Image segmentation using encoder-decoder architecture and region consistency activation

Abstract

An Encoder-Decoder Neural Network Architecture is combined with a novel strategy to improve global label consistency, to come with an improved image segmentation model. Label Distribution predictions extracted from the SegNet Network is investigated and used in the project for image labeling. An algorithm called Region Consistency Activation (RCA) to improve the global label consistency is implemented. RCA is based on a novel transformation between Ultra metric Contour Map (UCM) and the Probability of Regions Consistency (PRC). These algorithms were rigorously tested on the CamVid dataset. Superior performances were achieved compared with the state-of-the-art methods on this dataset.