Theory and context

Facial landmark detection algorithms automatically find the locations of key landmark points on facial images. Those key points are usually prominent points locating a facial component, such as eye corner or mouth corner, to achieve a higher-level understanding of the face shape. To detect a decent range of facial expressions, for example, points around the jawline, mouth, eyes, and eyebrows are needed. Finding facial landmarks proves to be a difficult task for a variety of reasons: great variation between subjects, illumination conditions, and occlusions. To that end, computer vision researchers proposed dozens of landmark detection algorithms over the past three decades.

A recent survey of facial landmark detection (Wu and Ji, 2018) suggests separating landmark detectors into three groups: holistic methods, constrained local model (CLM) methods, and regression methods:

• Wu and Ji pose the holistic methods as ones that model the complete appearance of the face's pixel intensities...