Automatic Facial Feature Points Detection In Bespectacled Faces
R.Saravanan, and Usha Sridhar, (2007) Automatic Facial Feature Points Detection In Bespectacled Faces. In: Research Excellence and Knowledge Enrichment in ICT: Proceeding of the 2nd International Conference on Informatics, 27th - 28th November 2007, Petaling Jaya, Selangor, Malaysia.
Full text not available from this repository.
Cranes Software International Limited, Foundations & Futures Group
Detection of eye-centers, a set of key facial feature points, in a bespectacled face has been a challenge in automated face recognition systems because glasses and frames can significantly alter the intensity profile in the region around the eyes. The altered profile usually is caused not only by the spectacle frames, but also by reflections from glass surfaces, which are dependent on illumination conditions and the tints in the glass, leading to poor accuracy in feature point detection. The problem of robust feature point detection in the presence of spectacles in combination of pose and illumination variations has not been fully addressed. This paper describes a new algorithm to extract the eye-centers accurately in the presence of spectacles. The algorithm uses image analysis in combination with a robust rule engine derived from elements of geometrical structure of a face. The novelty of the algorithm stems from the use of local adaptive thresholding to find the lip region, and then filtering out infeasible combinations of shape objects in the eye region using the lip as a reference. Further localization of eye-pupil center in the algorithm delivers eye-center locations with high accuracy. The method handles both variations in illumination and in-depth rotations up to 20 degrees. The paper presents results from the application of the algorithm on a corpus of 939 bespectacled faces obtained from many different public databases. The results show that in 96% of the images the algorithm locates eyecenters within 3 pixels of the pupil found with manual effort.
|Item Type:||Conference or Workshop Item (Paper)|
|Keywords:||Automatic Feature Detection, Image Processing, Face Recognition, Adaptive Thresholding|
Repository Staff Only: item control page