Archive

Tag Archives: facial recognition

Regardless of how you feel about facial recognition technology—is it a big time saver or a total privacy violator or both?—it’s here. Right now, military-grade technology is at work, watching who’s going in and out of buildings around the world. It’s also more easily accessible than ever. One of the most appealing things about the system is that it works with almost any existing security system. You just install the software and start spotting faces.

Advertisements

In modern face recognition, the conventional pipeline consists of four stages: detect => align => represent => 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. This deep network involves more than 120 million parameters using several locally connected layers without weight sharing, rather than the standard convolutional layers. Thus we trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities, where each identity has an average of over a thousand samples. The learned representations coupling the accurate model-based alignment with the large facial database generalize remarkably well to faces in unconstrained environments, even with a simple classifier. Our method reaches an accuracy of 97.25% on the Labeled Faces in the Wild (LFW) dataset, reducing the error of the current state of the art by more than 25%, closely approaching human-level performance.

DeepFace: Closing the Gap to Human-Level Performance in Face Verification – or, in other words, YIKES. Via O’Reilly Radar.

(See also HuffPo’s coverage, for one.)

Dubbed the Marauder’s Map after the magical map used by Harry Potter, the system takes security camera footage and analyses it using an algorithm that combines facial recognition, colour matching of clothing, and a person’s expected position based on their last known location. The main challenge in designing the map, says Shoou-I Yu, a graduate student at Carnegie Mellon University in Pittsburgh, Pennsylvania, was finding and following individuals in complex indoor environments where walls and furniture can block the cameras. He and his colleagues found a solution by combining several tracking techniques. For example, an individual whose clothes are the same colour, and whose facial features correspond to those of the person who appears just a few frames back in the footage suggests that the two are a match.

Previously, I focused on sets of images, for example taking multiple images from video or combining images from different angles, with one camera looking from the left, one from the right and so on,” he says. “You combine all these images and produce an image set, a composite representation of the face. Image sets provide more accuracy and are applicable in more realistic scenarios”. He says there are many sources from where researchers can get facial image sets of a person, including YouTube videos, Google Images, multiple surveillance cameras and personal photo albums. “After sampling different sets of images from different cameras, the next step was to go in different wave lengths,” he says. “Rather than sampling different poses from different cameras, let’s go into the other dimension which is the wave length or the frequency domain, viewing the images in different colours. That’s what we’re doing now with satellite technology.

Extracted from this Affectiva paper about “Crowdsourced Data Collection of Facial Responses”:

“The internet provides the ability to crowd-source lots of useful information [17]. People are willing to engage and share visual images from their webcams [21] and these can be used for training automatic algorithms for learning. Inspired by these approaches, we capture videos of natural engagement with media online and show that this can be elicited without payment, providing motivation for the viewers by combining the experiment with popular and engaging media shown during recent Super Bowl television coverage.”