Feature Based Similarity Measure

Abstract

This paper proposes a feature based similarity measure. The proposed measure calculates the likeness between two time series based on the number of common features between the series. This approach is different from the commonly used approach of estimating likeness based on the distance or similarity between two series. The approaches based on the distance or similarity between two series often interpret the series either as a numeric series or a text series. This however, may not always be desirable. Interpreting some series, such as video from a ’human vision’ perspective is favourable over interpreting them as simple numeric or text series. The feature based measure proposed in this paper attempts to interpret video series from a human viewpoint. It considers objects and their relative spatial and temporal positions while estimating likeness between two video time series. It also attempts to handle spatial and temporal variabilities, which abound in videos by extracting spatio-temporal invariant features. This paper introduces the measure and presents some very initial evaluation of the proposed measure. Initial results are encouraging and shows a future research possibility.