A novel method based on chan vese segmentation for salient structure detection

Visual saliency detection is an advanced approach which analyse the noticeable objects from hasty scenes in our day to day routines. But fixation prediction model i n messy environment is tough to put into action. Contour Based Spatial Prior(CBSP) method is currently used in extracting salient structures from messy scenes. Irregular edges is one among the shortcoming of salient detected image segments. In this paper, a novel technique is suggested to segment the salient objects from both complex and simple scenes. The method is a two path-way based searching scheme and provides processing of local and global information in parallel. Chan Vese based segmentation is used to extract the contours. Multilevel thresholding is used for YCbCr colorspace. The depth of the salient object is also estimated to classify the objects inside the scenes. This method is successfully tested with MSRA, ECSSD dataset and acquired an accuracy of above 94%.