A Novel Saliency Prediction Method Based on Fast Radial Symmetry Transform and Its Generalization

Abstract

Symmetry has been observed as an important indicator of visual attention. In this paper, we propose a novel saliency prediction method based on fast radial symmetry transform (FRST) and its generalization (GFRST). We made two contributions. First, a novel saliency predictor based on FRST is proposed. The new approach does not require a whole set of visual features (intensity, color, orientation) as in most previous works but uses only symmetry and center bias to model human fixations at the behavioral level. The new model is shown to have higher prediction accuracy and lower computational complexity than an existing saliency prediction method based on symmetry. Second, we propose using GFRST for predicting visual attention. GFRST is shown to outperform FRST, as it can detect symmetries distorted by parallel projection.