What is Face Recognition?

Verification

• Input image, name/ID
• Output whether the input image is that of the claimed person

Recognition

• Has a database of K persons.
• Get an input image
• Output ID if the image is any of the K persons.

One-Shot Learning

One-shot learning means recognizing a face or predicting the target vector with only 1 image or sample.

For example, there is a database of images of employees working at a firm. The problem is that it includes only 1 image per employee.

To do this, we make our Neural Network learn a similarity function where we feed in two images and the algorithm outputs the degree of difference between ($\tau$ being a certain threshold) then,

d(img1,img2)= $\tau$

• If d(img1,img2) $\leq \tau$, then it's the same image.
• While, if it's $\geq \tau$, then it's a different image.

This helps us to solve the one-shot learning problem.

When the Neural Network receives a new input image, it is fed into the similarity algorithm with the other images from the database and outputs $\tau$.