ABSTRACT
In this chapter, facial feature point extraction from a face data set is particularly of interest mainly for mobile robotics applications as well as for visual surveillance systems. A feature-based classification approach using fast discrete contourlet transform and Principal Component Analysis (PCA) is used in this method. The preprocessing and filtering steps are applied to each image of the face data set. These are the main steps of the proposed algorithms because they are used to sharpen the images that are able to extract efficient feature vectors from the edges in later stages. The discrete contourlet transform is applied to the preprocessed image. Each face is decomposed using the contourlet transform. Low-frequency and high-frequency contourlet coefficients are obtained at different scales and various angles. The frequency coefficients are used as feature vectors for later processes. PCA is used to reduce the dimensionality of the feature vector. Finally, the eigen feature vector is used for the classifier. The test database is projected on a Contourlet-PCA subspace to retrieve the reduced coefficients. These coefficients are used to match the feature vector coefficients of the training data set using a Euclidian distance classifier. The experiments were carried out using Face94 and IIT_Kanpur databases. Recognition rates using the contourlet transform with and without preprocessing were compared with two distance measure classifiers (Euclidean distance classifier and neural network). A Euclidean distance classifier with preprocessing provides an improvement of almost 4% to 10% in the recognition rate compared with other methods.
