A Privacy-Preserving Filter for Oblique Face Images Based on Adaptive Hopping Gaussian Mixtures

Abstract

Photographs taken in public places often contain faces of bystanders thus leading to a perceived or actual violation of privacy. To address this issue, we propose to pseudo-randomly modify the appearance of face regions in the images using a privacy filter that prevents a human or a face recogniser from inferring the identity of people. The filter, which is applied only when the resolution is high enough for a face to be recognisable, adaptively distorts the face appearance as a function of its resolution. Moreover, the proposed filter locally changes the values of its parameters to counter attacks that attempt to estimate them. The filter exploits both global adaptiveness to reduce distortion and local parameter hopping to make their estimation difficult for an attacker. In order to evaluate the efficiency of the proposed approach, we consider an important scenario of oblique face images: photographs taken with low altitude Micro Aerial Vehicles (MAVs). We use a state-of-the-art face recognition algorithm and synthetically generated face data with 3D geometric image transformations that mimic faces captured from an MAV at different heights and pitch angles. Experimental results show that the proposed filter protects privacy while reducing distortion, and is also robust against attacks.INDEX TERMS Image privacy protection, hopping Gaussian blur, micro aerial vehicles.

Privacy–Enhancing Face Biometrics: A Comprehensive Survey

“…Bonetto et al [63], for example, studied the efficiency of different privacy filters for privacy protection in video footage captured by minidrone based video surveillance systems. Sarwar et al [121], on the other hand, presented an approach that filters facial regions in video footage captured by MAVs adaptively as a function of image resolution. The proposed privacy filter is applied only if the input resolution is high enough to be used with a face matcher.…”

“…Because the meaning of the term biometric utility is very much application dependent, standard evaluation strategies typically define a secondary task that needs to be evaluated after privacy enhancement. With video surveillance systems, for example, utility is often linked to the ability to extract behavioural information [121], with deidentification techniques, utility is frequently measured with the ability to infer facial expressions after deidentification [9], [10], while for soft-biometric privacy problems, utility is routinely defined with the recognition performance after the removal of selected soft-biometric attributes [20], [66]. To quantify the amount of preserved utility, existing work is commonly looking at the performance differences achieved with automatic recognition techniques on the selected secondary task with the original and privacy-enhanced data.…”

“…Several such attack scenarios are considered in the literature in addition to the vanilla scenario discussed above, including [23], [121]:…”

Deep Learning for Privacy in Multimedia

“…In particular, we show how to generate natural-looking adversarial images either by selectively modifying colors within chosen ranges that we perceive as natural or by enhancing details in the image. The references that will be covered in the tutorial are [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18].…”

Trust Hardware Based Secured Privacy Preserving Computation System for Three-Dimensional Data

“…Moreover, data owners have no awareness about how important their personal data is. For example, the refined 3D model reconstruction reveals personal information, including sex and appearance [9][10][11][12][13][14][15]. 3D sensitive data confirmation, security and privacy problems have attracted extensive attention.…”