Abstract

MERL’s recognizer has two major components (Figure 1). The first is an aligner whose input is any arbitrary image and whose output is a cropped and rectified face if the image contains a face. The second is a comparator whose input is two aligned faces and whose output is a similarity score. The comparator is described in detail in [3]. The aligner first detects any faces in the image using a Viola-Jones frontal face detector. For any faces found, it tries to find nine facial landmark points by running nine Viola-Jones type facial landmark detectors. The facial landmark detectors are trained using the same Viola-Jones AdaBoost cascade method as the face detector. For each facial landmark, a set of aligned positive examples are manually marked. Some positive examples are shown for the right outside eye corner in Figure 2. The negative examples consist of other patches of faces and non-face patches. After the facial landmark detectors are run, if at least six landmarks are found, then the optimal scale, in-plane rotation and translation in x and y are

Citations