A shot classification method of selecting effective key-frames for video browsing

Publisher Website
Google Scholar
Abstract
A popular method for producing video summaries is to use cut detection to select representative key-frames for shots in a movie. The key-thrnes are then used as the summary. However, most movies contain dialogues where there are repeated similar key-frames. This paper presents anew method of reducing this redundancy by detecting this repetition using chromatic histograms and differences in local average luminance values. It classifies the key-frames into groups called “patterns” and “acts” using proposed link certainty metrics. Experimental results show that 1,362 key-frames extracted from a two hour movie are reduced to 599 dissimilar key-frames (44%). Additionally, 109 acts are detected to represent the movie sequence.

This publication has 0 references indexed in Scilit: