Physical constraints shape adversarial patches for aerial image recognition

Printability-Constrained Adversarial Decals for Near-Nadir Aerial Perception: Measured Ink Gamuts, Nested Realism Constraints, and a Physical-World Bound

Computer Vision and Pattern Recognition

Summary

Adversarial patches are patterns designed to confuse aerial image recognition systems, but most studies test these patches digitally, ignoring real-world printing limits. The authors studied how color, size, and viewing conditions affect patch effectiveness when physically printed and viewed from above. They found that using a real printer’s color range can change attack success depending on the target system, but limiting colors alone doesn’t consistently weaken attacks. Physical tests with printed patches showed some differences from simulations, setting a boundary on real-world effectiveness. This work helps better understand how printed adversarial patterns perform in aerial settings compared to purely digital designs.

What this means in practice

  • For drone surveillance teams: Test printed adversarial decals within operational deployments to understand real-world risks to aerial recognition systems under physical constraints.
  • For security device manufacturers: Develop evaluation tools that include printer color limitations and real viewing conditions for designing more robust aerial detection systems.

Authors

Sandesh Shrestha, K. T. Yasas Mahima, Asanka G. Perera

Abstract

Adversarial patches for aerial perception are typically evaluated as digital composites, with printing left as an implementation detail. This study imposes three physical constraints during optimization rather than after it: the color range a particular printer can reproduce, the size of the flat panel a vehicle offers, and the loss of fine detail incurred when the patch is imaged from altitude. The principal comparison isolates the ink set. Two patches share all seventeen recorded optimization settings and differ only in the colors available to them. One is constrained to a uniform color cube; the other to a gamut measured by printing and scanning a 216-patch chart. Each was optimized at three seeds and evaluated against thirteen victim conditions, with every rate reported against a size-matched optimized control. The effect of the measured gamut is victim-dependent rather than uniform. Net attack success rises on three of six closed-set segmentation victims, and for these the seed ranges of the two ink sets are disjoint: $+0.120$ on DeepLabv3-R101 and $+0.041$ on SegFormer-B0. The color-cube patch is consistently stronger on the open-vocabulary segmenter and on two of four detectors, though no detector exceeds a net of $+0.026$ under either ink set. The natural explanation is that a printable palette is simply less chromatic and lower in frequency than a digital one. Eleven further patches test this account and it does not hold. Once cardinality is matched, a palette as chromatic as the cube attacks equally well. Cardinality itself shows no trend from three inks to thirty-two. Palettes matched on cardinality, lightness and chroma, and differing only in hue placement, span $0.035$ to $0.136$. A physical evaluation with printed decals did not detect transfer; it bounds the transferred rate at $0.133$, which does not exclude the simulated value of $0.121$.