Wavelet PCA/LDA Neural Network eye detection

Abstract

Eye detection is an important step for face recognition and verification because it provides a reference point to normalize not only location but also the flat 2d orientation of face relative to the image border. The base technique that is referred to shows how Wavelet Transformation works hand in hand with Neural Networks. In this paper a proposition of a system that regiment the wavelet coefficient is introduced, as such it includes a reduction methods, namely Principle Component Analysis (PCA) and Linear Discriminant Analysis (LDA) on top of the Wavelet Transform as a feature extraction technique and Neural Network as an eye-detector classifier. Experimental results showed an increased performance (Internal 10%, ORL 9.2% and Yale 7.5%) across three datasets by using the proposed method(PCA) and 7% overall increase of performance when changing from PCA to LDA Eigen Vectors.