Can the government or police really use AI to deblur your photos and identify your face?
To investigate this question, I conducted a series of tests by processing several blurred photos through two online deblurring websites: Fotor and Remini.
The primary focus of this study is to assess the effectiveness of these tools in restoring facial clarity, allowing individuals to be recognized even in the presence of intentionally applied blur.
I collected a set of stock photos featuring a diverse range of models and applied three levels of blur to each one. The selection of models aimed to cover a spectrum of facial orientations, lighting conditions, and expressions, providing a comprehensive evaluation scenario.
The intentional blurring of facial features ranged from mild to severe, simulating real-world scenarios where image quality may be compromised. The chosen images also varied in lighting conditions, capturing scenarios with different ambient light and shadows. Additionally, facial orientations, including both direct and oblique angles, were considered to assess the adaptability of the tools in various photographic situations.
Selected AI Deblurring Websites
Fotor
Known for its user-friendly interface and versatile image editing capabilities, Fotor incorporates AI-driven deblurring algorithms. The platform claims to restore clarity and detail in images affected by blurring, making it an ideal candidate for this comparative analysis.
Remini
Remini stands out for its advanced image enhancement features, including deblurring functionality. Leveraging sophisticated AI algorithms, Remini promises to resurrect fine details in images, potentially offering a solution for recognizing faces obscured by intentional blurring.
Objective
The primary goal of this study is to evaluate and compare the effectiveness of Fotor and Remini in deblurring intentionally blurred facial images. Through an examination of the restored images, we aim to determine the extent to which these tools can successfully reveal facial features and enable recognition, even in challenging conditions.
SAMPLE PHOTO 1
This should be the easiest photo for AI to deblur. The subject’s face is clearly visible, well lit and facing the camera.
SAMPLE PHOTO 2
I wanted to see how the AI would deal with her having some hair covering her face. Seems like AI had no problems with the hair but trouble recreating her face.
SAMPLE PHOTO 3
For this photo I wanted to test how the AI would handle the different shadows on her face. For the AI to work best it needs a clear image of your face, so this was an interesting test to see how it would handle the harsh lighting conditions.
SAMPLE PHOTO 4
We can see that AI works best when the subject is well lit and facing directly to the camera but how would it handle facial piercing? We can see that AI completely missed it and didn’t include any piercings at all.
Although the Remini software is very impressive for its ability to recreate the face, even with a strong blur. It may not look like the original person but the tech is pretty impressive.
SAMPLE PHOTO 5
I wanted to try and push the abilities of the AI but showing it faces that are turned to the side. This makes it harder to “recreate” the face as it has to think about placement of eyes, nose, mouth etc. As you can see in these photos, both AI programs failed pretty badly.
SAMPLE PHOTO 6
The AI struggles to deblur the image when the face is turned away so I wanted to push it even farther to see what would happen when it can only see half the face.
We can see the AI really struggles with only half the face. This is good to know when you’re posting photos of yourself online. Either cover part of your face with your hair or turn away from the camera.
SAMPLE PHOTO 7
The AI struggles to deblur the image when the face is turned away so I wanted to push it even farther to see what would happen when it can only see half the face.
We can see the AI really struggles with only half the face. This is good to know when you’re posting photos of yourself online. Either cover part of your face with your hair or turn away from the camera.
The test results suggest that AI performs admirably when the subject is well-lit, facing the camera, and experiencing slight to medium blurring. In these conditions, the AI demonstrates an ability to recreate a face that closely approximates the original, unblurred image.
However, challenges arise when the subject’s face is turned away from the camera, is affected by harsh shadows, or experiences a strong blur. In such instances, the AI struggles to effectively remove the blur and generate a realistic-looking face.
In conclusion, for those seeking to safeguard their identity online, it is advisable to employ strategies that obscure facial features. This may include facing away from the camera, utilizing shadows, or applying a significant blur. Given the continuous advancement of AI deblurring technologies, taking proactive measures now to conceal one’s face can be instrumental in protecting identity in the evolving digital landscape. As these AI technologies evolve, the more precautions taken today, the better the chances of maintaining online anonymity in the future.