Data mining applied to face detection
OTHER License
Data mining applied to face detection
Do you want to see a presentation that's a summary of the work? Checkout the project website: https://galeone.github.io/face-miner/
Face Miner is based on the paper of Wen-Kwang Tsao [[1]]. To understand Face Miner, you have to read it before. You can find a copy of the paper into the docs/
folder.
The aim of Face Miner is to build a classifier capable of detecting faces in images.
To reach this aim, the paper's authors built three different classifiers, using a simple-to-complex and coarse-to-fine approach.
In Face Miner, a critical discussion of the paper has been done. Therefore Face Miner is not a reproduction of the work described in the paper, but a different implementation resulting from this critical discussion.
The original paper was not reproducible due to the usage of private datasets and neither the resulting work was publicly available.
Face Miner is completely reproducible and the resulting work is avaiable in this repo.
The main differences between the original paper and Face Miner are:
Load the project into Qt Creator. Build the subproject MAFIA before and than build Face Miner.
The theory behind face miner, and thus the critical discussion can be found in the docs/pdf
folder.
At the moment the complete theorical discussion is in Italian only.
The results/
folder contains some example of the results of Face Miner
The performance of Face Miner are bad in term of speed, especially if compared to the speed achieved by the state of art algorithm used for the face detection task: the Viola & Jones algorithm.
The other main difference, is the region detected. In Face Miner the ROI is smaller wrt the ROI detected with Viola & Jones. This can be good in applications of facial recognition, where you have to reduce noisy parts to focus on the face only.
Below you can see the different ROI detected with Face Miner and Viola & Jones.
The benchmark has been done on the Yale Face Database B [[2]]
Face Miner is licensed under Mozilla Public License version 2.0