Automatic shot boundary detection algorithm using structure-aware histogram metric

Abstract

This paper proposes a shot boundary detection algorithm based on a novel structure-aware histogram dissimilarity measure. In the proposed approach, local histogram feature is employed to capture spatial information for comprehensive visual variations including illumination, color and motions. To improve the reliability of histogram feature in differentiating motion related shot transitions, the structural similarity measure is involved to help suppress the motion sensitivity and hence minimize false detection rate due to transition disturbances. Therefore, the proposed combination of local histogram feature and structural similarity measure offers a more robust framework to correctly evaluate frame discontinuities. Moreover, in order to ensure good performances in detecting both cut changes and gradual transitions, we propose an adaptive dual-thresholding scheme which addresses both global statistics and local variations of the frame discontinuities. Experimental results evaluated with challenging testing videos show that the proposed method achieves better overall performances than other conventional algorithms.