Multifeature Fusion Based on Fisher Discriminant Criterion

The relationship between single feature vector together with its discriminant vector and the separability the pattern can get is discussed in this paper, from that we know optimal discriminant vector can result in the best separability. We propose a new method of multi-features fusion in this paper, it considers all discriminant performances made by different features and different discriminant vectors, and the new feature produced by multi-features fusion has the advantages hold by every single feature. The experiments of handwritten Chinese character recognition show the effectiveness of the proposed approach.