Improving face classification with multiple-clustering induced feature reduction

For modern-age security, many have turn to biometrics such as face classification to verify authority. Despite this, the accuracy of existing classifiers have been constrained by the curse of dimensionality typically observed in face images. In order to simplify the task, one may ruduce the original data to a more compact variation, where only key feature components are included in the classification process. Unlike conventional feature reduction techniques found in the literature, this paper presents a novel method that makes use of cluster ensemble, specifically the summarizing information matrix, as the transformed data for a supervised learning step. Among different state-of-the-art methods, link-based cluster ensemble approach (LCE) provides a highly accurate clustering, and thus particularly employed here. The performance of this transformation model is evaluated on published face dataset and its noise-added variations, using different classifers. The findings suggest that the new model can improve the classification accuracy beyond those of other benchmark methods investigated in this empirical study.