디지털 증거물 확보를 위한 인공지능 기반 화질 개선 및 정량적 검증기법 제안 (ESRGAN 알고리즘 중심으로)
Proposed AI-based quality improvement and quantitative verification techniques to secure digital evidence (Focusing on ESRGAN algorithms)
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
현재 우리 사회는 디지털 장비의 사용이 대중화되고 디지털 포렌식 기술이발전함에 따라 각종 디지털 기기로부터 획득되는 증거물에 대한 중요성도높아지고 있다. 또한 군 수사기관에서 다뤄지는 디지털 증거물의 범위와 양도기하급수적으로 늘어나는 추세이다. 하지만 여러 가지 기법으로 복원된 증거물의복원율과 원본과의 유사도에 대한 정량적인 평가는 어려운 실정이다. 본연구에서는 디지털 포렌식을 통해 획득되는 이미지 중 저화질 증거물을 딥러닝기반 알고리즘인 ESRGAN(Enhanced Super Resolution Generative Adversarial Network)을 활용하여 화질을 개선 시키고, FID(Frechet Inception Distance) 알고리즘을 통해 복원율과 원본과의 유사도를 객관적으로 검증하여ESRGAN으로 복원된 이미지가 증거물로서 인정받기 위한 연구를 진행하였다
Abstract
Currently, as the use of digital equipment becomes popular and digital forensics technology develops in our society, the importance of evidence obtained from various digital devices is also increasing. In addition, the scope and amount of digital evidence covered by military investigative agencies are increasing exponentially. However, it is difficult to quantitatively evaluate the restoration rate of evidence restored by various techniques and the similarity to the original. In this study, low-quality evidence obtained through digital forensics was improved using ESRGAN (Enhanced Super Resolution Generative Adversarial Network), a deep learning-based algorithm, and the similarity to the original was objectively verified.
참고문헌 (10)
- "Fake video, make sure to leave a trace."We will trace the invisible truth to the end"[URL]
- "video/voice reading and improvement"[URL]
- Ian J. Goodfellow, "Generative Adversarial Networks" (2014), University of Montreal
- 한주혁, "GAN을 활용한 백혈구 수치 예측 인공지능 모델에 관한 연구", 차세대융합기술학회논문지, 6(10), pp.1855~1862 (2022), 국제차세대융합기술학회[DOI]
- Christian Ledig, "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network", The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp.4681~4690 (2017)
- Xintao Wang, "ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks" (2018), The Chinese University of Hong Kong
- Martin Heusel, "GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium" (2017), Johannes Kepler University Linz
- Christian Szegedy, "Going Deeper with Convolutions", Google Inc, ~ (2014)
- Cho Chul-woo, "Improving recognition rate by improving vehicle license plate quality using ESRGAN", Law and Science, (6), pp.52~53 (2022), Scientific Investigation Department of the Supreme Prosecutors' Office
- "Roboflow"[URL]