The field of temporal image forensics is the science of exploiting age-dependent traces introduced by the image acquisition pipeline to approximate the age of a digital image relative to images from the same device. This task can be viewed as a classification problem, where the classes are defined by the temporal resolution of the considered age traces and the available images. It has already been shown that applying a conventional deep neural network (i.e., an image classifier) to this classification problem leads to unreliable predictions. A main reason for this is the content bias inherent in the data. Models from the field of image steganalysis usually include a preprocessing layer, which increases the signal-to-noise ratio (suppresses image content) by generating high-pass filter residuals. Similar to the stego signal, age traces (in-field sensor defects) are also high-frequency image components ‘hidden’ in an image. Thus, are steganalysis models suitable for temporal image forensics? Multiple experiments are conducted to investigate this question. In principle, when comparing steganalysis models with generic image classifiers, steganalysis models do not exhibit a better ability to detect weak age traces. Although image content is suppressed by the preprocessing layer, the models are still prone to content bias.

In the context of temporal image forensics, it is not evident that a neural network, trained on images from different time-slots (classes), exploits solely image age related features. Usually, images taken in close temporal proximity (e.g., belonging to the same age class) share some common content properties. Such content bias can be exploited by a neural network. In this work, a novel approach is proposed that evaluates the influence of image content. This approach is verified using synthetic images (where content bias can be ruled out) with an age signal embedded. Based on the proposed approach, it is shown that a deep learning approach proposed in the context of age classification is most likely highly dependent on the image content. As a possible countermeasure, two different models from the field of image steganalysis, along with three different preprocessing techniques to increase the signal-to-noise ratio (age signal to image content), are evaluated using the proposed method.

Image sensor defects that develop in field over a camera’s lifetime are at the core of temporal image forensics, as by knowing their onset time a temporal order can be assigned among pieces of evidence. In this context, only defects that have developed within the time interval of the available data set are relevant. The available methods for defect detection, based on regular scene images, aim to identify all present defects (e.g., to conceal them). In this paper, we introduce two novel defect detection techniques. Because of their properties, these methods only detect defects relevant for image age approximation. This is important since defects that do not provide additional age information can negatively affect the process of image age approximation.