Facebook have updated their social network to make it easier for blind people to understand the sort of photo content which is coming up on their news feed. The new system uses AI to describe an image aloud…
The world’s largest Social Network has taken a step towards making their services more accessible for the blind and partially sighted by improving their audio descriptive technology. In the past Facebook was unable to accurately describe specific images. The new feature uses screen reader technology to describe in detail what is on screen.
Visually impaired exclusion from Social Media
A recent paper written by the Facebook Research department explored the motivations, challenges and interactions experienced by blind people when engaging with visual content on Social Media. It was concluded that blind people often feel left out of this style of conversation.
An advancement on past technology
Image recognition technology is not a new thing for Facebook, having previously been used to give suggestions about which friends should be tagged within a particular image. More recently however, the Social Network have invested in the tech to begin to understand and recognise specific objects and features within an image.
According to reports, Facebook have been working on an audio-descriptive service for images, which is tailored to the visually impaired, for 10 months now. The new system will be able to identify and describe an image by detecting its contents. It will then use a computer-generated voice to convey this to the user.
Facebook have used mountains of previous photo content to better train their image recognition technology and have concluded that it is able to recognise around 100 specific feature concepts that are present in over half of photos on the network, such as smiles, babies and selfies, with an accuracy rate of at least 80%. They plan on continuing to develop the recognition capabilities going forward in order to continually improve network experience for the visually impaired.
Photo Source: By Elekes Andor [CC BY SA 3.0], via Wikimedia Commons
