| Previous work | Go back to the index | Templates |
The new face recognition algorithm performs a very computationally-demanding set of operations over the photo in order to detect all the faces present in the image.
The two main sources of information for this detection are:
First the original photo is loaded and resized such that its largest size is 800px. This is done to decrease the computational load, while making sure that the smallest faces that appear in natural scenes are big enough to produce a significant peak over the noise when they are correlated with a template face.
Then both the photo and the images are converted to the Lab color space and the photo is correlated with all the templates. Afterwards, the color of the photo is compared to the skin color, calculated from the templates, the results are combined and the initial guesses are taken. Finally, the initial guesses are clustered and refined so that the final guesses so more accurately the faces positions.
| Previous work | Go back to the index | Templates |