The goal of this project is to perform a face detection on images, using classical algorithms.

Training

A sliding window algorithm is applied, combined with a SVC (Support Vector Classifier) and HOG (Histogram of Oriented Gradients) features. The best hyper-parameters, classification algorithm and feature descriptors were obtained using a cross-validation.

Testing

The following results were obtained on the test images:

  • Precision: 99.07 %
  • Recall : 94.02 %
  • F1-score : 96.48 %
  • AUC : 93.46 %