DoFNet: Depth of Field Difference Learning for Detecting Image Forgery

Yonghyun Jeong, Jongwon Choi, Doyeon Kim, Sehyeon Park, Minki Hong, Changhyun Park, Seungjai Min, Youngjune Gwon; Proceedings of the Asian Conference on Computer Vision (ACCV), 2020


Recently, online transactions have had exponential growth and expanded to various cases, such as opening bank accounts and filing for insurance claims. Despite the effort of many companies requiring their own mobile applications to capture images for online transactions, it is difficult to restrict users from taking a picture of other's images displayed on a screen. To detect such cases, we propose a novel approach using paired images with different depth of field (DoF) for distinguishing the real images and the display images. Also, we introduce a new dataset containing 2,752 pairs of images capturing real and display objects on various types of displays, which is the largest real dataset employing DoF with multi-focus. Furthermore, we develop a new framework to concentrate on the difference of DoF in paired images, while avoiding learning individual display artifacts. Since DoF lies on the optical fundamentals, the framework can be widely utilized with any camera, and its performance shows at least 23% improvement compared to the conventional classification models.

Related Material

@InProceedings{Jeong_2020_ACCV, author = {Jeong, Yonghyun and Choi, Jongwon and Kim, Doyeon and Park, Sehyeon and Hong, Minki and Park, Changhyun and Min, Seungjai and Gwon, Youngjune}, title = {DoFNet: Depth of Field Difference Learning for Detecting Image Forgery}, booktitle = {Proceedings of the Asian Conference on Computer Vision (ACCV)}, month = {November}, year = {2020} }