Last May, the Redmond giant How-Old.net, a portal that lets us evaluate the ability of its Microsoft Azure algorithms to estimate a person's age. Now we come across a component of the so-called "Proyecto Oxford", which aims to recognize the emotions of all the people that appear in a photograph. The technology involved is undoubtedly experimental, but its results are quite accurate…
Machine learning, artificial intelligence, neural networks… the big companies are investing a lot of time and money in these technologies. The accumulated advances are supposed to enable the development of more intelligent services and devices, with the ability to adapt completely to each user's demands. Of course, the process is significantly slow, and all the elements involved tend to make serious mistakes, but time is on their side, and with each new generation the results are much more accurate than before.
Emotion Recognition on Proyecto Oxford
Today we come across one of the modules available in Proyecto Oxford from Microsoft, focused on emotion recognition. The official page does not provide any technical details about how the module works, although it highlights that the numbers are generated by the API known as "Emotion", when it applies machine learning techniques. The system can recognize up to a maximum of 64 faces in a single image, and if you want to try it with a personal photo, remember its size must be less than four megabytes, with a minimum resolution of 36 by 36 pixels. Supported formats include JPG, BMP, PNG and GIF (only its first frame).
Microsoft explains it from the start: The technology linked to the emotion recognition module is experimental, and many absurd results are expected. In any case, in our tests it didn't do badly at all. The numbers are divided into eight categories (surprise, fear, happiness, sadness, anger, contempt, disgust and neutral), and a perfect recognition equals 1, but the most common thing is to find several decimals behind each value. Microsoft's plan is for developers to use free of charge this and other modules that are part of Proyecto Oxford in their projects, including facial recognition, speech to text (and vice versa) and text detection in images, among other things.
