A Kernel Based Neighborhood Discriminant Submanifold Learning for Pattern Classification

Figure 4

The 2-dimension embedding result by KNDA using eigenvectors corresponding to the 2nd and the 3rd smallest eigenvalues of matrix (𝐾(𝐷_max−𝑊_max)𝐾)⁻¹𝐾(𝐷_min−𝑊_min)𝐾.