Modified Kolmogorov–Smirnov metric for shot boundary detection

Abstract

A new metric extended from the existing Kolmogorov–Smirnov measure is proposed for detecting shot boundaries of video. Exploiting more fine-grained differences between the cumulative histograms of two successive frames, the proposed metric improves both the recall and precision of shot boundary detection. Experiments conducted on a comprehensive video set show that the new metric outperforms other widely-used histogram-based metrics.