Software detects deepfakes by analyzing individual blood flow patterns

Intel recently unveiled a new program called FakeCatcher that tackles the rising problem of deepfakes. These fabricated videos pose a threat to trust and can spread false information. Even experts sometimes struggle to identify a deepfake video accurately. However, Intel claims to have found a solution by examining the blood flow under a person’s skin.

Traditional approaches fail to catch deepfakes since they focus solely on identifying manipulated elements, which can be manipulated further to evade detection. FakeCatcher takes a different approach.

It recognizes the genuine aspects within authentic videos, exploring what makes them real. To achieve this, the software analyzes the person’s heart, specifically the changes in blood color caused by pumping.

Ilke Demir, a senior research scientist at Intel, elaborates on the technique. As blood circulation alters the color of a person’s veins, a process known as photoflexmography (PPG), FakeCatcher detects these minuscule shifts.

By capturing PPG signals from numerous areas on the face and converting them into PPG maps, FakeCatcher employs deep learning to classify videos as either real or fake.

In addition to scrutinizing blood flow, FakeCatcher considers other telltale signs, such as the eyes. According to Demir, real humans tend to focus on a specific point, while deepfakes exhibit “googly eyes.” This distinctive clue further aids in the detection process.

Intel claims that FakeCatcher achieves a 96 percent accuracy rate and can identify various types of deepfakes. Nevertheless, when tested by a BBC journalist, the software’s performance was less than stellar. During the test, FakeCatcher incorrectly labeled genuine videos of President Biden and Donald Trump as deepfakes.

However, Intel acknowledges that the program still does not analyze audio, and this improvement could enhance its efficiency and accuracy.

Trendingthe week