A Comparison of Face Landmark Detection Techniques
Ritika Chandel, Rajnandini Bhowmick, Udhayakumar Hariharan · 2023
A lot of higher-order computer vision applications, like face recognition and facial expression analysis, depend on the precise detection of landmarks inside facial pictures. Although landmark localization is a straightforward and intuitive operation for humans, for it to be replicated in machines, it has taken multiple years of research, giving rise to large number of high-quality image data sets, and an abrupt increase in the processing capacity of computers to attain near-human proficiency. While recognizing physical characteristics on a face, such as types of noses, is an intuitive and ordinary task for human vision, it has proven to be quite tough for computer vision, as it has not profited from millennia of evolution. Although facial images generally have similar content, common variations in pose, lighting, facial expression, and facial features can give rise to a multitude of problems for many computer vision systems, leading to substantial errors and mistakes in the task of computer vision face landmarking accuracy. In this study, we examine the two facial landmark identification techniques already in use-the Mediapipe Face Landmarker and Dlib's 68-point face landmark detection algorithm-to determine the conditions in which each method performs best.