Face Alignment

Introduction

Face alignment revolves around spotting the geometric structure of a face shown in a digital image and trying to achieve a canonical alignment of the detected face based on translation, scale, and rotation. Face alignment plays a key role in most face analysis functions and is frequently used as a preprocessing phase in many face recognition and 3D face reconstruction applications.

Objective

This lesson will demonstrate the face alignment process as a data normalization process while relying on the facial landmarks that are extracted using the MediaPipe library. As a result, the rotated face image will have the eyes lying on a horizontal line.

The face alignment process consists of the following steps:

1. Detect the faces within a digital image.
2. Localize the eyes of each face detected.
3. Grabb the center points for each eye detected.
4. For each face, draw a line between the center points of its underlying eyes. This line will be diagonal if the face in question is inclined.
5. Draw a horizontal line between the eyes on each face.
6. Calculate the rotation angle.
7. Rotate the face image based on the calculated rotation angle.

Dependencies

The following Python external libraries are brought into play:

Let’s code the core functions

We will lay out the core functions of this utility:

Function:

...