Security analysis of image copy detection systems based on SIFT descriptors
Year: 2012
Type: dissertation
Abstract: Content-Based Image Retrieval Systems (CBIRS) are now commonly used as a filtering mechanism against the piracy of multimedia contents. These systems often use the SIFT local-feature description scheme as its robustness against a large spectrum of image distortions has been assessed. But none of these systems have addressed the piracy problem from a ''security'' perspective. This thesis checks whether CBIRS are secure: Can pirates mount violent attacks against CBIRS by carefully studying the technology they use? First, we present the security flaws of the typical technology blocks used in state-of-the-art CBIRS. Then, we present very SIFT-specific attacks. The attacks are performed either during the keypoint detection step or during the keypoint description step. We also present a security-oriented Picture in Picture attack deluding CBIRS using post filtering geometric verifications. Experiments with a database made of 100,000 real world images confirm the effectiveness of proposed attacks.
Source: HAL
Author: Thanh-Toan Do
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W113450189