Abstract
Over the past decade video acquisition rate has brokenthrough over 1000[Hz]. Current display systems, unfortunately, have a maximum frame-rate about 60 [frames/sec]. The display systems are definitely slower than the camera system. When a high frame-rate video (over 1000 [frames/sec]) is played in real time, however, it is necessary to down-sample the frame-rate of the video. Conventional methods keep constant the interval of neighbouring frames after down-sampling. It is based on the assumption that non-constant intervals would yield jerky motion, i.e. perceivable discontinuity in the optical flow field. However, this assumption is not true for high frame-rate video. Such video has many frames around the "constant interval" sub-sampled frame that do not yield jerky motion since the optical flow field displacement within the frames is too small to be perceived. In other words, constant frame interval is not always needed when down-sampling high frame-rate video. Relaxing the interval requirement allows greater flexibility in selecting the frames to be sub-sampled, which allows the number of bits in a sub-sampled sequence to be reduced, subject to realizing the same image quality. In this paper, we propose a temporal down-sampling method for high frame-rate video. The proposed method features the adaptive selection of down-sampled frames so as to minimize inter-frame prediction error. Our method can improve the PSNR of prediction signal by 0.13[dB] to 0.23[dB] compared to simple sub-sampling with constant interval.
