HEVC based tampered video database development for forensic investigation

HEVC based tampered video database development for forensic investigation

연구 분야: Analysis

학회: Multimedia Tools and Applications

Nowadays, smartphones are becoming the predominant source of video content that is being widely shared through various platforms of social media. According to recent studies, 78% of iPhones and 57% of android smartphones support hardware-accelerated HEVC decoding. Growing usage of videos on social media applications poses the challenge of distinguishing between authentic and manipulated content. In this paper, we have proposed the development of HEVC based Tampered Video Dataset (HTVD) consisting of diverse scenarios of authentic and forged videos for more comprehensive testing capabilities. The dataset will provide the researchers with a benchmark with a varied range of realistic and smartly tampered videos for validation and comparison of their forensic investigation techniques. The HTVD dataset is developed from videos captured under scenarios of indoor, outdoor, and surveillance shots. It includes 60 original videos, 966 tampered videos, their corresponding ground truth information, and masks. All the original videos are captured with HEVC supported smartphones. Various types of inter-frame forgeries such as frame insertion, frame deletion and frame duplication, and object-based intra-frame forgeries such as cloning, splicing, and inpainting are incorporated to create a diversified database of forged videos. Further, the tampered videos are provided with variations based on the encoding parameters of the codec namely, GOP size, CRF and frame types resulting in a total of 8,694 forged videos. To perform video tampering, the forger needs to perform recompression. However, videos may also undergo recompression while transferring over the internet or exchanged through social media applications. Thus, recompression doesn’t always mean that forgery has been performed. The proposed HTVD dataset provides a video dataset for experimenting with both circumstances. This video dataset is publicly available on https://drive.google.com/drive/folders/143NEyVjcHNVDDZVzIZm6nTEUk5-ezWb5?usp=sharing.