Although there are some very clever tools which can identify an image by detecting and recognising shapes, their powers are still limited.
Today’s invention is a way to communicate to a machine the essence of what you are drawing. Imagine sketching an eye on a touch screen. When you’ve done this, you can popup a dialogue box and enter the word ‘eye.’ Change the colour of the pen being used to the inputting colour, draw something else and type in its name. Continue doing this until the entire drawing is labeled.
The machine will be equipped with information about the spatial arrangements of visual elements (so that a car might be expected to have two wheels when viewed from the side and a frontal face will have two eyes a mouth and a nose).
This will enable a machine to infer eg that the thing just drawn above two eyes may be a hat (and ask the artist to confirm this). It will also allow eg automated animation in the sense that the positions of something marked ‘face’ and something marked ‘neck’ can never be farther apart than a certain critical distance, even when a dance is in progress.
Elements can be remembered as present even when one occludes another (just as in the real world) and they can also be automatically replaced by exemplars extracted from a database, making eg crazy photofit composite movies from combinations of photographic facial features.
This could be extended to allow tracing and labelling of elements in an existing image.
