LAUSR

Humans possess an intelligent system which effortlessly detect salient objects with high accuracy in real-time. It is a challenge to develop a computational model which can mimic human behavior such that the model achieves better detection accuracy and takes less computation time. So far the research community have suggested models which achieve better detection accuracy but at the cost of computation time and vice versa. In this paper, we attempted to realize a model that takes less computational time and simultaneously achieves higher detection accuracy. In the proposed model the original image is divided into m superpixels using SLIC superpixels algorithm and then these superpixels are clustered into k regions using k-means algorithm. Thereafter the result of the k-means clustering is used to build Gaussian mixture model whose parameters are refined using Expectation-Maximization algorithm. Finally the spatial variance of the clusters is computed and a center-weighted saliency map is computed. The performance of the proposed model and seventeen related models is evaluated both qualitatively and quantitatively on seven publicly available datasets. Experimental results show that the proposed model outperforms the existing models in terms of precision, recall and F -measure on all the seven datasets and in terms of area under curve on four datasets. Also, the proposed model takes less computation time in comparison to many methods.