| Title |
U-Net based Camera Model Identification for Facebook Images |
| Authors |
임상균(Sangkyun Im) ; 강상욱(Sang-ug Kang) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.3.35 |
| Keywords |
Camera model identification; Source camera identification; Post-processed image |
| Abstract |
Due to the recent development of network services such as social network services and streaming services, illegal images that forensic analyst acquire through networks are high possibility to be post-processed images rather than the original. However, most of the previous camera model identification forensic techniques that track the captured camera model have been developed using the original image that is not deformed after being photographed. It was also confirmed that the accuracy of camera model identification decreases if the previous research technology is applied to the post-processed image. Therefore, in order to realistically apply the camera model identification technology, it is necessary to develop a camera model identification technology using a post-processed image. In this paper, we propose a technique for finally identifying a camera model by restoring a converted image downloaded from Facebook that one of the most popular SNS to the original image, and applying the previous camera model technology to restored image. As a result of the experiment, the identification accuracy of the camera model for the original image was 92.05%, the identification accuracy for the converted image was 73.49%, and the identification accuracy for the restored image was 83.47%. Although the accuracy for restored image was lower than the accuracy for original image, it showed that it achieved higher accuracy than the converted image. |