Face Analysis Under Uncontrolled Conditions

2Effectiveness of Facial Landmark Detection

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

¹University of Lille, France

²Junia ISEN, Lille, France

Landmark detection is a common and often crucial preprocessing step in the context of facial analysis. Although its overall performance continues to improve, its impact on subsequent tasks must be considered. Let us consider, for example, a typical expression recognition process (see Figure 2.1). It relies on landmarks to normalize the face and extract appearance, geometry or motion features. Poor landmark detections may lead to confusion between expressions, decreasing the accuracy of the recognition process. Hence, it is necessary to ensure the robustness and stability of facial landmark detection under uncontrolled conditions and the suitability of the detected landmarks to the subsequent task (here, expression recognition).

In light of the review in Chapter 1, we can observe a large number and variety of approaches for facial landmark detection. Besides, these approaches are no longer limited to images as well as include video-based approaches. Nowadays, it may be difficult to clearly understand the current state of the problem. Hence, there is a need for benchmarking to better identify the benefits and limits of the various approaches proposed so far, especially deep learning ones. Beyond the overall performance, there is also a need to quantify the accuracy of these approaches ...