Face recognition algorithm uncovers ‘Photoshopped’ images
25 May 2016 15:39 GMT

Once the preserve of celebrity magazines, retouched facial images are being the norm for millions as user-friendly image-enhancing tools make it easier to mask blemishes. But as these images are increasingly used for tasks such as uploading ID document images, is there is a threat that beautification could confuse face recognition algorithms- and therefore security?

The trend has led a team of researchers to demonstrate the effect of digital alterations on the performance of automatic face recognition, and crucially to also introduce an algorithm that can classify face images as original or retouched.

In the report “Detecting Facial Retouching Using Supervised Deep Learning”, a team of IEEE specialists, including Aparna Bharati, Richa Singh, Mayank Vatsa and biometrics expert Kevin W. Bowyer, note that the use of retouched images has become “common practice” in social media.

The team notes that the retouching process includes altering facial features in various ways: “airbrush out” pimples, age spots and wrinkles, make the whites of the eyes whiter, make the teeth whiter, change shape of nose and eyebrows, remove wrinkles, add texture, adjust skin tone, and make the face slimmer.

The team compared two face image databases with unaltered and retouched images, finding that a retouched image is matched with its original image or an unaltered gallery image, the identification performance is “considerably degraded, with a drop in matching accuracy of up to 25%.”

As a solution, a “Boltzmann machine algorithm” has been proposed.  It uses facial parts to learn discriminative features to classify face images as original or retouched. The proposed approach for classifying images as original or retouched yields an accuracy of over 87% on the datasets introduced in this paper and over 99% on three other makeup datasets used by previous researchers.

In terms of real-world and research applications, the uses are many. For instance, the result has implications for face recognition studies that are based on image datasets of celebrities collected from the web.

The algorithim can also be used to identify users who have inadvertently used retouching when supplying their own images for identification documents – before their inclusion degrades the accuracy of facial recognition systems.


Related articles

3D boost for face recognition accuracy

Industry Events

connect:ID 2021 5-6 Oct 21