The latest Amped Authenticate update introduces the new Perspective filter, enabling experts to analyze perspective in images and spot inconsistencies in vanishing points to detect deepfakes and manipulation.
The update also delivers improvements to Video Mode, report generation, and more.
Dear friends, we’re excited to share another important update to Amped Authenticate! This time our developers worked hard to bring you a big new entry in the Geometric Analysis category: the Perspective filter. This new tool allows you to analyze perspective in images. It helps you evaluate vanishing points and geometric consistency to detect deepfakes and image manipulation.
Moreover, they’ve also improved many other things with the software. So let’s dive in and check them all out!
See the New Features in Action
The New Perspective Filter
This is huge! With this new filter, you’ll now be able to check if elements in the picture have a consistent perspective. I am not just talking about individual vanishing points. You can also check the coherence of multiple vanishing points that should lie on the same horizon. But let’s go one step at a time, and start from the basics.
The Importance of Geometric Analysis
Geometric analysis is an important asset to image authentication. When you create a fake image, possibly a deepfake, it is not trivial at all to keep shadows, reflections, and perspective technically coherent. Yes, they may look alright and cause no concern at first glance. But can they pass a technical analysis? Often not, as we showed recently in a scientific paper.
The best part of geometric analysis is its robustness to post-processing. Methods based on container, metadata, and pixel analysis all share a fundamental problem: the more you process the image after tampering, the more you conceal traces of intermediate manipulation. If you print-and-photograph a deepfake, for example, you’ll get a new image that will pass every file format and container check. However, it’s not authentic because its content is not genuine.
With geometric analysis, you can bypass this problem. Put simply, issues with shadows, reflections and perspective are not hidden by image and video re-encoding. Can you grasp the power of this?
Authenticate has been featuring the Shadows filter for years now, and we’ve recently introduced the Reflections filter. It follows a very similar mathematical model, as we showed in the mentioned scientific paper linked above. Now it’s time to raise the bar further and move to the analysis of perspective in images.
A Perspective Primer
Perspective is how we perceive depth and spatial relationships in a scene. It’s what makes a two-dimensional image appear three-dimensional.
In the analysis of perspective in images, the vanishing point is a crucial concept to consider. This is the point where parallel lines appear to converge in the image as they stretch into the distance. Think of a long road or a hallway: the edges appear to meet at a single point far away. That’s the vanishing point. Most real-world scenes contain one or more of these, depending on the angles and layout of the objects.
We said images can contain multiple vanishing points. Indeed, each direction in the 3D world corresponds to a vanishing point in the image, as shown in the example below.
Closely connected to this concept is the horizon line, where vanishing points for all directions of the ground plane lie on.
A very important rule of perspective is that all lines lying on the ground and on other planes that are parallel to the ground share the same horizon line. This is very useful, indeed, when analyzing images of rooms, cities, etc. where the floor is typically flat. In these cases, objects or buildings lie on it, creating lots of parallel planes, as in the example below.
In the example above, you may argue that the two lower red lines lie on a surface that is perpendicular to the floor, rather than parallel. But if you give it more thought, you’ll conclude that those lines are eligible for inclusion since they are parallel in the real world and their direction lies on planes parallel to the ground.
So, in summary, we know that:
- For a given direction, parallel lines in the real world converge to a single vanishing point in the 2D image.
- Vanishing points obtained from lines lying on planes parallel to the ground all lie on the horizon.
All of the above applies under the assumption that no optical distortion is present. Otherwise, lines in the image are no longer straight and all of the above no longer makes sense. If you want to check the perspective of an image where distortion is present, you first need to remove it, e.g. using Amped FIVE’s Undistort, Correct Fisheye, or Camera Calibration filters.
How the Perspective Filter Works
The Perspective filter is found under the Geometrical Analysis category. If you’re updating from a previous version of Amped Authenticate, you may find the filter disabled at the first run. Just right-click on it and select Enable, then click on Save Settings to replace the default loading settings.
The filter works in two steps:
- In the first instance, you identify up to 5 different vanishing points within the image by selecting groups of lines that you know to be parallel in the real world. Each line group should represent the vanishing point of one direction of the ground. If parallel lines of a single cluster do not converge to a single vanishing point, you’ll be notified and you’ll already know something is not quite right with the image.
- Assuming all selected vanishing points are consistent, the filter also checks that a line intersecting them all exists. As explained before, this is expected to happen when all vanishing points are computed from lines that lie on parallel planes. If no line exists, you’ll be notified and once again, you’ll know something is wrong.
Let’s now have a look at how to use the filter in practice.
Selecting Lines for a Vanishing Point
Each vanishing point is hosted by a tab in the right panel. Just as we did for the Shadows and Reflections filters, you can select wedges instead of simple lines. This approach helps reduce any errors caused by the limited resolution of images.
