The Elastic k-Nearest Neighbours Classifier for Touch Screen Gestures

Touch screen gestures are a well-known method of person authentication in mobile devices. In most applications it is, however, reduced to checking if the user entered the correct pattern. Using additional information based on the speed and shape of finger movements can provide higher security without significantly impacting the convenience of this authorization method. In this work a new distance function for the k-nearest neighbour (kNN) classifier is considered in the problem of person recognition based on touch screen gestures. The function is based on the well-known \(\mathrm {L}^p\) distance and the elastic distance considered in elastic shape analysis. Performance of the classifier is measured using 5-fold stratified cross-validation on a set of 12 people. Only four gesture performances per gesture for each person are used to train a model. The effects of sampling rate on the classifier performance is also measured. The kNN classifier with the proposed distance function has higher accuracy than both the \(\mathrm {L}^p\) distance and the elastic distance.