Lip Recognition Using Various Neural Classifiers
In this paper we attempt at presenting a detailed comparative study of lip recognition based on Principle
Component Analysis (PCA) which is used for feature extraction for various classifiers. The classifiers include different
Artificial Neural Networks (ANN) like Back Propagation (BP), Layer-Recurrent Network (LRN), Radial Basis Network
(RBN), Modular Neural Network (MNN) with input space partitioning and neural network ensembles. Here, we have chosen
“TULIPS1 database” which is a small audiovisual database of 7 subjects for the incorporation of above methods. The paper
presents a systematic approach where we start by trying to figure out the ideal number of PCA coefficients, network models,
and network architectures for the considered problem. We later attempt to use these readings to make effective hybrid
classifiers. Experimental results confirm that using the approach, we can get effective classifiers that are characterized by
effective recognition rates.
Index terms- Lip Features; Principal Component Analysis; Back Propagation; Layer-Recurrent Network; Radial Basis
Network; Modular Neural Network; Ensembles; Fuzzy C- Means Clustering.