图像异常检测研究现状综述  

Review of Image Anomaly Detection

作　　者：吕承侃 沈飞[1,2,3] 张正涛 张峰[1,2,3] LV Cheng-Kan;SHEN Fei;ZHANG Zheng-Tao;ZHANG Feng(Research Center of Precision Sensing and Control,Institute of Automation,Chinese Academy of Sciences,Beijing 100190;School of Artificial Intelligence,University of Chinese Academy of Sciences,Beijing 100049;CASI Vision Technology CO.,LTD.,Luoyang 471000)

机构地区：[1]中国科学院自动化研究所精密感知与控制研究中心,北京100190 [2]中国科学院大学人工智能学院,北京100049 [3]中科慧远视觉技术(洛阳)有限公司,洛阳471000

出　　处：《自动化学报》2022年第6期1402-1428,共27页Acta Automatica Sinica

基　　金：中国科学院青年创新促进会(2020139)资助。

摘　　要：图像异常检测是计算机视觉领域的一个热门研究课题,其目标是在不使用真实异常样本的情况下,利用现有的正常样本构建模型以检测可能出现的各种异常图像,在工业外观缺陷检测、医学图像分析、高光谱图像处理等领域有较高的研究意义和应用价值.本文首先介绍了异常的定义以及常见的异常类型.然后,本文根据在模型构建过程中有无神经网络的参与,将图像异常检测方法分为基于传统方法和基于深度学习两大类型,并分别对相应的检测方法的设计思路、优点和局限性进行了综述与分析.其次,梳理了图像异常检测任务中面临的主要挑战.最后,对该领域未来可能的研究方向进行了展望.Image anomaly detection is a hot research topic in the field of computer vision, which aims at training a model by normal images to detect potential anomalous images without requiring any real anomalous images for training. Therefore, it has high research and application value in the fields of industrial surface defect detection,medical image analysis, and hyperspectral image processing. This paper firstly introduces the definition of anomaly and its common type. Secondly, image anomaly detection methods are divided into traditional and deep learningbased methods according to whether the neural network is involved. Meanwhile, the design ideas, advantages and limitations of these methods are summarized and analyzed respectively. Thirdly, the major challenges faced in image anomaly detection are introduced. Finally, the future directions of image anomaly detection are discussed.

关 键 词：图像异常检测 计算机视觉 深度学习 神经网络 背景重构 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]