Abstract
In this paper, we introduce the fast-pan detection problem which aims to find out the fast motion / low resolution frames caused by fast camera pan. In addition, we propose an algorithm that employs neural network training process to accomplish the detection. The features used in the neural network process include motion features and texture features. In addition, we also take temporal information into account for determining fast-pan events. Experimental results show that the proposed method which combines motion and texture features could achieve satisfactory performance. © 2005 IEEE.