A Feature Weighed Clustering Based Key-Frames Extraction Method

Video key-frame extraction using unsupervised clustering is an effective method to get key-frame from video clips. When multi-features are used to cluster frames, different features usually have different weight and importance. This paper introduces a feature weight based clustering method which detects the optimize cluster number and performs clustering at the same time. Starting with an over-specified number of clusters, similarly clusters are merged by weighed-distance of clusters and the dispersion of key-frame. At last, the key-frames are extracted as nearest to the cluster centers. Experimental results show encouraging results compared with the traditional methods.