Semantic Video Carving Using Perceptual Hashing and Optical Flow
Abstract
Video files are frequently encountered in digital forensic investigations. However, these files are usually fragmented and are not stored consecutively on physical media. Suspects may logically delete the files and also erase filesystem information. Unlike image carving, limited research has focused on video carving. Current approaches depend on filesystem information or attempt to match every pair of fragments, which is impractical. This chapter proposes a two-stage approach to tackle the problem. The first perceptual grouping stage computes a hash value for each fragment; the Hamming distance between hashes is used to quickly group fragments from the same file. The second precise stitching stage uses optical flow to identify the correct order of fragments in each group. Experiments with the BOSS dataset reveal that the approach is very fast and does not sacrifice accuracy or overall precision.
Domains
Computer Science [cs]| Origin | Files produced by the author(s) |
|---|
Loading...
