|Emotion Detection Using Project Oxford|
|Written by Lucy Black|
|Saturday, 14 November 2015|
The Project Oxford team has released a new demo that uses its machine learning Face detection API. This one recognizes emotional states in photographs of people and seems to be impressively sensitive.
Project Oxford is Microsoft's set of artificial intelligence APIs that give developers the opportunity to take advantage of machine learning without having to do the training process for themselves as it has already been done.
In the case of the new Emotional Recognition tool, it has been trained to recognise eight core emotional states – anger, contempt, fear, disgust, happiness, neutral, sadness or surprise – which, according to the Project Oxford team are understood to be cross-culturally and universally communicated with particular facial expressions.
To try out the demo you can upload a JPEG, GIF, PNG, or BMP under 4MB in size or provide the url of an image. Microsoft advises that a near-frontal or full-frontal facial picture will yield the best result. An image of multiple people can be used, assuming it has no more than 64 faces total. The Emotion Tool will pick out faces in the photo and for each of them return the confidence levels for each of the eight emotions.
For the purposes of the demo Microsoft has supplied three sample pics but, as is often the case when people know their photos are being taken, these show happy faces.
To try out the tool we looked for photos that would reveal other emotions and chose as the subject Prince Charles who as a member of the British Royal family is frequently photographed both formally and candidly.
The universally recognized expression of Contempt is to have one side of the mouth curl upward. It is the dominant emotional recognised here with 0.88 confidence, there is also an element of happiness and "wry amusement" might best describe his expression.
The confidence level for Happiness in the official portrait photo above is 0.49 with 0.48 attributed to Neutral. Compare this to 0.9997 in the photo below and 0.96 neutral in the final photo.
This demo, and others based on the Face SDK which we've previously reported on (see How Old - Fun, Wrong, Potentially Risky? and Twin Detection Using AI), isn't just there for our amusement. As with the other Project Oxford APIs, the Face SDK is provided for other developers to and can be downloaded, including documentation, samples, and REST wrappers.
or email your comment to: firstname.lastname@example.org
|Last Updated ( Saturday, 14 November 2015 )|