Visual saliency detection using information contents weighting

Abstract

Visual attention is the human ability, which instinctively allows to direct our gaze towards the objects of interest in visual environment. Precise estimation of salient regions without prior knowledge is an important step in many computer vision applications. In this article, we propose a novel idea for saliency detection where we exploited ternary color space with the premonition that salient object is always salient in one or more color channels. Our contribution is threefold: first, use global information to partition image into two clusters namely foreground and background by utilizing contrast stretching and morphological theory. Second, design a weighting criteria based on proposed extended segmentation based fractal texture analysis (ESFTA), boundary connections, number of connected labels and distance from the center. Finally, we enhance saliency while diminishing the background details to generate adept, robust and appealing results. Comprehensive evaluation on standard benchmark datasets and comparison with nine state-of-the-art methods validate the effectiveness of our proposed work.