Hierarchical scene change detection in an MPEG-2 compressed video sequence

In this paper, we propose an efficient scene change detection algorithm for direct processing of MPEG-2 video bitstreams. The proposed algorithm utilizes the hierarchical structure of the compressed bitstreams and statistical characteristics of the coded parameters, thus greatly reducing computational requirement compared to pixel domain processing with full decompression. Occurrence of scene change is checked first in a GOP level, and if the result is affirmative it is checked again in lower levels: sub-GOP and each picture. We used several metrics for different levels: variance of DC images for I-pictures, number of macroblock types for P-pictures and motion vector types for B-pictures