Corneal Patterns Classification Based on Mel Frequency Cepstral Coefficients and SVMs

This paper presents a proposed method for corneal pattern classification using a Cepstral approach and SVMs. This approach based on the transformation of the corneal image to 1D signal, the feature extraction process and finally the classification process. MFCCs are one of the best feature extraction techniques used in 1D signal. This approach composed of two phases: a training phase and testing phase. In the first phase, a database of the corneal patterns is applied to obtain features from each corneal image, and then these features are used to train Support Vector Machines. In the second phase, features are extracted with the same steps in training phase from a set of new corneal images and finally a feature matching process is carried out. In this work, 1D signal used with time domain or in different discrete transform domains. The experimental results indicate that this technique achieves high classification rate up to about 100%.