Detecting facial features in images bares a strong resemblance to general object detection. OpenCV has a set of sophisticated functions for building general object detectors, the most well-known of which is the cascade of Haar-based feature detectors used in their implementation of the well-known Viola-Jones face detector. There are, however, a few distinguishing factors that make facial feature detection unique. These are as follows:
Precision versus robustness: In generic object detection, the aim is to find the coarse position of the object in the image; facial feature detectors are required to give highly precise estimates of the location of the feature. An error of a few pixels is considered inconsequential in object detection but it can mean the difference between a smile and a frown in facial expression estimation through feature detections.
Ambiguity from limited spatial support: It is common to assume that the object of interest in generic object detection...