Facial Recognition Tech Is Creepy When It Works--And Creepier When It Doesn't

WIRED

For the last few years, police forces around China have invested heavily to build the world's largest video surveillance and facial recognition system, incorporating more than 170 million cameras so far. In a December test of the dragnet in Guiyang, a city of 4.3 million people in southwest China, a BBC reporter was flagged for arrest within seven minutes of police adding his headshot to a facial recognition database. And in the southeast city of Nanchang, Chinese police say that last month they arrested a suspect wanted for "economic crimes" after a facial recognition system spotted him at a pop concert amidst 60,000 other attendees. These types of stories, combined with reports that computer vision recognizes some types of images more accurately than humans, makes it seem like the Panopticon has officially arrived. In the US alone, 117 million Americans, or roughly one in two US adults, have their picture in a law enforcement facial-recognition database.


Eliminating False Positives During Corner Finding by Merging Similar Segments

AAAI Conferences

We present a new corner finding algorithm based on merging like stroke segmentations together in order to eliminate false positive corners. We compare our system to two benchmark corner finders with substantial improvements in both polyline and complex fits. Sketch recognition is an emerging field that utilizes penbased interaction with computers. Handwriting recognition software in the modern operating systems allows users to write naturally, and applications have been created to recognize sketches in domains such as UML diagrams (Hammond & Davis 2002) and family trees (Alvarado & Davis 2004). In an attempt to perform free-sketch recognition, which allows users to draw as they would naturally without training or being trained by the system, certain geometric sketch recognition systems require a shape to be defined by a set of primitives (Hammond & Davis 2007).


Facial Recognition Used by Wales Police Has 90 Percent False Positive Rate

#artificialintelligence

Thousands of attendees of the 2017 Champions League final in Cardiff, Wales were mistakenly identified as potential criminals by facial recognition technology used by local law enforcement. According to the Guardian, the South Wales police scanned the crowd of more than 170,000 people who traveled to the nation's capital for the soccer match between Real Madrid and Juventus. The cameras identified 2,470 people as criminals. Having that many potential lawbreakers in attendance might make sense if the event was, say, a convict convention, but seems pretty high for a soccer match. As it turned out, the cameras were a little overly-aggressive in trying to spot some bad guys.


Metropolitan Police's facial recognition technology 98% inaccurate, figures show

The Independent

Facial recognition software used by the UK's biggest police force has returned false positives in more than 98 per cent of alerts generated, The Independent can reveal, with the country's biometrics regulator calling it "not yet fit for use". The Metropolitan Police's system has produced 104 alerts of which only two were later confirmed to be positive matches, a freedom of information request showed. In its response the force said it did not consider the inaccurate matches "false positives" because alerts were checked a second time after they occurred. Facial recognition technology scans people in a video feed and compares their images to pictures stored in a reference library or watch list. It has been used at large events like the Notting Hill Carnival and a Six Nations Rugby match.


Scalable Angular Discriminative Deep Metric Learning for Face Recognition

arXiv.org Artificial Intelligence

With the development of deep learning, Deep Metric Learning (DML) has achieved great improvements in face recognition. Specifically, the widely used softmax loss in the training process often bring large intra-class variations, and feature normalization is only exploited in the testing process to compute the pair similarities. To bridge the gap, we impose the intra-class cosine similarity between the features and weight vectors in softmax loss larger than a margin in the training step, and extend it from four aspects. First, we explore the effect of a hard sample mining strategy. To alleviate the human labor of adjusting the margin hyper-parameter, a self-adaptive margin updating strategy is proposed. Then, a normalized version is given to take full advantage of the cosine similarity constraint. Furthermore, we enhance the former constraint to force the intra-class cosine similarity larger than the mean inter-class cosine similarity with a margin in the exponential feature projection space. Extensive experiments on Labeled Face in the Wild (LFW), Youtube Faces (YTF) and IARPA Janus Benchmark A (IJB-A) datasets demonstrate that the proposed methods outperform the mainstream DML methods and approach the state-of-the-art performance.