Article Preview
TopIntroduction
Anomaly detection (Du et al., 2022) is an essential area of research in machine learning. It is a method of constructing a model using unlabeled or normal samples to detect anomalous samples that differ from the desired pattern. Anomaly detection has a wide range of applications in various fields, such as defect detection, medical image analysis, hyperspectral image processing, abnormal behavior detection, and image and video processing. Table 1 shows that many methods have been applied to image anomaly detection in various fields.
Table 1. Applications of Image Anomaly Detection
Areas of application | Specific application |
---|
Defect detection (Wu et al., 2023)
| Inspection of surface defects on a variety of products including textured surfaces, such as cloth (Zhao et al., 2010), glass (Hu et al., 2022), steel (Zheng et al., 2022), cement (Meng & Zhu, 2023), and objects such as printed circuit boards (Alghassab, 2022) and wine bottles (Napoletano et al., 2021).
|
Medical image analysis (Ramaraj et al., 2020; Tschuchnig & Gadermayr, 2022)
| Detection of possible lesion areas in medical images, such as MRI images (Ramaraj et al., 2020), iris images (Modwel et al., 2021), and fundus retinal images (Hussain & Holambe, 2015).
|
Hyperspectral image processing (Yan et al., 2021)
| Marine vessel detection (Yu et al., 2022), ground anomaly area detection (Ning et al., 2023), etc.
|
Power image analysis | Defective power equipment classification (Wang et al., 2022), etc. |
Early anomaly detection algorithms were primarily applied in the field of data mining (Pasini, 2021). In recent years, with the development of computer vision, deep learning, and related technologies (Gao et al., 2023; Kaur et al., 2022; Luo & Pundlik, 2022; You et al., 2023), many studies have introduced anomaly detection into image processing to solve the problem of target detection in the case of sample scarcity. Figure 1 shows the application of image anomaly detection in power equipment protection.
Figure 1. Purpose of Image Anomaly Detection Algorithm
The general computational framework of the image anomaly detection algorithm is shown in Figure 2. After the original image is input into the system, the image features are first extracted. Then, the image features are classified, and an image anomaly detection model is trained using machine learning algorithms and the input pre-labeled image feature data. It can then be used to detect and identify image data anomalies in a new image dataset (Ye & Zhao, 2023).
Figure 2. General Flow of Image Anomaly Detection Algorithm