The appropriately-named leading social media platform Facebook is developing facial recognition technology with “near-human accuracy.”
Dubbed DeepFace, the software maps three-dimensional facial features and creates colorless models to zero-in on specific, personal facial characteristics – a method of accuracy equal to 97.25 percent match-correction, comparatively just below a human being’s 97.5 percent.
“We present a system [DeepFace] that has closed the majority of the remaining gap in the most popular benchmark in unconstrained face recognition, and is now at the brink of human level accuracy,” researchers said in a Facebook developer API Group report. “It is trained on a large dataset of faces acquired from a population vastly different than the one used to construct the evaluation benchmarks, and it is able to outperform existing systems with only very minimal adaption.”
The software breaks down facial recognition into four separate steps: detect, align, represent and classify.
“We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network,” Facebook said in the post announcing the DeepFace research.
The network calculates more than 120 million parameters over several layers, and draws from four million facial images belonging to more than 4,000 identities – each on average having more than a thousand samples.
Facebook has yet to announce when or how the technology might be implemented into the platform at large (assuming parts of it haven’t been tested on the back end already), but the company is expected to present its findings at a tech conference this summer.