The automatic classification of erythrocyte is critical to clinic blood-related disease treatment in Medical Image Computer Aided Diagnosing(MICAD). After 3D height field recovered from the varied shading, the depth map of each point on the surfaces is applied to calculate Gaussian curvature and mean curvature, which are used to produce surface type label image. Accordingly the surface is segmented into different parts through multi-scale bivariate polynomials function fitting. The count of different surface types is used to design a classifier for training and classifing the red blood cell by means of support vector machine and particle swarm optimization. The experimental result shows that this approach is easily to implement and promising.