Anchor Shot Detection with Diverse Style Backgrounds Based on Spatial-Temporal Slice Analysis

Abstract

Anchor shot detection is a challenging and important task for news video analysis. This paper has put forward a novel anchor shot detection algorithm for the situations with dynamic studio background and multiple anchorpersons based on spatio-temporal slice analysis. Firstly, two different diagonal spatio-temporal slices are extracted and divided into three portions, after which sequential clustering is adopted to classify all slices from two sliding windows obtained from each shot to get the candidate anchor shots. And finally, structure tensor is employed, combining with the distribution properties to precisely detect the real anchor shots. Experimental results on seven different styles of news programs demonstrate that our algorithm is effective toward the situations described above. And the usage of spatio-temporal slice can also reduce the computational complexity.