| Presentation | 2022-05-17 A study of adversarial example detection using the correlation between adversarial noise and JPEG compression-derived distortion Kenta Tsunomori, Yuma Yamasaki, Minoru Kuribayashi, Nobuo Funabiki, Isao Echizen, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Adversarial examples cause misclassification of image classifiers. Higashi et al. proposed a method to detect adversarial examples by applying noise reduction filters of different strengths to input images and observing changes in the output of the image classifier. This method used 14 different filters, which was computationally expensive. In this paper, we propose a method that demonstrates high detection accuracy with a small number of filters. Based on Higashi et al.'s report, JPEG compression is considered to be a suitable filter for denoising adversarial noises. In the proposed method, a distortion signal created from the difference of images before and after JPEG compression is used as a denoising filter. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Adversarial examples / Convolutional neural network / JPEG compression / Scaling / JPEG compression-derived distortion |
| Paper # | IT2022-6,EMM2022-6 |
| Date of Issue | 2022-05-10 (IT, EMM) |
| Conference Information | |
| Committee | IT / EMM |
|---|---|
| Conference Date | 2022/5/17(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Gifu University |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Information Security, Information Theory, Information Hiding, etc. |
| Chair | Tadashi Wadayama(Nagoya Inst. of Tech.) / Ryoichi Nishimura(NICT) |
| Vice Chair | Tetsuya Kojima(Tokyo Kosen) / Masaaki Fujiyoshi(Tokyo Metropolitan Univ.) / Masatsugu Ichino(Univ. of Electro-Comm.) |
| Secretary | Tetsuya Kojima(Saitamai Univ.) / Masaaki Fujiyoshi(Yamaguchi Univ.) / Masatsugu Ichino(Utsunomiya Univ.) |
| Assistant | Masanori Hirotomo(Saga Univ.) / Shoko Imaizumi(Chiba Univ.) / Youichi Takashima(Kaishi Professional Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Information Theory / Technical Committee on Enriched MultiMedia |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A study of adversarial example detection using the correlation between adversarial noise and JPEG compression-derived distortion |
| Sub Title (in English) | |
| Keyword(1) | Adversarial examples |
| Keyword(2) | Convolutional neural network |
| Keyword(3) | JPEG compression |
| Keyword(4) | Scaling |
| Keyword(5) | JPEG compression-derived distortion |
| 1st Author's Name | Kenta Tsunomori |
| 1st Author's Affiliation | Okayama University(Okayama Univ.) |
| 2nd Author's Name | Yuma Yamasaki |
| 2nd Author's Affiliation | Okayama University(Okayama Univ.) |
| 3rd Author's Name | Minoru Kuribayashi |
| 3rd Author's Affiliation | Okayama University(Okayama Univ.) |
| 4th Author's Name | Nobuo Funabiki |
| 4th Author's Affiliation | Okayama University(Okayama Univ.) |
| 5th Author's Name | Isao Echizen |
| 5th Author's Affiliation | National Institute of Informatics(NII) |
| Date | 2022-05-17 |
| Paper # | IT2022-6,EMM2022-6 |
| Volume (vol) | vol.122 |
| Number (no) | IT-25,EMM-26 |
| Page | pp.pp.29-34(IT), pp.29-34(EMM), |
| #Pages | 6 |
| Date of Issue | 2022-05-10 (IT, EMM) |