Abstract: Shot boundary detection is the foundation of video indexing, retrieval and analysis.
However, the huge amount of data and the high computational cost in shot detection becomes a
bottleneck in the practical application. Using the distributed computing of Map Reduce model, a large
number of video data processing operations is splinted into several independent Map tasks for video
decoding and feature extraction. Then the feature value is detected by several Reduce tasks to get the
final shot boundary set. When extracting the features, the video is divided into 31 frames segment,
using the histogram of block with weight to calculate the difference between first frame and last
frame of the video segment. Through the adaptive threshold filtering out the non-boundary segments
and the candidate segments, the further detection has been done for the candidate segments. The twice
frame difference method of non-neighboring frame is proposed for the gradual shot detecting.
Experimental results show that the Map Reduce model and the improved shot detection algorithm is
effective in accelerating the shot detection process, and it can also achieve better detection accuracy.

Key words: shot boundary detection, Map Reduce, histogram