On Metods of Graphic Object Recognition Based on DTW Metric

Abstract

We consider the possibility of using the DTW metric for shape recognition of graphic objects with incomplete group of affine transformations, including shift, scaling and rotation. A DTW-based algorithm is proposed that allows the identification of the object shape invariant with respect to such affine transformations. The statistical studies carried out show that the value of DTW metric under distortions is directly proportional to the distortion magnitude. The modeling shows that in the case of identifying the objects that do not belong to the sample class the variance of DTW metric values is significantly greater than the one observed in the case of comparing the distorted objects of the same class. This fact allows assigning a definite value of classification tolerance for each class of figures providing the required probability of false recognition and missing of objects.