Abstract Many biometry methods extract task-relevant information from images. In forensic applications, these images may stem from uncontrolled sources like surveillance cameras in the wild. Such devices oftentimes strongly compress the data, which can significantly complicate biometric tasks. This issue is exacerbated by the emergence of AI compression, which may provide visually appealing images that are of questionable value for biometric identification. The purpose of this work is to investigate potential pitfalls of AI compression. We evaluate six AI compression methods including the recently standardized JPEG AI on the four biometric modalities of irises, fingerprints, fabrics and tattoos. We qualitatively show multiple cases when AI compression achieves misleading results. Tattoos in particular includes misrepresentations of color or shapes at strong compression rates. The quantitative evaluation shows impact on recognition rates when there are few identifying features, such as in low-resolution iris images. Further results show that compressors with MSE loss are prone to omit important image details, and MSE+LPIPS loss may hallucinate features. The findings in this paper aim at raising awareness to these pitfalls, and aiding the development robust biometric algorithms for images in the wild.

In digital forensics, the provenance of data being used as digital evidence in court is frequently challenged. This is often done to raise doubts about whether the possession of some piece of illegal data was intentional. However, there may also be situations where the provenance of data is legitimately questioned, such as in the case of second-hand devices that were not well sanitized. This may also lead to the disclosure of personal or corporate data. While poor sanitization practices have already been observed for second-hand hard disk drives, this has not yet been reported for new storage devices based on flash technology. Based on insights into the second-hand chip market in certain countries, we report on the results of the first large-scale study on the effects of chip reuse for USB flash drives. We provide clear evidence of poor sanitization practices for USB flash drives from the low-cost market that were sold as new. More specifically, we forensically analyzed 1211 low-cost USB flash drives and were able to recover non-trivial data on a total of 76 devices (6%). Furthermore, we forensically analyzed 435 high-cost USB flash drives on which we could not find any evidence of chip recycling in this market sector.

Audio recordings may provide important evidence in criminal investigations. One such case is the forensic association of a recorded audio to its recording location. For example, a voice message may be the only investigative cue to narrow down the candidate sites for a crime. Up to now, several works provide supervised classification tools for closed-set recording environment identification under relatively clean recording conditions. However, in forensic investigations, the candidate locations are case-specific. Thus, supervised learning techniques are not applicable without retraining a classifier on a sufficient amount of training samples for each case and respective candidate set. In addition, a forensic tool has to deal with audio material from uncontrolled sources with variable properties and quality. In this work, we therefore attempt a major step towards practical forensic application scenarios. We propose a representation learning framework called EnvId, short for environment identification. EnvId avoids case-specific retraining by modeling the task as a few-shot classification problem. We demonstrate that EnvId can handle forensically challenging material. It provides good quality predictions even under unseen signal degradations, out-of-distribution reverberation characteristics or recording position mismatches. Code is available athttps://faui1-gitlab.cs.fau.de/mmsec/few-shot-recording-environment-identification.

High-quality artificially generated images are widely available now and increasingly realistic, posing challenges for image forensics in distinguishing them from real ones. Unfortunately, building a single detector that generalizes well to unseen generators is very difficult, creating the need for diverse cues. In this paper, we show that natural and synthetic images differ in their color statistics, possibly due to the widely used perceptual loss, which is more sensitive to brightness than to chroma differences. Consequently, color statistics offer valuable cues for forensic analysis and the development of robust detectors. Our experiments using simple hand-crafted color functions with a random forest achieve 91% accuracy averaged over all tested Diffusion Models, even with limited training samples.

The classical JPEG compression is a rich source of cues for forensic image analysis. However, this compression standard will in the near future be complemented by a new, highly efficient learning-based compression standard called JPEG-AI. JPEG-AI is fundamentally different from classical JPEG. Hence, its forensic traces can also be expected to be fundamentally different. We argue that there is a pressing need for image forensics research to investigate these traces. In this work, we characterize forensic compression traces of different AI compression algorithms. Our analysis investigates AI compression artifacts in frequency domain and in spatial domain. Both domains exhibit similar artifacts that likely stem from upsampling operations of the decoders. Additionally, we report for one AI codec another artifact in homogeneous regions. We also investigate the artifact detectability in several scenarios including unseen AI compression traces and postprocessing. Here, frequency and autocorrelation features are better on additive noise and classical JPEG post-compression, while RGB features perform better on blurred and downsampled images.

Criminal investigations oftentimes need the identification of license plates of escape vehicles. The vehicles may be recorded by low-quality cameras in the wild. Their license plates may be unreadable for police officers. Recent efforts aim to use machine learning to forensically decipher license plates from such low-quality images. These methods operate near the information-theoretic limit of recognition and hence show quite high error rates. Unfortunately, it is unclear when such prediction errors occur, which makes it difficult to use these methods in practice. In this work, we propose a Bayesian Neural Network to inherently incorporate a reliability measure into the classifier. We additionally propose to integrate multiple estimations with an entropy weight to further improve the reliability. Our experiments show that this uncertainty metric dramatically reduces the number of false predictions while preserving most of the true predictions.

Forensic license plate recognition (FLPR) remains an open challenge in legal contexts such as criminal investigations, where unreadable license plates (LPs) need to be deciphered from highly compressed and/or low resolution footage, e.g., from surveillance cameras. In this work, we propose a side-informed Transformer architecture that embeds knowledge on the input compression level to improve recognition under strong compression. We show the effectiveness of Transformers for license plate recognition (LPR) on a low-quality real-world dataset. We also provide a synthetic dataset that includes strongly degraded, illegible LP images and analyze the impact of knowledge embedding on it. The network outperforms existing FLPR methods and standard state-of-the art image recognition models while requiring less parameters. For the severest degraded images, we can improve recognition by up to 8.9 percent points.1