General Analysis Overview

The following section outlines general analysis information, offering an overview of the key aspects of the study.

Evaluation

This overview summarises the general data relating to the image and the analysis.

Forensic Analysis

The forensic analysis shows how our system has evaluated the image and which indicators this assessment is based on.

Two key aspects are considered:

1. Traces identified within the image

Our system specifically searches for digital traces that typically arise during the creation of an image. These can be grouped into the following categories:

  • Synthetic: indications of content generated by AI
  • Real: typical patterns produced by genuine camera pipelines
  • Compression: traces resulting from heavy image processing or compression
  • Anomaly: unusual or irregular patterns that stand out from the other detected traces

The number of traces reflects how many indicators point in a particular direction.

Important:
If only a small number of traces is detected overall, the analysis is less reliable. This can occur, for example, with heavily compressed or low-quality images, where much of the detail has been lost.

If traces from different categories significantly overlap (e.g. Synthetic and Compression), this may indicate that the image has undergone multiple processing steps (such as repeated compression or redistribution), making individual signals more difficult to distinguish.

2. Uncertainty of the assessment

In addition to the detected traces, the analysis also indicates how confident the model is in its evaluation.

This is visualised in the chart as a line:

A short line indicates high confidence
A long line indicates high uncertainty

Higher uncertainty means that the available traces do not form a clear overall picture.

This can have several causes, such as:

  • the presence of unusual or emerging deepfake techniques
  • image material that deviates significantly from known patterns
  • or an out-of-distribution scenario, where the model encounters data it has not previously learned to interpret

Example:
The system detects many indicators of AI generation but remains uncertain at the same time. In such cases, the results should be interpreted with particular caution.

In short

Our analysis provides not just a result, but also shows:

  • which traces were identified
  • how strong these signals are
  • and how reliable the assessment is

This ensures that the decision remains transparent and verifiable – not a black box.

Heatmaps

Heatmaps are visual aids that highlight where specific irregularities have been detected within an image.

Colours make these indicators easy to interpret:

  • Warm colours (e.g. red, yellow): strong indicators
  • Cool colours (e.g. blue, green): weaker or no indicators

The heatmap is overlaid directly onto the image, making it clear which areas are particularly relevant to the analysis.nd.

Our heatmap types

We distinguish between the following types of heatmaps, each revealing different kinds of traces:

  • Real: highlights areas where typical patterns of a genuine camera have been detected
  • Synthetic: marks areas with indications of AI-generated content
  • Compression: emphasises regions affected by heavy processing or compression
  • Anomaly: identifies unusual or irregular patterns that stand out from other detected traces

How to interpret heatmaps correctly

Heatmaps do not represent a final decision; rather, they visualise the basis of the analysis:

  • They help identify where irregularities occur within the image
  • They show whether indicators are evenly distributed or locally concentrated
  • Overlapping heatmaps may suggest that an image has undergone multiple processing steps

Important:
A single noticeable area does not automatically indicate manipulation or generation. Heatmaps must always be interpreted in relation to the number of detected traces. Only the combination of all signals leads to a reliable assessment.

In short

Heatmaps make it visible where and why the system has identified certain patterns — and, together with the detected traces, ensure a transparent and understandable analysis rather than a black box.

Authenticity Traces

This heatmap highlights areas of the image where indicators of a genuine camera capture have been detected.

  • Warm colours (red/yellow):
    Indicate strong traces of a natural camera pipeline
  • Cool colours (blue/green):
    Show areas with little or no evidence of authenticity

The detected patterns are based on subtle, typically invisible signatures from the camera sensor. These so-called “fingerprints” arise during image capture and are characteristic of real photographs.

Important:
This heatmap is a visual aid and does not represent a standalone decision. It should always be interpreted in the context of the other traces and the overall analysis.

Synthetic Traces

This heatmap highlights areas of the image where indicators of artificially generated content have been detected.

  • Warm colours (red/yellow):
    Indicate strong signs of AI-generated content
  • Cool colours (blue/green):
    Show areas with little or no evidence of synthesis

The analysis is based on subtle statistical patterns that arise during the generation process of AI models. These patterns result from the computational processes of the generators and leave behind characteristic artefacts that differ from natural image structures.

These irregularities are visualised directly within the image through the heatmap.

Important:
This heatmap is a visual aid and does not represent a standalone decision. It should always be interpreted in the context of the other traces and the overall analysis.

Anomaly Traces

This heatmap highlights areas of the image where unusual or irregular patterns have been detected that stand out from the other identified traces.

  • Warm colours (red/yellow):
    Indicate strong deviations from the typical patterns within the image
  • Cool colours (blue/green):
    Show areas with little or no irregularities

Anomalies arise when certain regions or patterns differ from the rest of the image. They point to inconsistencies that may result from post-processing or manipulation, with or without the use of AI.

These deviations are visually emphasised in the heatmap.

Important:
This heatmap is a visual aid and does not represent a standalone decision. It should always be interpreted in the context of the other traces and the overall analysis.

Compression Artefacts

This heatmap highlights areas of the image that have been affected by compression or subsequent processing.

JPEG compression, for example, is a widely used method in which image data is simplified to reduce file size. In this process, pixels are grouped into blocks and similar content is merged, which can lead to the loss or alteration of fine details.

  • Warm colours (red/yellow):
    Indicate strong compression artefacts
  • Cool colours (blue/green):
    Show areas with little or no compression

Although compression can remove certain image features, it also leaves behind its own characteristic artefacts within the image noise – these are what we make visible here.

Interpretation

This heatmap primarily helps to contextualise the other analysis results:

  • If similar regions appear as in the Synthetic heatmap (e.g. “Synthetic Traces”), this may indicate that compression has overlapped with or altered existing traces
  • In such cases, the results should be interpreted with particular care, as individual indicators may be more difficult to distinguish.

Important:
This heatmap is a visual aid and does not represent a standalone decision. It should always be interpreted in the context of the other traces and the overall analysis.