Improved Recognition Performance with LDA and ICA Using Vertically and Horizontally Partitioned Facial Images

Abstract: A comparative recognition performance of LDA- and ICA-based multiple classifier systems for face recognition is presented using vertically and horizontally partitioned facial images. A face image is partitioned into several vertical and horizontal segments and a multiple classifier based divide-and-conquer approach is used to combine these segments to recognize the whole face. The experiments demonstrate that vertical and horizontal partitioning result in a better recognition performance compared to the performance results of the holistic methods. Key-words: LDA, ICA, multiple classier systems, appearance-based statistical methods, classifier combination, feature-based face recognitio