Training of neural networks for detecting manipulations in digital images이미지 조작 탐지를 위한 뉴럴 네트워크 학습 방법

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 233
  • Download : 0
Editing images has become easier than ever before with the development and distribution of smartphones with high-end cameras and editing applications. Meanwhile, the distribution of these images become intractable due to social network service and it continues to arise due to fake news using theses manipulated images, therefore, it is important to detect image manipulations. To achieve this goal, it has been common to train a neural network. In this dissertation, the two concepts of training a neural network that is specialized for manipulation detection in digital images. First, we propose BitMix, a data augmentation method for spatial image steganalysis. BitMix mixes a cover and stego image pair by swapping the random patch and generates an embedding adaptive label with the ratio of the number of pixels modified in the swapped patch to those in the cover-stego pair. We explore optimal hyperparameters, the ratio of applying BitMix in the mini-batch, and the size of the bounding box for swapping patch. The results reveal that using BitMix improves the performance of spatial image steganalysis and better than conventional data augmentation methods. Second, a multi-domain data processing of an JPEG compressed image to identify twenty different manipulations is proposed. We proposed MCNet that learns forensic features for each domain through a multi-stream structure and distinguishes manipulations by comprehensively analyzing the fused features. Our work jointly considers visual artifacts caused by image manipulations and compression artifacts due to JPEG compression; therefore, rich forensic features can be explored and learned in the training phase. To demonstrate the effectiveness of the proposed MCNet, extensive experiments were conducted using state-of-the-art baselines. Compared to these baselines, our proposed method outperforms in terms of multi-class manipulation classification. In addition, we experimentally proved that the fine-tuned model based on the multi-class manipulation task was effective for different forensic tasks such as DeepFake detection, seam-carving detection, or integrity authentication of JPEG images.
Advisors
Lee, Heung Kyuresearcher이흥규researcher
Description
한국과학기술원 :전산학부,
Publisher
한국과학기술원
Issue Date
2021
Identifier
325007
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전산학부, 2021.2,[vi, 68 p. :]

Keywords

Image forensics▼aNeural networks▼aImage steganalysis▼aData augmentation▼amulti-domain data processing; 이미지 포렌식▼a뉴럴 네트워크▼a스테가날리시스▼a데이터 증강법▼a다중 도메인 데이터 처리

URI
http://hdl.handle.net/10203/295742
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=956461&flag=dissertation
Appears in Collection
CS-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0