A progressive image semantic segmentation method using recurrent neural network

To improve the accuracy of image detail structure segmentation is a challenging research topic. In this paper, a progressively refined image semantics segmentation method based on recurrent neural network is presented. By introducing a recurrent processing layer into the neural network, a recurrent network structure is constructed, which makes the image segmentation from coarse to fine. Using the coarse-grained segmentation result of the previous segmentation module, it provides an effective context clue for the next fine-grained segmentation, and achieves progressive, multi-grained semantic image segmentation from coarse to fine. Comparative experiments on several datasets show that: image semantic segmentation based on recurrent neural network can effectively utilize the image context information and improve the accuracy of fine-grained image segmentation.