Beard/Moustache Detection & Segmentation 

Facial hair analysis has recently received significant attention from forensic and biometric researchers because of three important observations as follows. Firstly, changing facial hairstyle can modify a person's appearance such that it effects facial recognition systems. Secondly, most females do not have beard or moustache. Therefore, detecting facial hair helps to distinguish male against female with high confidence in the gender classification problem. Finally, opposed to babies and young adults, only male senior adults generally have beard or moustache. The facial hair detection can help to improve the accuracy of an age estimation system.

 

SparCLeS is composed of three stages. In the first stage, which is a preprocessing one, an input facial image is enhanced using an illumination normalization approach. In the second stage ofSparCLeS, we use a binary decision dynamic sparse classifierto determine if a region contains facial hair or not (i.e. skin).The dictionary of regions that is used for classification isbuilt from Histogram of Oriented Gradients (HOG) feature[9] descriptors extracted from images captured under identicalconditions. The third and final stage of our approach usesa new level set based algorithm to accurately and robustlysegment the detected beard and moustache regions. Our proposedsegmentation approach takes advantages of both localand global information that is modeled by four optimizationterms: (1) a local fitting term, (2) a global fitting term, (3)a contour length term, and (4) a regularization term.

The second team represents the local force of the active contour in which f_1 and f_2 are the internal and external forces respectively determinedby the local information inside and outside the contour

T. Hoang Ngan Le, Luu, K., and Savvides, M., "SparCLeS: Dynamic L1 Sparse Classifiers with Level Sets for Robust Beard/Moustache Detection and Segmentation," IEEE Trans. on Image Processing (TIP), Vol. 22, No. 8, pp. 3097-3107, Aug. 2013.
 
@article{NganLe_TIP,
author={T. Hoang Ngan Le, Luu, K., and Savvides, M},
title={SparCLeS: Dynamic L1 Sparse Classifiers with Level Sets for Robust Beard/Moustache Detection and Segmentation},
volume = {22},
number = {8},
paper = {3097 -- 3107},
journal={IEEE Trans. on Image Processing (TIP)},
year=2016
}

The third team is used to preserve the regularity of the level set function, which is necessary for an accurate computation and a stable evolution of the level set

The second team represents the global force in which c_1 and c_2 represent the average intensities inside and outside the contour and lambda_12 > 0 and lambda_22 > 0 are theparameters that control the force inside and outside the contour

The fourth team is used to regularize the zero level set and is thus to derive a smooth contour