Modified Multiscale Vesselness Filter for Facial Feature Detection

Abstract

The paper introduces a new filter inspired by Frangi's Vesselness filter. The latter has been used for enhancement and noise/background suppression in medical images. The proposed modification allows for the filter parameters adjustment to detect facial features, including eyes, eyebrows, nose and lips. The main contributions of this paper include: 1) re-defining the multiscale filter previously used for vessel detection, and applying it to facial biometric, 2) proposing a way to adjust scales of the filter for facial images. In addition, the modified filter does not require applying different scales to perform filtering. This allows us to avoid the complex procedure for finding suitable scales, and, therefore, computational complexity. The method has been tested over multiple controlled and uncontrolled face databases and the results show the effectiveness of algorithm.