A Video Splicing Forgery Detection and Localization Algorithm Based on Sensor Pattern Noise

Abstract

Video splicing forgery is a common object-based intra-frame forgery operation. It refers to copying some regions, usually moving foreground objects, from one video to another. The splicing video usually contains two different modes of camera sensor pattern noise (SPN). Therefore, the SPN, which is called a camera fingerprint, can be used to detect video splicing operations. The paper proposes a video splicing detection and localization scheme based on SPN, which consists of detecting moving objects, estimating reference SPN, and calculating signed peak-to-correlation energy (SPCE). Firstly, foreground objects of the frame are extracted, and then, reference SPN are trained using frames without foreground objects. Finally, the SPCE is calculated at the block level to distinguish forged objects from normal objects. Experimental results demonstrate that the method can accurately locate the tampered area and has higher detection accuracy. In terms of accuracy and F1-score, our method achieves 0.914 and 0.912, respectively.

OIDS-45: A large-scale benchmark insect dataset for orchard pest monitoring

“…We evaluated 12 more advanced or typical target detection methods on the OIDS-45 dataset, and based on the features of the OIDS-45 dataset, we also made targeted improvements to the corresponding network, trying to be more relevant to the actual application situation. We employ both single-stage target detection models, such as YOLOv8 [53], YOLOv7…”

Spatio-temporal co-attention fusion network for video splicing localization

“…The traditional artificial feature approaches rely on detecting the artifacts left by compression, 9 imaging pipeline, 10 and inter-frame motion. 11,12 A sensor pattern noise (SPN)-based approach 13 is proposed to detect spliced foreground regions in videos, addressing the common scenario where spliced videos involve two different modes of SPN. To discover manipulated videos, Avino et al 14 and Johnston et al 15 leverage autoencoders to learn the intrinsic model of genuine videos.…”