A new similarity measure for histogram comparison and its application in time series analysis

摘要

We suggest a new measure to find similarity between ordered histograms. In our approach the similarity is related to the closeness of positions and shapes of peaks in the compared histograms. Using this measure for comparison of a big number of histograms we obtain results that agree reasonably well with the expert decisions. Applying our peak matching algorithm for studies of the histogram series obtained from experimental data, we reproduce a phenomenon that was found earlier using the expert decisions on the histogram similarity, but was not reproduced with the computer histogram analysis utilized in earlier studies.