Identification of facial neural torpid images based on Support Vector Machines

Support Vector Machines (SVMs) were machine learning algorithm based on the Statistical Learning Theory, which had strong study and classification ability and were used in facial neural anaesthetization examination. The facial feature points such as 1 nose needle, 4 canthuses, 2 mouth edges and 1 jaw point, etc. were extracted using the method: Firstly, 24 colored BMP image was preprocessed by the way of median filter and noisy data was dispelled and the boundary was detected; Secondly, the available facial information boundary was determined using the methodology of vertical gray projection. Within the boundary, the horizontal and vertical projection of eyes and mouth were performed respectively because of their different colors from that of skin. Lastly, the gray value of pixels were summed up. After the steps mentioned above, 11 dimension eigenvectors consisted of feature points were formed. After the huge simples of 11 dimension eigenvectors were studied and trained by SVMs, doctors were satisfied with the accuracy of 92.52336% of separating neural anaesthetization figures from that of normal ones.