Semantic Image Segmentation using Deep Convolutional Neural Networks and Super-Pixels

摘要

In this paper Deep Convolution Neural Network (DCNN) with multiple layers are projected. Multiple layers work to build an improved feature space. First layer learns 1st order features (e.g. edge). Second layer learns higher order features (combinations of edges, etc.). Some models learn in an unsupervised mode and identify general characteristics of the input space. Final layer of transformed features is fed into supervised layers and an entire network is often subsequently tuned using supervised training using the initial weightings learned in the unsupervised phase. This deep convolution neural network gives a solution in the form of color segmentation. Deep CNN based segmenta-tion model shows 93% accuracy on BSDS300 dataset.