Learning to Find Discriminative Null Space for Face Recognition

Face recognition is when we use classification to identifying people through images of their face. In simple words, given an image of the face of an unknown person, we can use face recognition to match their face to a known person, usually from a gallery of images.

In this project, we’ll solve a problem of face recognition by finding a discriminative subspace, more formally known as the Null Foley-Sammon Transform (NFST). NFST is a distance-metric learning approach of finding a discriminative null space. The intention behind this approach is to find a subspace where the discrimination among different classes increases and data points belonging to the same class are close together, ideally at a single point.

We’ll work on a real datasetGary B. Huang, Manu Ramesh, Tamara Berg, and Erik Learned-Miller. University of Massachusetts, Amherst, Technical Report 07-49, October, 2007. called “Labeled Faces in the Wild” (LFW). The dataset consists of a gallery of JPEG images of famous people. Each picture, collected from the internet, focuses on a single person’s face. We’ll use Python’s sklearn, matplotlib, and numpy libraries to load, visualize, and perform operations on the data.