You start by identifying a group of parallel lines, and by left-clicking on a point in the image at the beginning of the first. The second click will be at the other extreme of the line, and that will set the line direction. By moving the mouse, you’ll then be able to adjust the wedge width, which depends on the uncertainty caused by image resolution and definition. Once you’re happy with the wedge, a third left-click will finalize it. You’ll still be able to move it to adjust its direction, until you click a fourth time to register it in the list of constraints. You’ll see the coordinates of the three points defining the wedge listed in the corresponding vanishing point’s tab. You can cancel the wedge creation at any time using a single right-click.
Remember that you can zoom in and out by using the mouse wheel, and that you can pan across the image by holding the CTRL key and by dragging over the image. This eases out the navigation while drawing wedges.
You can now add more lines in the same tab following the steps above. While you keep adding lines, the filter will search for an intersection among the wedges, which would contain the vanishing point for that direction.
Checking the Consistency of an Individual Vanishing Point
In our example, the floor tiles seem to have a consistent perspective, as indicated by the green text below the tab.
You can visually check the intersection of wedges (or “feasible region”) by clicking on the dedicated button. A new window will appear showing the feasible region in yellow. You can see it’s pretty tight with the constraints we’ve drawn.
We can now consider more lines that we know for sure to be parallel, or work on a new tab. In our example, we are not entirely sure that the kitchen’s furniture is perfectly aligned with the floor tiles. However, we can safely assume they lie on a parallel plane, so we simply add new lines to a new tab.
Also in this case, wedges identify a feasible region.
Let’s now turn to yet another direction, still using the furniture’s lines. Also, in this case, we see that they identify a feasible region. Something, however, has now changed at the bottom right of the filter’s GUI.
Checking the Consistency of Combined Vanishing Points
The result shown above is telling us that, although each individual tab identifies a feasible region, no line also intersects (crosses) all the feasible regions. If the image were real, all vanishing points should lie on the horizon. This is not the case for this image!
To check the overall situation, you can click on the “Show Multiple Feasible Regions” button. We can see that the floor’s vanishing point is indeed positioned well below the remaining feasible regions. Once we introduced the third tab’s lines, it became impossible to find a line that crosses them all.
By clicking on the checkboxes at the bottom, you can identify the minimal configuration that leads to an inconsistent status.
We see that, if we exclude the third set of constraints, the system reverts to a feasible state, as expected.
Tip: Check for Perspective Where There Is Perspective
When choosing directions for your perspective analysis, we recommend using lines that feature some perspective (i.e. they are not parallel in the 2D image). If you use directions in which the perspective is negligible, the vanishing point will be considerably far from the image, thus preventing a view of the feasible region.
Analyzing Perspective in Video Frames
The Perspective filter is a powerful addition to Amped Authenticate for analyzing the perspective in images. It completes the Geometrical Analysis category. Remember that geometric filters can be used on video frames as well! In Authenticate’s Video Mode, just click on the “Send frame to Image mode” button and you’re all set!
Automated Rotation in Authenticate’s Video Mode
Authenticate’s Video Mode GUI now features a new button allowing you to rotate each filter’s output.
When you save a bookmark, the status of the button will be recorded and presented in the report.
Saving Advanced File Info Data to a Log File
When using Advanced File Info in Authenticate’s Video Mode, you’ll now be able to save several tabs’ data to a dedicated log file. That makes it easier to attach such information to your report.
If you want to include the data in your Authenticate’s report, you can copy-paste the info from the panel and paste it into a bookmark’s description. Another option is to create an empty folder and use it as a container for the data.
The information will be printed in your report as shown in the illustration below.
Improved Report Generation to DOCX Format
We’ve discontinued the option to export to the legacy DOC format and added support for the more recent and flexible DOCX format. In addition to having a much neater and more customizable report, you’ll now be able to export to this format even if Microsoft Word is not installed on your computer.
Other Improvements and Bug Fixes
We’ve made other improvements to the software, including:
- Improved handling of the case where the user attempts to load an audio-only file.
- Annotate’s Magnify tool will now show the set Contrast and Brightness values in the report.
- We’ve updated the PortAudio library to version 19.7.0
- We’ve removed the Slack values from the Shadows and Reflections filter and improved the filter GUI layout.
- We’ve improved the behavior of the “Expand all / Collapse All” feature in the filter’s panel.
- We’ve added keyboard shortcuts for zooming in and out (“CTRL” + “+” and “CTRL” + “-”, respectively.
- We’ve improved the GUI of the RIFF and Hex viewers.
- We’ve improved the PRNU Source Identification filter so that it displays the detected rotation and resize factors when a match is found.
- We’ve fixed an issue with the CTU filter that caused wrong motion vectors to be displayed sporadically.
Don’t Delay – Update Today
The new Perspective filter marks a major step forward in image perspective analysis for image authentication. It enables investigators to assess vanishing points and geometric consistency.
If you have an active support plan, you can update straight away by going into the menu About > Check for Updates within Amped Authenticate. If you need to renew your SMS plan, please contact us or one of our authorized distributors. And remember that you can always manage your license and requests from the Amped Support Portal.