Neural Network Provides Tips on How to Take the Perfect Selfie

Andrej Karpathy, a researcher at Stanford University's Computer Vision Lab, trained a neural network computer to learn what makes the perfect selfie by analysing millions of them and identifying specific traits from the ones that were most popular.

Karpathy used a Convolutional Neural Network (ConvNet), a powerful processing system that is more akin to a human brain than conventional computer, 'ibtimes.co.uk' reported.

An initial five million images tagged #selfie from the Internet were fed into the system, then the ConvNet narrowed it down to two million applying parameters such as number of likes it received, number of followers and number of tags.

The system found that most popular were female selfies.

In all the top ranked images, the face always occupied about one-third of the image, was slightly tilted, and was positioned in the centre and at the top.

The research also suggested women should cut off foreheads in the snaps, show their long hair and use filters.

The top images also had a frequent occurrence of oversaturated lighting, which often made the face look much more uniform and faded out.

Selfie taken in low lighting and group shots were ranked very low by the ConvNet.

Karpathy has also built a Twitter tool that will analyse any image you tweet to it and give you feedback.