On the basis of video image captured by the cornea measuring instrument Corvis ST, this paper proposes an idea to improve the accuracy of distinguishing normal corneas from keratoconic corneas by extracting new feature parameters, Firstly, the original images were preprocessed by filtering and segmenting to detect the upper and lower boundaries of the cornea and calculate the curvature of anterior cornea. Then, the change of corneal curvature was analyzed by wavelet transformation method to obtain features related to the trend of corneal movement, including the trend of the whole corneal motion as well the norm and the standard deviation of corneal vibration. Furthermore, the feature parameters were extracted in succession and the optimal parameter was obtained by the minimum mean square error algorithm. The Support Vector Machine (SVM) was finally applied to distinction of normal corneas from keratoconic corneas. The experiment results on the frequency indicate that there are corneal vibrations along with the basic movement process. Besides, the proposed parameters are better than traditional parameters such as Deformation Amplitude (DA), Peak Distance(PD) at the highest concavity, which improves the accuracy, sensitivity and specificity by 10.2%, 5.7% and 6.9%, respectively. Moreover, the area under the receiver operating characteristic curve (ROC) is 0.948, close to unity. The automatic extracted feature parameters in this paper are able to improve the accuracy of classification between normal and keratoconic corneas and contribute to the clinical diagnoses.