A Polynomial Neural Network Classifier based on Gabor Features for the Extraction of Ear Tragus and Eye Corners
Version: 1,
Uploaded by: Administrator,
Date Uploaded:
26 November 2022
Warning
You are about to be redirected to a website not operated by the Mauritius Research and Innovation Council. Kindly note that we are not responsible for the availability or content of the linked site. Are you sure you want to leave this page?
This paper presents the results obtained with the application of a Polynomial Neural Network (PNN) classifier for the detection and localisation of craniofacial landmarks, namely the ear tragus and eye corners. The input
feature vector of the classifier is derived by Gabor filtering, using masks over two scales and four orientations. With the use of a PNN as classifier, the feature input experiences a
dimensional expansion so that a small neighbourhood for the landmarks is preferable. This in turn influences the size
of the Gabor masks that can be used, namely the coverage of the Gaussian envelope at the smallest frequency. This paper analyses the trade-off between coverage of filter envelope
and the dimensionality of the feature vector. Detection rates obtained from tests on images from three face databases are given. The robustness of the classifier to variations in
intensity, noise, scale and rotation is analysed. The results show that a PNN based on Gabor features, gives good performance for the extraction of the ear and eye features.
Language
English
Publisher
World Academy of Research in Science and Engineering
