Skip to Main Content
The eigenfaces algorithm has long been a mainstay in the field of face recognition due to the high dimensionality of face images. While providing minimal reconstruction error, the eigenface-based transform space de-emphasizes high-frequency information, effectively reducing the information available for classification. Methods such as linear discriminant analysis (also known as fisherfaces) allow the construction of subspaces which preserve the discriminatory information. In this article, multiscale techniques are used to partition the information contained in the frequency domain prior to dimensionality reduction. In this manner, it is possible to increase the information available for classification and, hence, increase the discriminative performance of both eigenfaces and fisherfaces techniques. Motivated by biological systems, Gabor filters are a natural choice for such a partitioning scheme. However, a comprehensive filter bank will dramatically increase the already high dimensionality of extracted features. In this article, a new method for intelligently reducing the dimensionality of Gabor features is presented. The face recognition grand challenge dataset of 3-D face images is used to examine the performance of Gabor filter banks for face recognition and to compare them against other multiscale partitioning methods such as the discrete wavelet transform and the discrete cosine transform.