Machine learning features
In the absence of text descriptions provided by content authors, screen readers are turning to machine learning. Screen readers have used a form of machine learning called Optical Character Recognition (OCR) for some time. OCR examines a graphical representation of a document (like a scan of a paper document) for text content, then converts it into actual text that screen reader users can read. More recently screen readers have begun introducing image recognition capabilities. VoiceOver on iOS uses image recognition to describe images and identify common objects like icons and buttons; Jaws Picture Smart describes images and controls sourced from files, websites and the clipboard; NVDA recognises objects and with addons can be further extended.
Content authored text descriptions are still needed
If you're thinking to yourself that providing text descriptions is no longer something you need to bother with though - think again.Here are the Picture Smart image recognition results for this image of The Metamorphosis of Narcissus by Salvador Dali:
- Caption is a painting of a person
- These tags describe the photo: Art, cartoon, drawing, text
- This tag probably describes the photo: Illustration
- This tag possibly describes the photo: Sketch
To borrow from Douglas Adams, it's almost, but not quite entirely nothing like it.