One-Class Classification Based Eigen Contours

Aiming at the shortcomings of the current object contour classification methods,we presented a novel method of one-class classification based on eigen contour.First a common normalization method of object contour using Fourier descriptor is corrected;Then eigen contour is created,which is similar with "eigen face"by applying PCA to the point range of signed centroid distance;At last,the contour according to the threshold value is classified,which is based on the principle that different kind of contour has different projecting value on the eigen contour. Experimental results verify the effectiveness of this method with classification accuracy 20% higher than that of one-class SVM.