Artificial Intelligence

Facial Recognition Comes of Age

According to Google Scholar, in 2016 alone, 938 face recognition algorithms were published, 34 patents were filed, and dozens of face recognition startups were established. As quickly as they are being founded, these startups are being acquired- the latest deals include Emotient, RealFace (Apple), Looksery, Seene & Obvious Engineering (Snapchat), Viewdle, Moodstocks (Google) etc.

 

The M&A action is not limited to the consumer segment either. Earlier this year, French Aerospace & Defence group Safran negotiated the sale of their identity and security division (Morpho) to global private equity firm, Advent International, the owner of Oberthur Technologies since 2011, in a deal worth some €2.5 billion Euro. The resulting merger of Oberthur & Morpho has resulted in a company with some $3 Billion in annual identify & security sales globally.

 

The applications and solutions enabled by face recognition technologies are many and varied. On the lighter side, in late 2016, Baidu teamed up with KFC in China to develop a facial recognition system to predict what customers would like to order based on their age, gender etc, details here.

 

There is however a more serious and rapidly growing opportunity for face recognition technologies- one that can broadly be described as public security. For example, in May 2017, it was announced that in future, mainland users of UnionPay, China’s sole clearinghouse for bank card transactions, would have to insert their identity cards into ATMs and have their identity verified by facial recognition software to withdraw cash, details from the Financial Times here. According to the article, The move appears designed to target gamblers and middleman who have been flouting withdrawal limits by using multiple ATM cards registered to different customers. In June 2017, it was reported that facial recognition technology was being trialed by several Chinese municipalities to name and shame jaywalkers.

 

While search and social media giants such as Google, Facebook and more recently Tencent and Baidu, have done significant foundational work on improving the underlying face recognition algorithms, their resulting cloud-based image tagging & facial expression categorisation services offer little practical value (note.. Their deployment of these capabilities for internal use – such as face-tagging by Facebook- are a different matter).

 

At the same time, the emergence of real-world-type challenges for face recognition algorithms such as MegaFace has highlighted the ability of relatively unknown startups to demonstrate superior performance to incumbents like Google- Russia’s NTech Lab being a recent case in point. National security concerns are likely to play to the strengths of local startups- providing them with the ideal opportunity to gain a foothold in their country’s public defence. To this end, Chinese startups like MegaVii (with their Face++ offering) and SenseTime have attracted venture capital investments in the $150+ million range. These contrast significantly with US-based startups such as Kairos and FaceFirst, each of which attracted less than $5 million in funding since their inception.

 

In a very real sense, the technology behind facial recognition has become mainstream with low-entry-barriers through its deployment via open source organizations such as OpenCV and OpenFace. However, the key to commercial success lies with access to datasets- the records containing countless millions of faces scanned for passport & identification documents, immigration control etc. Gaining access to these datasets gives facial recognition algorithms unprecedented opportunity to train their underlying neural nets- making inference, the process of checking any given facial image with those contained in one of those databases, a breeze. In this context, Russian startup NTech Labs achieved overnight fame with its ability to instantly search Russian social network VK (the local equivalent of Facebook) for any picture snapped on a smartphone- with greater than 70% success rate. By the same token, Chinese startups with access to local police records of citizens ID images have a significant advantage in terms of training and optimising their algorithms.