Abstract:

Based on the maximum inter-class margin of Support Vector Machine (SVM) and the minimum intra-class volume of Support Vector Data Description (SVDD), a discriminant algorithm is proposed, named Support Vector Data Description Discriminant Analysis (SVDDDA). This algorithm establishes two different concentric hyperspheres. The positive class samples are packed in the small hypersphere and the negative class samples are excluded from the large hepersphere. The objective function of the model maximizes the inter-class margin and minimizes the volume of the small hypersphere simultaneously. The projection coordinates are defined by the distance between the sample and the center of the hyperspheres. SVDDDA can preserve the inter-class discriminant information and intra-class scatter distribution. Results of experiment on public facial expression database demonstrate the efficiency of the proposed method.

Key words: computer application, support vector discriminant analysis, support vector machine, support vector data description