图像分割述评:基本概貌、典型算法及比较分析
Review of Image Segmentation:Basic Overview,Typical Algorithms and Comparative Analysis
张婧 1张策 1张茹 2王宇彬 1张展 3苏子旸 1吕为工1
作者信息
- 1. 哈尔滨工业大学(威海)计算机科学与技术学院,山东 威海 264209
- 2. 哈尔滨商业大学 管理学院,黑龙江 哈尔滨 150076
- 3. 哈尔滨工业大学 计算机科学与技术学院,黑龙江 哈尔滨 150001
- 折叠
摘要
图像分割作为计算机视觉领域的一个重要分支,在可穿戴计算、自动驾驶、医学图像分析等方面都发挥着重要作用,并有着广泛应用.为了更好地了解图像分割领域的发展以及研究现状,该文对图像分割进行了深入梳理和系统述评.首先,对图像分割的含义以及其工作流程、指标等进行阐释;然后,对图像分割方法按照时间的跨度进行分类——基于阈值和区域、基于图论和聚类,以及基于深度学习的图像分割,对每类方法的代表性算法进行分析介绍,较为全面地总结了每类方法的基本思想和优缺点;最后,对该领域目前存在的问题和未来的发展方向进行展望,提出实时图像语义分割、弱监督或非监督语义分割和三维场景的语义分割是目前研究中的主要挑战.
Abstract
As an important field of computer vision,image segmentation plays an important role in medical image analysis,automatic driving,wearable computing and so on,and has wide application.In order to better understand the development and research status of image segmentation,we make a thorough review and systematic review of image segmentation.Firstly,we explain the meaning of image segmentation and its workflow and indicators.Then,the image segmentation algorithms are classified according to the time span-based on threshold and region,based on graph theory and clustering,as well as image segmentation based on deep learning.The representative algorithms of each type of methods are analyzed and introduced,and their basic ideas,advantages and disadvantages are comprehensively summarized.Finally,we look forward to the current problems and future development direction in this field.Real-time image semantic segmentation,weakly supervised or unsupervised semantic segmentation and three-dimensional scene semantic segmentation are the main challenges in current research.
关键词
图像分割/分割算法/语义分割/深度学习/像素分类Key words
image segmentation/segmentation algorithm/semantic segmentation/deep learning/pixel classification引用本文复制引用
基金项目
国家自然科学基金(61473097)
山东省重点研发计划项目(GG201703130116)
山东省重点研发计划项目(GG201703040002)
2021年度山东省自然科学基金面上项目(ZR2021MF067)
威海市科技发展计划项目(ITEAZMZ001807)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据