Two novel graph theory based algorithms for salient object detection

Salient object detection deals with location of the salient object in a given scene followed by segmentation. For efficient detection of salient object from the background, graph theory proved to be a very powerful tool. To improve the performance of the graph cut algorithm for salient object detection, two different models based on color channels have been proposed. Both of the models deals with the estimated features from each available color channels of a given scene. Both of the models also considered the neighbourhood information based on the estimated features from given color channels. To prove the superiority of the proposed models, different experiments are carried out on images taken from Berkley database and Corel database. The effectiveness of the proposed algorithms over Otsu and K-means clustering are successfully tested based on three validity indices.