Face Analysis Under Uncontrolled Conditions

1Facial Landmark Detection

Romain BELMONTE¹, Pierre TIRILLY¹, Ioan Marius BILASCO¹, Nacim IHADDADENE² and Chaabane DJERABA¹

¹University of Lille, France

²Junia ISEN, Lille, France

As stated by Jin and Tan (2017), the face corresponds to a deformable object that can vary in terms of shape and appearance. The first attempts at facial landmark detection can be traced back to the 1990s. The active shape models (ASMs) (Cootes et al. 1995) represent one of the seminal works on the subject. Such a generative approach consists of a parametric model that can be fitted to a given face by optimizing its parameters. This process of applying the model to new data is called inference. Rapid progress has been made through the development of active appearance models (AAMs) (Cootes et al. 2001), constrained local models (CLMs) (Cristinacce and Cootes 2006) and other extensions (Baltrusaitis et al. 2013; Belhumeur et al. 2013; Antonakos et al. 2015; Tzimiropoulos 2015), to the point where the problem is now considered well addressed for constrained faces (Jin and Tan 2017). The research has therefore shifted to unconstrained faces with multiple and complex challenges, for example, occlusion, variations in pose, illumination and expression. Faster and more robust discriminative methods such as cascaded shape regression (CSR) (Dollár et al. 2010) have been proposed to address these challenges. They differ from generative approaches as they directly learn a mapping function between images and facial ...