Dynamic Threshold Based Keyframe Detection

This paper proposes a simple dynamic threshold based method to detect abrupt scene changes for the coding of a video sequence. The detected frames are then encoded as keyframes aiming at improving the coding efficiency of the video sequence. The proposed detection method first replies on the number of INTRA macroblocks that are available during encoding process. The dynamic threshold, which is updated at the frame level, not only utilizes the statistical properties of video sequence but also takes the resolution and QP value into consideration. The detection process is fast and robust. Experimental results show that the proposed method achieves PSNR improvement.