Three-dimensional modular discriminant analysis (3DMDA): A new feature extraction approach for face recognition

In this paper, we present a novel multilinear algebra based feature extraction approach for face recognition which preserves some implicit structural or locally-spatial information among elements of the original images. We call this method three-dimensional modular discriminant analysis (3DMDA). Our approach uses a new data model called third-order tensor model (3TM) for representing the face images. In this model, each image is partitioned into the several equal size local blocks, and the local blocks are combined to represent the image as a third-order tensor. Then, a new optimization algorithm called direct mode (d-mode) is introduced for learning three optimal projection axes. Extensive experimental results conducted on four benchmark face image databases, demonstrate that 3DMDA is much more effective and robust than state-of-the-art facial feature extraction methods on both classification accuracies and computational complexities.

Graphical abstractFigure optionsDownload full-size imageDownload as PowerPoint slideHighlights► A new image representation model called third order tensor model is introduced. ► Our approach can capture the spatial redundancies in the local blocks of the image. ► Our approach can control SSS problem by changing the size of local blocks. ► Computational cost of the proposed method is much less than that of the LDA.