Locating facial features in images with uncontrolled context

摘要

This paper proposes an automatic method of detection of facial features - eyes and mouth, based on color and on the symmetry map. The improvement of the quality of the image and the application of the correction techniques, such as the re-escalation of the image, the ellipsoidal vignetting or related processing lead to a detection rate of the facial features of up to 82.21%, as compared to 56.18% the value that was obtained initially.