Crafting and Learning Features from Faces

Identification of similar faces to a face image

The goal of this assignment is to explore more advanced techniques for constructing features that better describe objects of interest and perform a few tasks using these features.

More practical, this assignment is broken down into four main parts:

  • construct your own dataset of faces;
  • build some feature representations using handcrafted (HOG) and non-handcrafted techniques (PCA and transfer learning);
  • use and compare the feature representations in context of classification of faces and assessing similarity between faces;

We picked images from four different celebrities A, B, C, and D. C looks like A and D looks like B. The classifiers are trained on 40 images of A and B then predict labels on 40 images of A, B, C, D. The identification step provides k similar faces from training faces on a randomly picked testing face.

This code can be found here.

Feiyang Tang
Feiyang Tang
Ph.D. Candidate in Machine Learning

Data Enthusiast, ENFJ-T. Travelling, hiking and crime series lover. Multilingual.