Template-based detection of anchorperson shots in news programs

For easy browsing and retrieval processes in a large news database, stored news programs have to be indexed in a suitable way. One possibility of doing this is to index each news report based on covered topic. However, before the topic recognition and indexing steps can be carried out, each news program has first to be segmented into single reports. The first step in this segmentation process is the classification of all video shots of a news program in different categories, such as anchorperson (news reader) shots, news (report) shots and possible commercial breaks. We present a new reliable approach for detection of all anchorperson shots in an arbitrary news program. Increase in detection reliability, compared to the referred methods, is achieved by using a sequence-own video shot as the template for detecting all anchorperson shots of that sequence. The template itself is determined in a robust, threshold-free way