基于DCNN的图像语义分割综述

图像的语义分割是计算机视觉中重要的基本问题之一,其目标是对图像的每个像素点进行分类,将图像分割为若干个视觉上有意义的或感兴趣的区域,以利于后续的图像分析和视觉理解.近年来,深度卷积神经网络(Deep Convolutional Neural Network,DCNN)的出现,极大地推动了语义分割的发展.本文从语义分割的基本定义出发,对语义分割中存在的困难和挑战进行了分析和描述.总结了目前用于评测语义分割算法的典型数据库,并以PASCAL VOC数据库为主线对近年来基于DCNN的语义分割算法进行了梳理和总结.最后对语义分割未来的研究重点进行了探讨和预测.

A review on image semantic segmentation based on DCNN

Image semantic segmentation is one of the most important problems in computer vision, whose target is to assign a semantic label for each pixel of a given image and segment the image into several visually meaningful or interest regions,in order for the image analysis and vision un-derstanding .In recent years,tremendous progress has been made for semantic segmentation due to the development of deep convolutional neural network (DCNN).In this paper,we firstly dis-cuss the difficulties and challenges of semantic segmentation.Then,DCNN-based achievements in the study of semantic segmentation are reviewed.In addition,an analysis for semantic segmen-tation based on PASCAL VOC dataset is given,which is generally acknowledged as a popular public evaluation for semantic segmentation.Finally,some probable development directions of semantic segmentation are discussed.