Classification of Signals with Unknown Time Scale by Autoregressive Modeling

Abstract

This paper discusses the problem of correct classification of a given test signal of unknown classification and time scale (only integer time scales are considered). We suggest the choice of “dominant frequency” of a signal as an appropriate feature to determine the time scale of a given test signal with respect to the original signal. Autoregressive (AR) models of various orders have been fitted to the signals to estimate the dominant frequency and compared with different methods of estimating it. Numerical results have been shown with 12 different rows of images of real textures, each scaled up by 7 increasing integer time steps, to show the relative efficacy of the different methods of estimating dominant frequency.

Indexing terms:

• Time scale
• Invariant classification
• Autoregressive
• Classification of Signals
• Dominant frequency