Ever wondered if you have a doppleganger on this planet? Perhaps a clone? Just a look-alike, or maybe even a real twin?
While science takes the DNA route to resolve these issues, online users can log on to twinsornot.net, a site launched by technology company Microsoft Corp. on Friday, and upload photos to have these questions answered.
Microsoft insists that it does not retain the pictures you upload, making it more of a fun exercise without accompanying privacy concerns.
A user needs to upload two photos to assess how similar the people in these photos are, giving a score from 0 to 100. On his official blog, Mat (Mateus) Velloso, senior software development engineer at Microsoft, said he used the Face API (application programming interface) in Project Oxford — a platform of intelligent services — in just four hours. “…and yes, 4 hours means I haven’t had time to test it with most devices, improve the UI experience or even test so do expect bugs…Yes, it is that simple…”
Face recognition typically provides the functionalities of automatically identifying or verifying a person from a selection of detected faces. It is widely used in security systems, celebrity recognition and photo tagging applications.
Microsoft’s Face API, for instance, can detect up to 64 human faces in an image. Face detection can be done by uploading an entire picture (.jpeg file) or specifying the online address (URL) of an existing image (.jpeg) on the web. The detected faces are returned with rectangles (left, top, width and height) indicating the location of faces in the image in pixels.
Optionally, face detection can also extract a series of face related attributes from each face such as pose, gender and age. Users can also use the Face API to look for similar-looking faces of a specified face from a list of candidate faces. The list has a limit of 100 faces.
Microsoft has a host of such APIs on gallery.azureml.net. For instance, there’s a Face API— a cloud-based face algorithms to detect and recognize faces in images. There’s a Computer Vision API, which allows image processing algorithms (method of solving a problem) to return information based on visual content and generate a thumbnail. Speech APIs use algorithms to process spoken language while ‘Recommendations’ help customers discover items on website catalogues.
Developers can use Microsoft’s Project Oxford to download the software development kits (SDKs) and develop such products.
According to Velloso, twinsornot.net got 74,000 users “in a few hours”, mostly from countries like the US, Brazil and Hungary. But he acknowledged that mobile users were “getting a lot of failed requests”, which he attributed to his “zero testing on mobile devices”.
It was only in late April that Microsoft launched a ‘How Old Do I Look?’ website (how-old.net) to mark its Build developer conference in the US. In a 4 May post, Corom Thompson and Santosh Balasubramanian, engineers in the Information Management and Machine Learning unit of Microsoft, explained that the idea of the site was to “showcase how developers can easily and quickly build intelligent applications using Azure (Microsoft’s cloud platform) services”. The site lets users upload a picture and have the API predict the age and gender of any faces recognized in that picture.
It took a couple of developers just a day to put this whole solution together — from the web page to the Machine Learning APIs to the real time streaming analytics and real time BI (business intelligence), accordinG to Thompson and Balasubramanian. The key components of this application comprise extracting the gender and age of the people in these pictures, obtaining real-time insights on the data extracted above, and creating real-time dashboards to view the above results.
The engineers said in their post that “…within hours, over 210,000 images had been submitted and we had 35,000 users from all over the world”. Over half the pictures analyzed were of people uploading their own images.
Facial recognition is not new, and technology companies are very interested in this for obvious reasons–knowing who their users are and using that analysis to build better and more marketeable products around it.
On 20 June, 2014, a University of Central Florida (UCF) research team said it has developed a facial recognition tool that promises to be useful in rapidly matching pictures of children with their biological parents, and potentially identifying photos of missing children as they age.
Graduate student Afshin Dehfghan and a team from UCF’s Center for Research in Computer Vision started the project with more than 10,000 online images of celebrities, politicians and their children.
Companies such as Twitter Inc, Facebook Inc and International Business Machines Corp (IBM) have also invested in startups to boost deep learning research. And many companies deploy sophisticated systems for the application of border control and biometric identification. However, these systems are sensitive to various factors such as lighting, expression and aging, which substantially deteriorate their performance in recognizing people in such unconstrained settings.
Deep learning is a new area of Machine Learning. It inspects large data sets that include human faces and tries to develop a high-level abstraction of a human face by looking for recurring patterns such as cheeks and eyebrows. An example of a Deep Learning model is the Convolutional Neural Nets, or ConvNets. Deep Learning, thus, takes Machine Learning closer to Artificial Intelligence.
Take the case of Facebook’s DeepFace that uses technology designed by an Israeli startup called face.com, a company that Facebook acquired in 2013.
Facebook’s facial recognition research project, DeepFace, now works almost like the human brain. DeepFace can look at two photos, and irrespective of lighting or angle, can say with 97.35% accuracy whether the photos contain the same face. Humans can perform the same task with 97.53% accuracy (benchmarked on the widely used Labeled Faces in the Wild (LFW) dataset). The software, developed by the Facebook AI (artificial intelligence) research group, uses a deep learning neural network–software that simulates how real neurons work.
Google does not want to be left behind. According to a paper published in arvix.org, a publication owned and operated by Cornell University, in March, three researchers from Google Inc. have also developed a similar deep neural net architecture and learning method that also uses a facial alignment system based on explicit 3D modeling of faces. Google claims its FaceNet system recognises the right person 99.96% of the time on LFW.
