Landscape More Secure Than Portrait? Zooming Into the Directionality of Digital Images With Security Implications

TL;DR

This paper identifies and systematizes causes of directionality at several stages of a typical acquisition pipeline, measures their effect, and demonstrates for three selected security applications how the performance of state-of-the-art methods can be improved by properly accounting for directionality.

Abstract

The orientation in which a source image is captured can affect the resulting security in downstream applications. One reason for this is that many state-of-the-art methods in media security assume that image statistics are similar in the horizontal and vertical directions, allowing them to reduce the number of features (or trainable weights) by merging coefficients. We show that this artificial symmetrization tends to suppress important properties of natural images and common processing operations, causing a loss of performance. We also observe the opposite problem, where unaddressed directionality causes learning-based methods to overfit to a single orientation. These are vulnerable to manipulation if an adversary chooses inputs with the less common orientation. This paper takes a comprehensive approach, identifies and systematizes causes of directionality at several stages of a typical acquisition pipeline, measures their effect, and demonstrates for three selected security applications (steganalysis, forensic source identification, and the detection of synthetic images) how the performance of state-of-the-art methods can be improved by properly accounting for directionality.

Figures (17)

• Figure 1: Directionality matters across security applications. The figure summarizes the performance loss of selected state-of-the-art classification tasks when the (square-shaped) test images are rotated by 90 degrees. Note the different scales.
• Figure 2: The power spectra (center panels) reveal that the right image has low energy in the high horizontal frequencies. (Motion blur has been applied for illustration.)
• Figure 3: The steerable pyramids directionality score compares the horizontal and vertical energy on two scales.
• Figure 4: Causes of directionality are present in several stages of the image acquisition.
• Figure 5: The relative distribution of horizontal and vertical scene content is skewed towards images with more horizontal edges. On the sides are example images from the three databases with very high absolute directionality scores.