RTS-ELM: an approach for saliency-directed image segmentation with ripplet transform

In spite of great advancements in the field of computer vision in recent times, efficient identification of salient regions in an image/scene and applying the results to image segmentation are a fertile area to be explored by researchers. This paper deals with a novel approach for image segmentation called RTS-ELM which uses cues from salient region identification. Initially, salient regions of an image are identified using ripplet transform. Based on the saliency map, a trimap is generated for an image which highlights the dominant regions of an image. Using histogram analysis, the dominant pixels of foreground and background are grouped together to produce the positive and negative groups of training data. The salient regions are then segmented using the trained ELM classifier. After a rigmarole process of comparing with eleven extant approaches using three benchmark datasets, RTS-ELM is found to be an efficient method for reifying effective segmentation in different types of images with only a few errors.