Video shot boundary detection based on JND color histogram

Abstract

The method of having detected the boundaries of a shot automatically in a stream of video is referred as shot boundary detection (SBD). This paper presents a novel method of shot boundary detection, based on the JND (Just Noticeable Difference) color histogram. Using JND color model, for every two consecutive frames in a video, histograms are computed and for this pair of histogram; a degree of similarity is computed. This is treated as an indicator for discontinuity between frames. In the presented method, adaptive threshold based on sliding window is used to detect cuts and dissolve/fade gradual transitions. The performance of the presented method is evaluated on open database videos using criteria Precision, Recall and combined measures. Results show that the performance of the presented method is outstanding in its category.