more effective.

Keywords

bottom-up model; image feature; saliency map; visual attention mechanism (VAM)

Hrčak ID:

131327

URI

https://hrcak.srce.hr/131327

Publication date:

21.12.2014.

Article data in other languages: croatian

Visits: 1.994 *