Key Frame Extraction Towards Kernel-SIFT Identification
Zhen Yu Wu, Rui Qing Wu, Hong Yu, Bin Tang · 2013
Key frame detection is a fundamental technique in video retrieving and video summarizing.In this paper, an effective and efficient content-based key-frame detection method is proposed.This method is based on content analysis and features detection from key frames.There are two main contributions in the proposed method.Firstly, shot cuts dividing scheme is studied here to exclude transition frames (such as fade in and fade out frames), which may disturb key-frame detection.On the other hand, key frames are selected according to each frame's status in the whole shots, which is determined by kernel-SIFT matching rate.Experimental results show that the proposed method can extract the most effective key frames, which have represented the whole shots very well and contained significant features than other histogram-based methods.