Visual Saliency Map Detection method using Iterative Outlier Cluster Elimination

摘要

The present invention relates to a visual saliency map detection method using iterative outlier cluster elimination. According to the present invention, provided is the visual saliency map detection method using iterative outlier cluster elimination including: (a) a multiple level cluster generation step of extracting an image as a division map; (b) a saliency calculation step of acquiring a saliency map by calculating a saliency score for each cluster in each of the extracted saliency maps; (c) a first saliency map coupling step of acquiring a reference saliency map by coupling a plurality of saliency maps which are acquired; (d) an outlier cluster elimination step of eliminating an outlier cluster by calculating similarity of each saliency map based on the reference saliency map; (e) a second saliency map coupling step of updating the reference saliency map by coupling the saliency map from which the outlier cluster is eliminated. If there is a difference by comparing the updated reference saliency map with a prior reference saliency map, until the difference disappears, the (d) and (e) steps are iteratively performed with the updated reference saliency map. Accordingly, the present invention can reduce a performance loss and improve a finally acquired saliency map.