Saliency detection using centroid weight map
Kang Han Oh, Soo-Hyung Kim, In Seop Na · 2014
This paper presents a novel approach to compute an image saliency map using Centroid Weight map (CWM). Detection of salient image regions is useful for applications like object segmentation and tracking. We developed and tested a method to detect saliency area using 4 features which are color, intensity, Different of Gaussian (DOG), CWM. A key idea is the CWM technique which is followed with N-classes using mean-shift and pixel location variation. The scheme successfully detects saliency region compare with five state-of-the-art saliency region detection method on the 1000 MSRA benchmark database. The proposed method has achieved superior detection accuracy with both higher precision and better recall than traditional method.