Darwinian-Based Feature Extraction Using K-Means and Kohonen Clustering

This paper presents a novel approach to feature extraction for face recognition. This approach extends a previously developed method that incorporated the feature extraction techniques of GEFE ML (Genetic and Evolutionary Feature Extraction – Machine Learning) and Darwinian Feature Extraction). The feature extractors evolved by GEFE ML are superior to traditional feature extraction methods in terms of recognition accuracy as well as feature reduction. From the set of feature extractors created by GEFE ML , Darwinian feature extractors are created based on the most consistent pixels processed by the set of feature extractors. Pixels selected during the DFE process are then clustered in an attempt to improve recognition accuracy. Our new approach moves clusters towards large groups of selected pixels using techniques such as k-means clustering and Kohonen clustering. Our results show that DFE clustering (DFE C) has statistically better recognition accuracy than DFE without clustering.