Now that I have finally returned to Madrid after MWC, it’s time to update some draft posts. A couple of days ago many people was amazed when I mentioned the base algorithm for face detection is from the year 2001. That’s it, more than 10 years since the work by Paul Viola and Michael Jones. This algorithm has matured so much and now we can find it as an example in many programming languages, and almost all the cameras and smartphones with cameras have it built to detect faces in real time.
If anyone is interested in seeing how the algorithm of Viola-Jones, the wikipedia entry (link) is a work of art.
Now, the basis of this algorithm is that the faces must be “looking to the camera” for the image can be processed correctly. This is what is being sought are the shadows and imaginary lines that are created between the eyes and the nose. This leaves us out of the results of detection, faces that are not straight and looking to the camera.
A few days ago Sachin Farfade and Mohammad Saberian in the laboratories of Yahoo in California, in addition to Li-Jia Li at Stanford University have shown a new approach to face detection, which includes also the faces that are not straight. The name of this algorithm is “Deep den Face Detector” and stands out among other similar speed and “simplicity”. That Yes, the purification and refining of the same is being performed with Machine Learning algorithms quite complex.
If someone wants to stick a fun time reading scientific articles in the official paper explained how one gets to this approach.
Greetings Finally @ Madrid
Multi-view Face Detection Using Deep Convolutional Neural Networks, http://arxiv.org/abs/1502.02766
Viola-Jones face detection, http://en.wikipedia.org/wiki/Viola%E2%80%93Jones_object_detection_framework