Semantic Segmentation using Deep Convolutional Neural Network: A Review

Abstract

Image segmentation is the process of assigning each pixel of the image to a class label. It is a sub-field of computer vision, in which the aim is to divide an image into multiple segments. It has various applications such as automated cars, delivery drones, object recognition, security, and monitoring, etc. With the advent of neural networks, deep convolutional neural networks (DCNNs) provide benchmarking results in the problems related to computer vision. Manifold DCNNs have been proposed for semantic segmentation such as UNet, DeepUNet, ResUNet, DenseNet, RefineNet, etc. The general procedure is common for all the models. It has three phases - pre-processing, processing and output generation. The outputs of the processing phase are the masked image and segmented image. In this paper, a systematic critique of the existing DCNNs for semantic segmentation has been manifested. The datasets and the architectures of the existing models have also been discussed in this paper with illustrations.

Keywords: Semantic Segmentation, DCNNs, Deep Learning, Computer Vision, Autoencoder, Deep Convolutional Neural Network