SRI Security Research Institute, Edith Cowan University, Perth, Western Australia
Hashing algorithms such as MD/SHA variants have been used for years by forensic investigators to look for known artefacts of interest such as malicious files. However, such hashing algorithms are not effective when their hashes change with the slightest alteration in the file. Fuzzy hashing overcame this limitation to a certain extent by providing a close enough measure for slight modifications. As such, image forensics is an essential part of any digital crime investigation, especially in cases involving child pornography. Unfortunately such hashing algorithms can be thwarted easily by operations as simple as saving the original file in a different image format. This paper introduces a novel technique for measuring image similarity using Dynamic Time Warping (DTW) of fractal features taken from the frequency domain. DTW has traditionally been used successfully for speech recognition. Our experiments have shown that it is also effective for measuring image similarity while tolerating minor modifications, which is currently not capable by state-of-the-art tools.