Combination of Kullback–Leibler divergence and Manhattan distance measures to detect salient objects

Abstract

Salient object detection is a computer vision technique that filters out redundant visual information and considers potentially relevant parts of our visual field. In this paper, we modify the Liu et al. model for salient object detection, which combines multi-scale contrast, center–surround histogram and color spatial distribution with conditional random fields. A combination of Symmetric Kullback–Leibler divergence and Manhattan distance instead of chi-square measure is employed to determine center–surround histogram difference. The modified Liu et al. model also uses a less computational intensive color spatial distribution map. To check the efficacy of the modified Liu et al. model, the performance is evaluated in terms of precision, recall, $$F$$ -measure, area under curve and computation time. Experiment is carried out on a publicly available image datasets, and performance is compared with Liu et al. model and six other popular state-of-the-art models. Experimental results demonstrate that the modified Liu et al. model outperforms Liu et al. model and other existing state-of-the-art methods in terms of precision, $$F$$ -measure, area under curve and has comparable performance in terms of recall with Liu et al. model.