Anomaly detection in images with shared autoencoders
Anomaly detection is a classical problem in computer vision, namely the determination of the normal from the abnormal when datasets are highly biased toward one class (normal) due to the insufficient sample size of the other class (abnormal). We introduce a novel model that utilizes two decoders to...
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2023-01-01
|
| Series: | Frontiers in Neurorobotics |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fnbot.2022.1046867/full |
| _version_ | 1797961657590218752 |
|---|---|
| author | Haoyang Jia Haoyang Jia Wenfen Liu Wenfen Liu |
| author_facet | Haoyang Jia Haoyang Jia Wenfen Liu Wenfen Liu |
| author_sort | Haoyang Jia |
| collection | DOAJ |
| description | Anomaly detection is a classical problem in computer vision, namely the determination of the normal from the abnormal when datasets are highly biased toward one class (normal) due to the insufficient sample size of the other class (abnormal). We introduce a novel model that utilizes two decoders to share two encoders, respectively, forming two sets of network structures of encoder-decoder-encoder called EDE, which are used to map image distributions to predefined latent distributions and vice versa. In addition, we propose an innovative two-stage training mode. The first stage is roughly the same as the traditional autoencoder (AE) training, using the reconstruction loss of images and latent vectors for training. The second stage uses the idea of generative confrontation to send one of the two groups of reconstructed vectors into another EDE structure to generate fake images and latent vectors. This EDE structure needs to achieve two goals to distinguish the source of the data: the first is to maximize the difference between the fake image and the real image; the second is to maximize the difference between the fake latent vector and the reconstructed vector. Another EDE structure has the opposite goal. This network structure combined with special training methods not only well avoids the shortcomings of generative adversarial networks (GANs) and AEs, but also achieves state-of-the-art performance evaluated on several publicly available image datasets. |
| first_indexed | 2024-04-11T01:01:11Z |
| format | Article |
| id | doaj.art-aa0dbca1fd4c4ce98f914586bfebf51e |
| institution | Directory Open Access Journal |
| issn | 1662-5218 |
| language | English |
| last_indexed | 2024-04-11T01:01:11Z |
| publishDate | 2023-01-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Neurorobotics |
| spelling | doaj.art-aa0dbca1fd4c4ce98f914586bfebf51e2023-01-04T16:12:32ZengFrontiers Media S.A.Frontiers in Neurorobotics1662-52182023-01-011610.3389/fnbot.2022.10468671046867Anomaly detection in images with shared autoencodersHaoyang Jia0Haoyang Jia1Wenfen Liu2Wenfen Liu3Guangxi Key Laboratory of Cryptography and Information Security, Guilin University of Electronic Technology, Guilin, Guangxi, ChinaSchool of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin, Guangxi, ChinaGuangxi Key Laboratory of Cryptography and Information Security, Guilin University of Electronic Technology, Guilin, Guangxi, ChinaSchool of Computer Science and Information Security, Guilin University of Electronic Technology, Guilin, Guangxi, ChinaAnomaly detection is a classical problem in computer vision, namely the determination of the normal from the abnormal when datasets are highly biased toward one class (normal) due to the insufficient sample size of the other class (abnormal). We introduce a novel model that utilizes two decoders to share two encoders, respectively, forming two sets of network structures of encoder-decoder-encoder called EDE, which are used to map image distributions to predefined latent distributions and vice versa. In addition, we propose an innovative two-stage training mode. The first stage is roughly the same as the traditional autoencoder (AE) training, using the reconstruction loss of images and latent vectors for training. The second stage uses the idea of generative confrontation to send one of the two groups of reconstructed vectors into another EDE structure to generate fake images and latent vectors. This EDE structure needs to achieve two goals to distinguish the source of the data: the first is to maximize the difference between the fake image and the real image; the second is to maximize the difference between the fake latent vector and the reconstructed vector. Another EDE structure has the opposite goal. This network structure combined with special training methods not only well avoids the shortcomings of generative adversarial networks (GANs) and AEs, but also achieves state-of-the-art performance evaluated on several publicly available image datasets.https://www.frontiersin.org/articles/10.3389/fnbot.2022.1046867/fullanomaly detectionautoencoder (AE)unsupervised learningadversarial networkimage identification |
| spellingShingle | Haoyang Jia Haoyang Jia Wenfen Liu Wenfen Liu Anomaly detection in images with shared autoencoders Frontiers in Neurorobotics anomaly detection autoencoder (AE) unsupervised learning adversarial network image identification |
| title | Anomaly detection in images with shared autoencoders |
| title_full | Anomaly detection in images with shared autoencoders |
| title_fullStr | Anomaly detection in images with shared autoencoders |
| title_full_unstemmed | Anomaly detection in images with shared autoencoders |
| title_short | Anomaly detection in images with shared autoencoders |
| title_sort | anomaly detection in images with shared autoencoders |
| topic | anomaly detection autoencoder (AE) unsupervised learning adversarial network image identification |
| url | https://www.frontiersin.org/articles/10.3389/fnbot.2022.1046867/full |
| work_keys_str_mv | AT haoyangjia anomalydetectioninimageswithsharedautoencoders AT haoyangjia anomalydetectioninimageswithsharedautoencoders AT wenfenliu anomalydetectioninimageswithsharedautoencoders AT wenfenliu anomalydetectioninimageswithsharedautoencoders |